CN119809828A

CN119809828A - A product portfolio analysis method and system based on large model agent

Info

Publication number: CN119809828A
Application number: CN202411784433.0A
Authority: CN
Inventors: 杨轶淳; 朱槐发; 谢能; 陈志明; 叶荣烜
Original assignee: China Citic Bank Corp Ltd
Current assignee: China Citic Bank Corp Ltd
Priority date: 2024-12-06
Filing date: 2024-12-06
Publication date: 2025-04-11

Abstract

The invention relates to a product warehouse-holding combination analysis method and a product warehouse-holding combination analysis system based on a large model agent, comprising the steps of defining the large model-based intelligent agent to obtain the large model agent; the method comprises the steps of obtaining the holding data and the transaction data by using a large model agent, preprocessing the holding data and the transaction data, inputting the preprocessed holding data and the preprocessed transaction data to the large model agent for analysis, and generating a report according to an analysis result. The method comprises the steps of automatically completing data cleaning and preprocessing by using a large model agent technology, improving data processing efficiency, processing complex unstructured data by using an agent by using deep learning and natural language processing technology, identifying complex modes in user warehouse holding data, generating personalized suggestions by analyzing historical data of users, improving user experience and satisfaction, processing and analyzing large-scale data in real time, providing timely investment suggestions and combination optimization schemes, and optimizing product combination benefits and risk features by using the agent by deep analysis and real-time monitoring.

Description

Product combination analysis method and system based on large model agent

Technical Field

The invention relates to the technical field of data analysis, in particular to a product combination analysis method and system based on a large model agent.

Background

In the field of financial technology, there are a variety of techniques and systems for analyzing and managing a customer's assets and portfolios. For example, conventional asset management systems may help analyze a customer's warehouse status and risk bearing capacity and provide a basic investment advice to the customer. These systems typically rely on manual analysis and fixed rules to identify good investment opportunities and build robust portfolios. At the same time, some modern systems have also begun to utilize data analysis and machine learning techniques to provide users with more accurate and personalized investment advice by analyzing large amounts of market data and user behavior data.

In addition, some specialized asset management software and services can also provide basic off-premise analysis functions, such as displaying the customer's asset distribution, investment return, risk level, etc. These systems typically provide basic charts and reports that help customers understand their investment status and market dynamics.

The traditional asset management and warehouse-holding analysis system often relies too much on manual analysis and fixed rules, resulting in inefficient and time-consuming analysis processes and being susceptible to subjective judgment by analysts, the existing systems often lack sufficient individualization, they often cannot fully understand and reflect unique needs, risk preferences and long-term investment goals of each customer, the traditional systems have limited data processing capacity, difficulty in processing large amounts of real-time market data and customer data, resulting in inaccurate and timely analysis results, the existing techniques and systems often lack intelligent algorithm optimization, inability to automatically learn and optimize from historical data and market dynamics, resulting in inadequate scientific and accurate investment advice, and the existing systems often lack friendly user interfaces and interactive designs, resulting in difficulty in understanding and using the systems by customers, and lowering user satisfaction and practical value of the systems.

Disclosure of Invention

In order to solve the defects in the prior art, the invention provides a more efficient, accurate and personalized product combination analysis method by using agent (agent) technology, and solves the problems of complex data processing, poor model adaptability, insufficient individuation, weak real-time performance and the like in the traditional method.

In order to achieve the above object, the present invention adopts the technical scheme that:

a product warehouse-holding combination analysis method based on a large model agent is characterized by comprising the following steps:

s1, defining an agent based on a large model to obtain the agent of the large model;

S2, acquiring holding data and transaction data by using a large model agent and preprocessing the holding data and the transaction data;

s3, inputting the preprocessed warehouse holding data and the preprocessed transaction data into a large model agent for analysis;

S4, generating a report according to the analysis result, wherein the report comprises an analysis report, an evaluation report, a diagnosis report, an optimization suggestion and an adjustment reason.

Further, the training of the large model agent includes:

Data collection and preprocessing, namely collecting financial text data, cleaning, de-duplication and formatting the financial text data, wherein the financial text data comprises research reports, news and supervision files;

Designing a pre-training task, and performing initial training on a large model, wherein the pre-training task comprises a mask language model task, a next sentence prediction task and a document-level continuity prediction task;

Multitask fine tuning, namely constructing a subtask data set based on each subtask of product analysis, and carrying out fine tuning on the subtasks;

Strengthening learning optimization, constructing a simulation environment, designing a comprehensive rewarding function, and introducing an advanced strengthening learning algorithm;

and (3) a continuous learning mechanism, designing an incremental learning algorithm, establishing a feedback loop, and strengthening the decision-making capability of the agent.

Further, the acquiring the taken-in-house data and the transaction data by using the large model agent and preprocessing the taken-in-house data and the transaction data comprises the following steps:

collecting data through an API interface or a data export function;

And identifying and processing the abnormal value in the data by using an abnormal detection algorithm of the large model agent, and correcting or deleting the abnormal value.

Further, the inputting the preprocessed out-of-warehouse data and the transaction data into the large model agent for analysis includes:

Acquiring asset characteristic data in the warehouse holding data, and performing identification classification and duty ratio calculation on the asset characteristic data through a large model agent;

Performing risk assessment based on the taken-in-warehouse data by using a large model agent, wherein the risk assessment comprises fluctuation analysis and pressure test;

evaluating the risk bearing capacity by utilizing a large model agent according to questionnaire survey and transaction data to obtain a risk grade;

Analyzing the warehouse holding data by using the large model agent to obtain the proportion and quantity of various funds, recommending the quantity of the funds, wherein the types of the funds comprise stock type funds, bond type funds, mixed type funds and exponential type funds;

Analyzing the stock holding plate by using a large model agent, and optimizing the holding combination configuration, wherein the stock holding comprises directly held stocks and stock-type funds;

and analyzing and diagnosing the holding funds in the holding warehouse data by using the large model agent.

Further, the acquiring the asset feature data in the warehouse holding data, and performing identification classification and duty ratio calculation on the asset feature data through the large model agent includes:

analyzing the asset characteristic data by using a large model agent, identifying and classifying various assets to obtain asset class identifiers;

Acquiring the quantity and the value of various assets in the warehouse-holding data, and calculating the duty ratio of the various assets by utilizing a large model agent according to the quantity and the value of the various assets to obtain an asset duty ratio result;

Calculating a equity asset ratio, the equity asset including stocks and stock-type funds;

The risk assessment based on the taken-in-place data by using the large model agent comprises the following steps:

Calculating the fluctuation of the asset by using a large model agent by using a statistical method to obtain a fluctuation index;

the large model agent simulates market extremes, generates multiple market scenarios using Monte Carlo simulation and evaluates the combined performance;

The evaluating the risk bearing capacity by using a large model agent according to questionnaire and transaction data to obtain a risk grade comprises the following steps:

obtaining a preliminary score by using a large model agent based on questionnaire survey;

obtaining a behavior score by using the large model agent based on the transaction data;

The large model agent analyzes based on the matching degree of the current product combination and the risk preference to obtain a combination score;

Adjusting weight proportion based on the large model agent, and carrying out weighted calculation on the preliminary score, the behavior score and the combined score to obtain a comprehensive score result;

Mapping the comprehensive scoring result to specific risk levels, wherein the risk levels comprise conservation, robustness and positive type;

Analyzing the warehouse holding data by using the large model agent to obtain the duty ratio and the quantity of various funds, recommending the quantity of the funds, comprising:

Acquiring fund data in the warehouse holding data, performing type distribution analysis on various funds by using a large model agent, and counting the occupation ratio of various funds;

Calculating the recommended quantity of funds by the large model agent based on the investment scale, the proportion of various funds and the risk bearing capacity of the user;

The large model agent evaluates the dispersion degree of the investment portfolio according to the ratio of various funds, and calculates the dispersion degree;

The large model agent calculates the recommended fund holding quantity and gives out dynamic optimization suggestions;

The large model agent is utilized to analyze stock holding plate blocks, optimize holding combination configuration, and comprises the following steps:

Acquiring holding stock data, wherein the holding stock comprises directly held stocks and stock-type funds, and identifying a plate to which a stock holding bin belongs;

analyzing by using a large model agent to obtain the performance and market characteristics of each plate, wherein the performance and market characteristics of each plate comprise investment risk and income characteristics of the plate;

optimizing the combination configuration of the holding bin by using the large model agent;

The analysis and diagnosis of the holding funds in the holding data by using the large model agent comprises the following steps:

Acquiring detailed information of funds held by a user;

the large model agent evaluates risk and benefit features of holding funds.

Further, the method further comprises:

Collecting new data and feedback data;

training and fine-tuning a large model agent;

Updating the knowledge base and the analysis model.

The invention also relates to a product warehouse-holding combination analysis system based on the large model agent, which is characterized by comprising the following steps:

the agent definition module is used for defining an agent based on a large model to obtain the large model agent;

The data preprocessing module is used for acquiring the holding data and the transaction data by using the large model agent and preprocessing the holding data and the transaction data;

the agent analysis module is used for inputting the preprocessed warehouse holding data and transaction data into the large model agent for analysis;

And the report generation module is used for generating a report according to the analysis result, wherein the report comprises an analysis report, an evaluation report, a diagnosis report, an optimization suggestion and an adjustment reason.

The invention also relates to a computer-readable storage medium, characterized in that the storage medium has stored thereon a computer program which, when executed by a processor, implements the method described above.

The invention also relates to an electronic device, which is characterized by comprising a processor and a memory;

The memory is used for storing program instructions;

The processor is configured to execute the above method by calling a program instruction.

The invention also relates to a computer program product comprising a computer program and/or instructions, characterized in that the computer program and/or instructions, when executed by a processor, implement the steps of the above-mentioned method.

The beneficial effects of the invention are as follows:

The product combination analysis method and system based on the large model agent automatically complete data cleaning and preprocessing by utilizing the agent technology, greatly reduce manual operation time and improve data processing efficiency, the agent can process complex unstructured data and identify complex modes in user holding data by utilizing deep learning and natural language processing technology, the agent can generate personalized suggestions by analyzing historical data of a user, user experience and satisfaction are improved, the agent can process and analyze large-scale data in real time, a timely investment suggestion and combination optimization scheme is provided, and product combination benefits and risk characteristics are optimized by means of deep analysis and real-time monitoring of the agent.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings that are required to be used in the description of the embodiments of the present application will be briefly described below.

FIG. 1 is a schematic flow chart of a product combination analysis method based on a large model agent;

FIG. 2 is a schematic diagram of a product combination analysis system based on a large model agent;

Fig. 3 is a schematic structural diagram of an electronic device according to the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the technical solutions of the present application will be clearly and completely described below with reference to specific embodiments of the present application and corresponding drawings. It should be understood that the described embodiments are only some, but not all, of the embodiments of the present application and are intended to illustrate the present application and not to limit the scope of the present application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.

As shown in fig. 1, the embodiment of the invention discloses a product combination analysis method based on a large model agent, which comprises the following steps:

s1, defining an agent based on a large model to obtain the agent of the large model.

In the invention, the large model agent is an agent based on a large-scale language model and is specially designed for product combination analysis.

Specifically, training of large model agents includes:

Designing a pre-training task, and performing initial training on the large model, wherein the pre-training task can comprise a Mask Language Model (MLM) task, a Next Sentence Prediction (NSP) task and a document-level consistency prediction task;

In this embodiment, the specific implementation process of large model agent construction and training includes the following key steps:

(1) The GPT3.5 large-scale language model is selected as a basic model, is excellent in performance in the field of natural language processing, can understand and generate natural language texts, and provides a strong basis for customized training in the subsequent financial field.

(2) The method comprises the steps of pre-training in the financial field, integrating a large amount of text data such as research reports, news and supervision files in the financial field, preprocessing the data, designing pre-training tasks including a Mask Language Model (MLM) task, a Next Sentence Prediction (NSP) task and a document-level consistency prediction task, enhancing the understanding capability of the model on the financial text, and adopting distributed training to perform efficient training by using multiple GPUs or multiple clusters in the training process so as to improve training efficiency.

(3) Data is collected and marked for each subtask of product analysis, a plurality of related tasks such as product classification, risk assessment and the like are optimized, the performance of the model on the related tasks is optimized, and training of a specific task is accelerated by using a progressive fine tuning or adapter method, so that the model can be quickly adapted to a new task.

(4) The method comprises the steps of establishing a simulated financial market environment, providing a relatively safe environment for testing and optimizing decision making capability of a model, designing a comprehensive rewarding function based on multiple dimensions of return on investment, risk control, user satisfaction and the like to train the model to make a better decision, and using a near-end strategy optimization (PPO) to realize an advanced reinforcement learning algorithm so as to improve adaptability and decision quality of the model in a complex financial market.

(5) And establishing a feedback loop, wherein the large model agent can perform self-adjustment and optimization according to actual performance to continuously optimize the model performance.

(6) The method comprises the steps of carrying out back measurement by using historical data, verifying the performance of a model on the historical data, carrying out A/B test in a simulated market and an actual market environment, comparing different models or strategy performances, determining an optimal model or strategy, establishing a continuous monitoring mechanism, evaluating the decision quality of a large model agent in real time, and adjusting in time according to the requirement.

Through the steps, a large model agent which has deep financial knowledge and strong analysis capability and can be adapted and reliably operated in an actual market environment can be obtained, and the large model agent is the core of product combination analysis.

In the embodiment of the invention, the large model agent constructed according to the steps has the following characteristics:

(1) The multi-mode input processing capability can process a plurality of data types such as text, numerical values, images and the like;

(2) Deep financial knowledge understanding, namely deep understanding of financial markets, fund operations and investment strategies;

(3) Personalized analysis capability, namely customized analysis and suggestion can be provided according to the specific situation of a user;

(4) Real-time learning and adaptation, which enables continuous learning and optimization from up-to-date market data and user feedback.

S2, acquiring the holding data and the transaction data by using the large model agent and preprocessing the holding data and the transaction data.

Related holding and transaction data is obtained from the user's bank account, dealer account and foundation platform. Through an API interface or a data export function, collecting product warehouse-holding combination information of a user, including product names, asset classification and duty ratio, resource output, historical transaction and the like, ensuring that comprehensive and accurate user warehouse-holding information is obtained, and providing basic data for subsequent analysis.

Preprocessing data by using the large model agent, identifying and processing an abnormal value in the acquired data by using an abnormal detection algorithm of the large model agent, and correcting or deleting the abnormal value. And identifying an abnormal value obtained from the user bin holding data by using an abnormal detection algorithm, correcting (e.g. filling in a missing value) or deleting the identified abnormal value to ensure the data quality and the reliability of the data, and cleaning, characteristic engineering and standardization the data to form a structured data set to obtain the preprocessed user bin holding data.

And S3, inputting the preprocessed holding data and the preprocessed transaction data into a large model agent for analysis.

In the embodiment of the invention, the analysis process mainly comprises the following steps:

(1) And acquiring asset characteristic data in the warehouse holding data, and performing identification classification and duty ratio calculation on the asset characteristic data through a large model agent.

The asset characteristic data in the holding bin of the user is acquired, the characteristics are analyzed by using a pre-trained classification model through a large model agent, various assets are automatically identified and classified, the processed asset class identification is obtained, and the assets can be rapidly and accurately classified into different classes such as stocks, bonds, funds, cash and the like by using the large model agent, so that a foundation is provided for subsequent duty ratio calculation.

And acquiring the quantity and value data of various assets in the user holding warehouse, and calculating the duty ratio of the various assets according to the data by utilizing the large model agent to obtain a processed asset duty ratio result. The method helps users intuitively understand the asset allocation situation of the users so as to evaluate investment risks and gain potential, thereby supporting more reasonable investment decisions. For example, the data of a certain user is input and output:

total asset 1,035,000 yuan

The cash ratio is 48.31%

Stock ratio 15.94%

Foundation ratio of 16.43%

Other investment ratio of 19.32%

The equity asset ratio is calculated, the equity asset including stocks and stock-type funds. And acquiring the stock component data of the stocks and the stocks in the funds directly held by the user, and integrating and calculating the data by the large model agent to obtain the total value of the equity assets and the equity asset ratio of the user, thereby helping the user evaluate the investment proportion of the user on the equity assets, knowing the risk and income characteristics and assisting in decision-making. For example, a large model agent is used to output based on the taken-in data of a certain user:

[ equity asset Total value: 225,000 yuan ]

Rights and interests asset ratio 21.74%

(2) And carrying out risk assessment based on the taken-warehouse data by using the large model agent, wherein the risk assessment comprises fluctuation analysis and pressure test.

And acquiring historical price data of the assets in the holding warehouse of the user, and calculating the volatility of the assets by the agent by using a statistical method to obtain volatility indexes of various assets, so as to provide quantitative analysis on the price fluctuation risk of the assets for the user and help the user understand the potential market risk.

The market history data and the user warehouse holding data are acquired, and the agent generates various market situations and evaluates the combination performance by simulating market extreme conditions (such as stock market breakout or economic decay) and using Monte Carlo simulation, so that the user is helped to evaluate the stability and the risk resistance of the investment combination under different market environments, and thus, the investment strategy is optimized. For example, in a simulated market dip 20% scenario, a combination of certain users is expected to lose 12%.

(3) And evaluating the risk bearing capacity by utilizing the large model agent according to the questionnaire and the transaction data to obtain the risk grade. The assessment of the risk bearing capacity of the user is a key step of establishing a personalized investment strategy for an investor, and can obtain a more accurate assessment result by combining the behavior analysis of the traditional questionnaire and the large model agent.

Risk tolerance assessment includes multidimensional data collection, behavioral pattern analysis, portfolio assessment, risk tolerance scoring, risk classification, personalized risk capability promotion advice.

And collecting questionnaire and historical transaction data, and analyzing the behavior patterns of the users. The method comprises the steps of filling a risk bearing capacity questionnaire by a user, comprising the problems of financial conditions, investment experience, risk preference and the like, sorting and analyzing data filled by the user by utilizing a large model agent, quantifying the risk preference and the investment target of the user, obtaining a preliminary score, evaluating the risk bearing capacity of the user in a quantitative and qualitative mode, providing basic information for subsequent analysis, obtaining historical transaction data of the user, comprising transaction frequency, loss bearing capacity, profit and loss performance and the like, analyzing the historical transaction data by the agent, identifying past investment behaviors and decision modes of the user, obtaining a behavior score, extracting the real risk bearing capacity of the user from the historical transaction record based on the transaction mode and the risk bearing characteristics of the user, further improving the accuracy of risk assessment, and providing support for personalized risk assessment. In this embodiment, the historical transaction data can be used to obtain the transaction frequency, the maximum single loss, the maximum single profit and loss ratio, and the like in the past year, and in this embodiment, the score can be set to be 100 points, and the user behavior is analyzed to obtain the user preliminary score and the behavior score.

The large model agent analyzes based on the matching degree of the current product investment combination of the user and the risk preference to obtain a combination score. The risk and benefit characteristics of the current portfolio are evaluated, providing a recommendation for the user to combine optimization. In this embodiment, the combined score is set to be 100 points.

The method comprises the steps of obtaining various scoring data (including preliminary scoring, behavioral scoring and combined scoring), determining weight proportion by a large model agent, calculating to obtain a comprehensive scoring result by using a weighting algorithm, providing comprehensive risk bearing capacity assessment for users, helping the users to better understand the risk bearing capacity of the users in investment decision, and periodically reevaluating the risk bearing capacity of the users, wherein a reevaluation mechanism is triggered in each quarter or after a major market event in the embodiment.

Mapping the composite scoring result to a specific risk level, in this embodiment, the grading risk level may include conservation (score < 30), robustness (score > =30 and score < 60), aggressiveness (score > =60), and the like. The risk level can provide visual risk cognition for the user, and is convenient for the user to make corresponding investment decisions.

And acquiring the risk level and related data of the user, and generating personalized risk capacity improvement suggestions by the agent based on the specific situation of the user. The method helps users identify their own risk management shortboards, and provides corresponding education and resources to improve their risk bearing capacity and effectiveness of investment strategies.

(4) And analyzing the warehouse holding data by using the large model agent to obtain the duty ratio and the quantity of various funds, and recommending the quantity of the funds. This step uses primarily holding fund data, the holding fund quantity analysis is aimed at evaluating the dispersion of user portfolios and providing optimization advice, which considers a number of influencing factors to balance dispersion and management efficiency.

And acquiring the fund data held by the user from the warehouse-holding data, integrating the data by the large model agent based on the type, scale, history performance and the like of the fund to obtain integrated fund data, carrying out distribution analysis on the funds of different types by the agent, and counting the investment ratio of various funds. The types and proportions of funds invested by the user are identified, so that the user is helped to know the diversity and risk characteristics of the investment portfolio. In this embodiment, the fund type may include stock type, bond type, hybrid type, index type, and the like.

Based on the investment scale, the proportion of various funds and the risk bearing capacity of the user, the large model agent calculates the recommended quantity of the funds, provides the user with a fund selection suggestion suitable for the risk preference of the user, and optimizes the investment portfolio configuration. In this embodiment, 5-10 funds may be proposed for the primary investor, 10-20 funds for the medium-scale investor, and 15-25 funds for the professional investor.

And acquiring the investment proportion of various funds held by the user, evaluating the dispersion degree of the investment portfolio by the agent, calculating the dispersion degree, quantifying the risk dispersion of the investment portfolio of the user, reminding the user to avoid excessive concentrated investment, and reducing the potential risk. In the present embodiment, the concentration and dispersion may be evaluated using indices such as the herdall-herhman index (HHI).

According to the dispersion degree, market conditions and user characteristics, the agent provides dynamic optimization suggestions for users, timely adjusts investment strategies according to market condition changes and investment conditions of the users, and improves the overall performance and risk management capability of the investment portfolio.

(5) And analyzing the stock holding plate by using the large model agent, and optimizing the holding combination configuration. Stock holding plate analysis is a key step in assessing portfolio diversification and potential risk, which involves a comprehensive analysis of directly held stocks and indirectly held stocks through funds to ensure that portfolios are properly dispersed across different industries and plates.

And acquiring the fund data held by the user from the warehouse holding data, classifying and analyzing various assets by the agent, collecting and classifying stock warehouse holding information, analyzing the fund season report, and extracting the top ten heavy warehouse stocks. The main stocks of the investment of the user and the plates to which the main stocks belong are identified, and a foundation is laid for the subsequent deep analysis.

And according to the analysis result of the agent, the performance and market characteristics of each plate are obtained, and the investment risk and income characteristics of each plate are particularly concerned.

Market trend data (such as industry indexes, market news and the like) are acquired, and the agent is combined with the user's holding data to identify the influence of the current market trend on the user holding, so that a more time-efficient analysis basis is provided.

Based on the user characteristics, plate analysis and market trend, the agent provides personalized investment portfolio adjustment suggestions for the user, helps the user optimize the warehouse holding configuration according to market changes and investment targets, and improves the overall investment return and risk management capability.

(6) And analyzing and diagnosing the held funds in the holding data by using the large model agent, and analyzing the funds of the users one by one.

The method comprises the steps of obtaining detailed information of funds held by a user, and inquiring latest data of each fund by using agents, wherein the latest data comprise historical net value, warehouse holding information, fund manager information, similar fund performances and the like.

The large model agent comprehensively analyzes the data based on the historical performance, risk indexes and cost structures of the funds, evaluates the risk and income characteristics of each fund and provides a comprehensive investment view for users.

Market data and macro economic indexes are acquired, and agents evaluate future performances of funds by using a prediction model, so that users are helped to understand potential risks and benefits, and more scientific investment decisions are supported.

And combining the analysis result, the agent generates personalized investment advice for the user, and guides the user to optimize the configuration of the holding funds so as to better meet the investment targets and risk bearing capacity.

And S4, generating a report according to the analysis result, wherein the report comprises an analysis report, an evaluation report, a diagnosis report, an optimization suggestion and an adjustment reason.

In the embodiment of the invention, the large model agent can generate investment suggestions, optimization schemes, risk early warning and the like, and the final result is presented through a user interaction layer.

In the invention, a deep learning model is adopted to replace a traditional rule base system or an expert system for financial analysis, deep behavior analysis is used to replace a standardized questionnaire for risk assessment, weight distribution is dynamically adjusted in a warehouse holding analysis instead of a fixed industry weight distribution model, multidimensional analysis and prediction are adopted instead of only paying attention to historical performance and basic surface indexes of a fund for fund diagnosis, and advanced AI technology and comprehensive data analysis are utilized to realize the processing of complex market dynamics, personalized analysis, dynamic adjustment and comprehensive assessment.

The invention also provides an expansion method for continuous learning and updating, which comprises the steps of collecting new data and feedback data, training and fine-tuning a large model agent, and updating a knowledge base and an analysis model.

As shown in fig. 2, the embodiment of the invention discloses a product warehouse-holding combination analysis system based on a large model agent, which structurally comprises:

By using the system, the above-mentioned operation processing method can be executed and the corresponding technical effects can be achieved.

Examples:

A specific application scenario of a product warehouse-holding combination analysis system based on a large model agent is described. An investor user owns a portfolio of funds and stocks, and the system is used to assist the user in analyzing and optimizing his portfolio by the following steps.

Agent training and acquisition

Input:

Pre-training model GPT-3 (175B parameters);

the financial data set comprises 10TB text data including financial teaching materials, research reports, market analysis and the like;

Market data, stock, bond, foreign exchange date line data over the past 20 years.

1.1 Financial Domain Pre-training

Continuous pre-training of GPT-3 using 10TB financial data

Training time of 2 weeks

As a result, the confusion (perplexity) decreases from original 20 to 12

1.2 Multitasking trimming

5 Tasks are designed:

a) Market trend prediction

B) Risk assessment

C) Asset configuration

D) Financial statement analysis

E) Macroscopic economic analysis

Each task uses 10 ten thousand pieces of annotation data.

Training time is 5 days

Results:

The average accuracy rate is improved from 65% to 88%, and the F1 fraction is improved from 0.60 to 0.85.

1.3 Reinforcement learning optimization

A stock exchange simulation environment is constructed and trained using the PPO algorithm.

Training step 100 ten thousand steps

Training time 1 week

Results:

in the simulated environment, the annual yield of the Agent is improved from 5% to 15%, and the maximum withdrawal is reduced from 30% to 20%.

1.4 Implementation of continuous learning mechanism

An online learning algorithm is designed, and the model is updated with the latest 1000 pieces of market data daily.

Results:

the response speed of the model to the new event is shortened from 24 hours to 2 hours, and the prediction accuracy under the abnormal market condition is improved from 50% to 70%.

And (3) outputting:

the final training results in a financial Agent with the following capabilities:

the average accuracy over 5 financial tasks reaches 88%;

realizing 15% annual yield in a simulated transaction environment;

can adapt to new market conditions within 2 hours;

The prediction accuracy of 70% is maintained under abnormal market conditions.

2. Whole account equity bin analysis

Input data:

USER ID USER 12345

Account information:

500,000 yuan deposit

Stock account:

a stock 1000 shares, current price 90 yuan/share

B stock 500 shares, current price 150 yuan/share

Fund account:

foundation (stock type) 50,000 parts, net 1.2 yuan/part

Fund D (bond type) 100,000 parts, net 1.1 yuan/part

Other investments:

E financial products of 200,000 yuan, expected annual income 4.5%

The treatment process is as follows:

2.1 data Collection and Pre-processing

The Agent calls an API to acquire real-time market data and update the current value of each asset.

And (3) inputting an Agent query prompt sentence, namely collecting and preprocessing all account data, cleaning abnormal data and updating the current market value of each asset. "

And (3) outputting results:

A stock total value 90,000 yuan

Total stock value of 75,000 yuan

Total value of C foundation 60,000 yuan

Total value of D fund 110,000 yuan

2.2 Asset Classification and duty cycle calculation

Assets are divided into cash, stocks, funds and other investments, and various asset duty ratios are calculated.

And (3) inputting an Agent inquiry prompt sentence, namely classifying the assets and calculating the duty ratio of various assets. "

And (3) outputting results:

Total asset 1,035,000 yuan

The cash ratio is 48.31%

Stock ratio 15.94%

Foundation ratio of 16.43%

Other investment ratio is 19.32%

2.3 Equity bin analysis

Equity assets (stock + stock type funds) duty are calculated.

And (3) inputting an Agent query prompt sentence which is the total duty ratio of the analysis equity class asset. "

And (3) outputting results:

225,000 yuan of rights and interests asset total value

Equity asset ratio 21.74%

2.4 Risk assessment

The risk is assessed using the historical volatility and VaR (Value at Risk) models.

And (3) inputting an Agent query prompt sentence, namely 'evaluating the overall risk of the investment portfolio, calculating historical fluctuation rate in a packet expansion way and calculating a VaR value by using Monte Carlo'. "

And (3) outputting results:

annual fluctuation rate of investment portfolio is 8.5 percent

95% VaR (10 days): 2.3%

2.5 Report Generation

Generating analysis reports and suggestions:

And (3) inputting an Agent query prompt sentence, namely generating a detailed analysis report, wherein the detailed analysis report comprises a current condition summary, risk assessment and personalized advice.

Please ensure that the analysis is comprehensive, accurate and presents the results in an easy-to-understand manner. "

Outputting a result:

"esteem user_12345:

Your account total asset is 1,035,000 yuan, the current equity class asset is 21.74%, at a lower level. Considering your risk bearing capacity assessment results (assuming medium), it is recommended to moderately increase equity asset proportion to 30-40% to increase long term revenue potential.

The annual fluctuation rate of your portfolio is 8.5%, the maximum possible 10-day loss at 95% confidence is 2.3%, and the overall risk is controllable.

Specific suggestions are:

Consider that a portion of cash (about 10-15 ten thousand yuan) is transferred to a stock or stock-type fund to increase the equity ratio.

The option to increase the C foundation or to purchase a new exponential foundation, dispersing the single stock risk.

Closely paying attention to the expiration of E-financial products, consider transferring a portion of funds to investment varieties with higher risk-to-benefit ratios.

Please take care of the above advice according to your specific situation and market changes. In the event of any doubt, welcome consultation at any time. "

3. User risk tolerance assessment

3.1 Multidimensional data collection

Data were obtained in the form of an online questionnaire:

For example:

USER ID USER 12345

Questionnaire survey results:

Age of 35 years

Annual income of 50 ten thousand yuan

Investment experience of 5 years

Financial objectives retirement savings and child education fund

Loss-bearing capacity, capable of receiving 20% of short-term fluctuations

And inquiring historical transaction data to the Agent according to the obtained user data, and then inputting an Agent inquiry prompt sentence, namely analyzing the historical transaction data of the user, and calculating the transaction frequency, the maximum single loss and the profit and loss ratio of the past year. "

And (3) outputting results:

transaction frequency over the past year, average 3 times per month

Maximum single loss of 8%

Ratio of deficiency to surplus 1.2

Risk tolerance assessment process:

3.2 behavioral pattern analysis

A) Questionnaire analysis

And analyzing the questionnaire result, and giving weight to each factor.

And (3) inputting an Agent inquiry prompt sentence, namely analyzing the questionnaire survey result of the user, and giving out preliminary scores by considering factors such as age, income, investment experience and the like. "

Output of results preliminary score 75 points (100 points)

B) Transaction behavior analysis

Historical transaction data is evaluated.

And (3) inputting an Agent query prompt sentence, namely evaluating historical transaction behaviors of the user, including transaction frequency, loss bearing capacity and profit and loss performance, and giving a behavior score. "

Output of results behavior score 70 points (full 100 points)

3.3 Portfolio assessment

And analyzing the matching degree of the current investment portfolio and the risk preference.

And (3) inputting an Agent inquiry prompt sentence, namely 'combining the current investment combination condition, evaluating the actual risk preference of the user, and giving a combination score'. "

Output of results, combined score 80 points (full 100 points)

3.4 Comprehensive score calculation

The final score is calculated using a weighted average method.

An Agent query prompt sentence is input, namely 'the risk bearing capacity score of a user is calculated by using a multi-factor model'.

The results output are composite scores (75×0.4) + (70×0.3) + (80×0.3) =75 points

3.5 Risk type determination

The users are classified according to the score.

An Agent query prompt sentence is input, namely, the user is classified into proper risk bearing types according to the score.

Outcome output (robust investors (between 65-80 partitions)

3.6 Personalized Risk capability promotion advice

And (3) inputting an Agent query prompt sentence, namely generating a detailed evaluation report, wherein the detailed evaluation report comprises an analysis result, risk type judgment and personalized advice. Please ensure that the assessment is comprehensive, objective and provides clear explanations and advice. "

Agent outputs the result:

"esteem user_12345:

Based on the information you provide and our comprehensive analysis, your risk bearing ability assessment results are as follows:

Comprehensive risk score 75 points (100 points)

Type of risk tolerance robust investors

Analysis shows that:

your 35 years of age and 5 years of investment experience indicate that you have the ability to assume a moderate risk.

Your transaction shows a rational attitude towards risk, being able to accept short term fluctuations.

The current portfolio substantially matches your risk preferences but there is still room for optimization.

Proposal:

It is contemplated that equity asset proportions may be moderately increased to 30-40% to increase long term revenue potential.

Some medium-risk investment products, such as balanced funds or premium blue-laid strands, may be explored.

A periodic investment plan is established and the risk is dispersed using dollar-cost averaging strategy.

The risk bearing capacity of you is reevaluated each year, ensuring that investment strategies remain consistent with your life changes.

Note that the present evaluation is based on information and history data provided by you. Changes in market conditions and personal conditions may affect your actual risk bearing capacity. Advice you review and adjust the investment strategy regularly.

In the event of any doubt, welcome consultation at any time. "

4. Analysis of the number of held funds

And inputting the acquired user holding data into the Agent.

Analysis and treatment process:

4.1 analysis of the status of the current fund

The 4 funds currently held are analyzed.

And (3) inputting an Agent inquiry prompt sentence which is' all funds warehouse information of the integrated user_12345, including types, scales, historic performances and the like of each fund. "

Results:

"USER ID: USER_12345

Current fund holding bin:

foundation (stock type) 50,000 parts, net 1.2 yuan/part

Fund D (bond type) 100,000 parts, net 1.1 yuan/part

F-based (mixed) metal 30,000 parts, net 1.5 yuan/part

20,000 Parts of G-base (exponential type), net 1.0 yuan/part

User total assets 1,035,000 Yuan (obtained from previous analysis)

Risk tolerance robust (obtained from previous evaluations)

Total value of funds = 253,000 yuan, 24.44% of total assets% "

4.2 Fund type distribution analysis

The distribution of the fund types is evaluated.

An Agent inquiry prompt sentence is input, namely 'analysis of the distribution of the foundation types'.

As a result, stock form 23.72%, bond form 47.43%, mixed form 19.37%, and index form 9.48%

4.3 Recommended fund quantity calculation

The recommended number is calculated based on the total asset size and risk tolerance.

An Agent query prompt sentence is input, namely, a suitable fund quantity range is recommended in consideration of the total asset size and risk bearing capacity of a user "

Calculation recommended fund quantity=5+ (total assets/500,000) =5+ (1,035,000/500,000) ≡7-

4.4 Evaluation of dispersity

Concentration was assessed using the heradar-herschmann index (HHI).

Input Agent query prompt sentence "based on the bin holding data, evaluate concentration using the hevincal-hehnman index"

Calculation HHI=0.2372≡2+0.4743≡2+0.1937≡2+0.0948≡2= 0.3374

As a result, the medium dispersity (HHI is between 0.15 and 0.45)

4.5 Optimization suggestion generation

And generating an optimization suggestion according to the analysis result.

And (3) inputting an Agent query prompt sentence, namely generating a detailed analysis report comprising the current state, the suggested quantity and type of funds and adjusting reasons.

Please ensure that the analysis is comprehensive, the advice is feasible, and taking into account the user's specifics. "

"Esteem user_12345:

Based on your fund holding conditions and overall financial status, our analysis results are as follows:

Current fund holding status:

Number of held funds 4

The total value of the fund is 253,000 yuan, accounting for 24.44 percent of the total asset

Analysis results:

The number of funds, you currently hold 4 funds, is below the recommended range of numbers.

The type distribution is that the bond type funds occupy a relatively high proportion, and the proportion of the equity type funds can be considered to be increased.

The number of funds recommended-considering your asset size and risk bearing capacity, 6-8 funds are recommended to be held.

The dispersity is that the concentration index of the current combination is 0.3374, which belongs to the medium dispersion level, and still has the optimization space.

Optimization advice:

Adding 2-3 funds, giving priority to the following types:

1-disc stock type foundation

1 Industry theme foundation (e.g. science and technology or medical)

1 QDII foundation (for overseas market configuration)

The holding ratio of the bond-type funds is moderately reduced, and the funds are distributed into the newly added funds.

Consider adding 1 monetary market fund for cash management and to deal with short term liquidity demands.

Adjustment reason:

Increasing the number of funds helps to disperse the risk and increase the anti-wave capability of the combination.

Adding different types of funds can optimize the asset configuration, keeping track of more market opportunities.

Moderately increasing equity funds accords with the robust risk bearing capacity of you and contributes to improving long-term benefit potential.

Note that these advice are based on your current financial situation and market circumstances. Before making any investment decisions, you are advised to carefully evaluate the specifics of each fund and consider your individual investment goals and market expectations. If you need more detailed advice or have any doubt, welcome consultation at any time. "

5. Stock holding plate analysis

And inputting the acquired user holding data to the Agent.

Input data:

USER ID USER 12345

5.1 Metal-based data analysis

And (3) inputting an Agent inquiry prompt sentence, namely calculating share of directly holding stocks and indirectly holding stocks through the funds according to the information of the fund holding warehouse of the USER 12345, and classifying according to industry plates. "

Agent output:

the stock keeping analysis result of "user_12345 is as follows:

Directly holding the strand:

technological plate 90,000 yuan (8.70%)

Financial block 75,000 yuan (7.25%)

Holding strands through foundation:

9,000 yuan (0.87%)

Consumption plate 6,000 yuan (0.58%)

New energy plate 4,800 yuan (0.46%)

Finance plate 7,200 yuan (0.70%)

Medical plate 3,600 yuan (0.35%)

Distribution of the combined plates:

Science and technology/Internet 99,000 Yuan (9.57%)

Finance 82,200 yuan (7.94%)

Consumption of 6,000 yuan (0.58%)

New energy source 4,800 yuan (0.46%)

Medical treatment 3,600 yuan (0.35%)

Total stock investment 195,600 yuan (18.90% total assets)'

5.2 Depth plate analysis

And (3) inputting an Agent query prompt sentence, namely analyzing plate distribution of stock holding, and evaluating risk exposure, deviation degree relative to the market and future potential of each plate. "

Agent output:

"depth plate analysis results:

Risk exposure:

The technology/internet panel ratio is highest and risk exposure is large. The board volatility is typically higher than the average market level.

The financial block is the second largest holding bin, also with higher systematic risks.

Other plates occupy smaller space and provide a certain dispersing effect.

Market deviation degree:

Science and technology/Internet: +3.57% (assuming an average market of 6%)

Finance: +1.94% (assuming an average market of 6%)

Consumption: -2.42% (assuming an average market of 3%)

New energy source-1.54% (assuming that the average market is 2%)

Medical treatment: -1.65% (assuming an average market of 2%)

Future potential:

technology/internet-long term growth potential is large, but short term may face regulatory and valuation pressures.

Finance is greatly affected by macro economy and may be challenging in mid-term.

The consumption has better defensive property and stable long-term growth.

New energy accords with national strategy and has strong long-term growth potential.

Medical treatment, under the trend of aging population, has long-term growth potential. "

5.3 Market trend integration

And (3) inputting an Agent inquiry prompt sentence, namely 'combining the current market trend and industry development prospect' to evaluate the future performance potential of each plate. "

Agent output:

"market trend integration analysis:

Science/internet:

There is a current trend to estimate high-enterprise, but the innovation power is strong.

The future potential is 5G, AI, the new technology drive of cloud computing and the like, and the long-term growth prospect is good.

Finance:

current trends are being influenced by economic resuscitation and interest rate environments.

Potential in the future, financial technological innovations may bring new growth points.

Consumption:

the current trend is that consumer upgrades continue and high-end consumer product demands increase.

The potential in the future is that under the national large circulation strategy, the consumption plate is expected to benefit continuously.

New energy:

The current trend is that the policy support is strong, and the industry chain is rapidly developed.

The future potential is that the long-term growth space is huge under the aim of carbon neutralization.

Medical treatment:

The current trend is that the medical investment is increased after epidemic situation, and the research and development of innovative drugs are accelerated.

Future potential-population aging and medical needs increasing support for long-term growth. "

5.4 Personalized adjustment advice

And (3) inputting an Agent inquiry prompt sentence, namely providing personalized plate adjustment advice based on the stock holding proportion, risk preference, investment target and current market trend of the user. "

Agent output:

"personalized adjustment advice:

Considering your robust risk preference and current market trend, the following is suggested:

science and technology/internet board:

it is suggested that the reduction is small from 9.57% to about 8%.

The reason is to reduce the risk of concentration but leave open enough to gain long-term growth.

Financial plate:

It is suggested to maintain the existing configuration.

The reason is that current configurations are roughly equivalent to the market and can benefit from economic resuscitation.

Consuming the plate:

It is recommended to increase the holding to 2-3%.

The reason is that the defenses of the combination are improved, and the consumption upgrade bonus is shared.

New energy source board:

It is recommended to increase the holding time to about 2%.

The reason is to grasp the high-speed growth opportunities of the industry and improve the growth of the combination.

Medical plate:

It is recommended to increase the holding time to about 2%.

The reasons are that risks are dispersed, and the industry long-term growth potential is obtained.

Consider the new addition:

1 to 2 percent of high-end manufacturing plate

1-2% Of semiconductor plate

The reason is to further disperse risks and grasp the opportunity of emerging industries.

Operation advice:

And (3) reducing part of science and technology/internet stocks or funds, and transferring funds to consumption, new energy and medical plates.

Consider purchasing industry ETF or subject funds focused on high-end manufacturing or semiconductors.

The adjustment is carried out step by step, avoiding the combination structure from being changed greatly in a short period.

Note that the above advice is based on the current market situation and your risk preferences. In actual operation, please combine your specific situation with market judgment. Advice you to review and adjust the tile configuration regularly. "

6. Bin-holding personal diagnosis

And analyzing the funds of the users one by one according to the obtained user warehouse holding data.

For example:

USER ID USER 12345

Fund to be diagnosed C fund (stock type)

50,000 Parts of fund holding capacity

Current net value 1.2 yuan/serving

Input to Agent

6.1 Fund data query

And (3) inputting an Agent inquiry prompt sentence which is data such as historical net value, warehouse holding information, fund manager information, similar fund performances and the like of the inquired C fund. "

Agent output:

"C fund data query complete:

Historical net value data: net value of day of last 3 years

Currently ten heavy stock (stock list)

Fund manager Zhang San, practitioner for 5 years

Average performance of similar foundation, annual yield rate of 8%, maximum withdrawal of 15% "

6.2 Multidimensional analysis

And (3) inputting an Agent query prompt sentence, namely analyzing the foundation from multiple dimensions such as risk, income, management capability, market adaptability and future potential based on the collected data, and calculating indexes such as maximum withdrawal, fluctuation rate, summer average rate, annual income rate, excess income, relative reference win rate and the like. "

Agent output:

"C-based multi-dimensional analysis results:

Risk index:

Maximum withdrawal of 18%

Fluctuation ratio 15% (years)

Revenue index:

Annual gain rate of 10.5 percent

Ratio of Charpy 0.7

The relative performance:

excess profit 2.5% (average over the same class)

Relative to the reference winning rate 65%

Management capability assessment:

historical performance of the third party of the fund manager is better than average level of the same party

Stable investment style and good choice of strand

Market adaptability:

Performance balance under different market environments

The downlink risk control capability is stronger

Future potential:

Currently, the holding structure is reasonable, and most of heavy warehouse stock is industry tap

Investment theme accords with current market hot spot and long-term trend "

6.3 Predictive analysis

And (3) inputting an Agent query prompt sentence, namely predicting the possible performance of the fund for 6-12 months based on historical data and market prediction. "

Agent output:

"future 6-12 months forecast of foundation C:

Benchmark scenario expected annual yield of 9-11%

Optimistic scenario with expected annual profitability 12-15%

Pessimistic scenario expected annual yield of 5-7%

The main influencing factors are as follows:

Macroscopic economic trend

Industry policy changes

Market style conversion

Risk of concern:

heavy strands have higher concentration and may increase volatility

Market style, if converted to value, may affect short term performance'

6.4 Personalized advice Generation

And (3) inputting an Agent inquiry prompt sentence, namely generating personalized investment advice based on the fund analysis result by considering the investment target, the risk bearing capacity and the current market viewpoint of the user. "

Agent output:

"personalized advice for user_12345:

Considering your robust risk bearing capacity and the investment goal of long term wealth, the proposals for C funds are as follows:

Holding advice to maintain the current holding bin

The reason is that the risk and benefit characteristics of the fund are matched with the investment targets of the user, and the fund has excellent long-term performance

The configuration proportion can consider small increase of the warehouse holding, but the control is recommended to be within 10-15% of the total assets

The reason is that the dispersion is kept moderately, and the excessive occupation ratio of a single foundation is avoided

Fixed investment strategy suggesting a periodic fixed investment strategy

The reason is that the market fluctuation can be smoothed and the timing risk is reduced

Attention points:

Periodically checking fund reports, focusing on heavy stock changes

Concern about change of fund manager

Keeping in mind the impact of market style changes on fund performance

Rebalancing, the reason for rebalancing every half year or every year according to the overall asset allocation situation, is to ensure that the investment portfolio always meets the risk and benefit goal of you "

6.5 Report Generation

And (3) inputting an Agent query prompt sentence, namely generating a detailed and easily understood report based on the fund analysis result and the advice, and explaining the analysis logic and the advice reason. "

Agent output:

"C-fund diagnostic report

The respectful USER 12345,

Based on a comprehensive analysis of the C foundation, we generated the following diagnostic report for you:

fund profile

The C fund is a stock fund, which is established in year and managed by a Zhang Sanjiu manager with abundant experience. The fund is primarily invested in [ primary investment direction ].

Performance analysis

Annual gain rate 10.5%, higher than 8% of average of the same class

Risk control, maximum withdrawal 18%, slightly higher than the average of the same class, but within acceptable limits, gain after risk adjustment, a summer ratio of 0.7, performed well

Investment style

The fund adopts [ describe investment style ] and has balanced performance in different market environments. Fund managers are good at selecting strands, and heavy warehouse strands are mostly industry tap enterprises.

Future prospect

Based on current market trends and the structure of the fund warehouse, we expect the fund to continue to perform well for the next 6-12 months, with expected annual profitability between 9-11%.

Personalized advice

Considering your robust risk preference and long term investment goals, we propose:

maintaining the current hold, a small increase can be considered

By means of a regular quota investment strategy

Controlling the ratio of the foundation to the total assets to be not more than 15%

Periodic attention to fund reports and market changes

Risk prompt

Stock markets have a risk of fluctuation, and the net value of funds can fluctuate with the market

Past performance does not represent future performance

Please make investment decisions based on own risk bearing capacity

Conclusion:

Foundation C performs well, matching your investment goals. Through reasonable configuration and regular monitoring, the method is expected to bring long-term stable benefits to your investment portfolio.

If you have any doubt about the report, welcome consultation at any time. You can make a congratulation on the successful investment- "

The embodiments of the present invention also provide a computer-readable storage medium capable of implementing all the steps of the method in the above embodiments, the computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements all the steps of the method in the above embodiments.

The embodiment of the application also provides an electronic device for executing the method, and the electronic device 300 is shown in fig. 3, and comprises a processor 301 and a memory 303 as an implementation device of the method. The processor 301 is coupled to a memory 303, such as via a bus 302. Optionally, the electronic device 300 may also include a transceiver 304. It should be noted that, in practical application, the transceiver 304 is not limited to one. It should be understood that the illustrated structure of the embodiment of the present application does not constitute a specific limitation on the specific structure of the electronic device 300. In other embodiments of the application, electronic device 300 may include more or less components than illustrated, or certain components may be combined, or certain components may be split, or different arrangements of components. The illustrated components may be implemented in hardware, software, or a combination of software and hardware. Optionally, the electronic device may further comprise a display screen 305 for displaying images or receiving user operation instructions if necessary.

The processor 301 is applied to the embodiment of the present application, and is configured to implement the method shown in the above embodiment of the method. Transceiver 304 may include a receiver and a transmitter, with transceiver 304 being employed in embodiments of the present application to perform functions that enable an electronic device of embodiments of the present application to communicate with other devices.

The Processor 301 may be a CPU (Central Processing Unit ), general purpose Processor, DSP (DIGITAL SIGNAL Processor, data signal Processor), ASIC (Application SPECIFIC INTEGRATED Circuit), FPGA (Field Programmable GATE ARRAY ) or other programmable logic device, transistor logic device, hardware component, or any combination thereof. Which may implement or perform the various exemplary logic blocks, modules and circuits described in connection with this disclosure. Processor 301 may also be a combination that implements computing functionality, e.g., comprising one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

The processor 301 may also include one or more processing units, for example, the processor 301 may include an application processor (application processor, AP), a modem processor, a graphics processor (graphics processingunit, GPU), an image signal processor (IMAGE SIGNAL processor, ISP), a controller, a memory, a video codec, a digital signal processor (DIGITAL SIGNAL processor, DSP), a baseband processor, and/or a neural network processor (Neural-network Processing Unit, NPU), etc. Wherein the different processing units may be separate devices or may be integrated in one or more processors. The controller may be a neural hub and a command center of the electronic device 300, among others. The controller can generate operation control signals according to the instruction operation codes and the time sequence signals to finish the control of instruction fetching and instruction execution. A memory may also be provided in the processor 301 for storing instructions and data. In some embodiments, the memory in the processor 301 is a cache memory. The memory may hold instructions or data that the processor 301 has just used or recycled. If the processor 301 needs to reuse the instruction or data, it may be called directly from the memory. Repeated accesses are avoided and the latency of the processor 301 is reduced, thus improving the efficiency of the system.

Processor 301 may perform the methods provided by embodiments of the present application. The processor 301 may include different devices, such as a CPU and a GPU, where the CPU and the GPU may cooperate to perform the method provided by the embodiments of the present application, such as a portion of the algorithm being performed by the CPU and another portion of the algorithm being performed by the GPU, so as to obtain a faster processing efficiency.

Bus 302 may include a path to transfer information between the components. Bus 302 may be a PCI (PERIPHERAL COMPONENT INTERCONNECT, peripheral component interconnect standard) bus or an EISA (Extended Industry Standard Architecture ) bus, or the like. Bus 302 may be divided into an address bus, a data bus, a control bus, and the like. For ease of illustration, only one thick line is shown in fig. 3, but not only one bus or one type of bus.

The Memory 303 may be, but is not limited to, ROM (Read Only Memory) or other type of static storage device that can store static information and instructions, RAM (Random Access Memory ) or other type of dynamic storage device that can store information and instructions, EEPROM (ELECTRICALLY ERASABLE PROGRAMMABLE READ ONLY MEMORY ), CD-ROM (Compact Disc Read Only Memory, compact disc Read Only Memory), high-speed random access Memory, nonvolatile Memory such as at least one magnetic disk storage device, flash Memory device, universal flash Memory (universal flash storage, UFS), or other optical disk storage, optical disk storage (including compact disc, laser disc, optical disc, digital versatile disc, blu-ray disc, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and that can be accessed by a computer.

Optionally, the memory 303 is used for storing application program codes for executing the inventive arrangements, and is controlled by the processor 301 for execution. The processor 301 is configured to execute application code stored in the memory 303 to implement the methods provided in any of the embodiments of the present application.

Memory 303 may be used to store computer executable program code that includes instructions. The processor 301 executes instructions stored in the memory 303 to thereby perform various functional applications and data processing of the electronic device 300. The memory 303 may include a stored program area and a stored data area. The storage program area may store, among other things, an operating system, code for an application program, and the like. The storage data area may store data created during use of the electronic device 300 (e.g., images, video, etc. captured by a camera application), and so on.

The memory 303 may also store one or more computer programs corresponding to the methods provided by the embodiments of the present application. The one or more computer programs are stored in the memory 303 and configured to be executed by the one or more processors 301, the one or more computer programs comprising instructions that can be used to perform the various steps in the respective embodiments described above.

Of course, the codes of the method provided by the embodiment of the application can also be stored in an external memory. In this case, the processor 301 may run the code of the method stored in the external memory through the external memory interface, and the processor 301 may control the running process.

The display screen 305 includes a display panel. The display panel may employ a Liquid Crystal Display (LCD) screen (liquid CRYSTAL DISPLAY), an organic light-emitting diode (OLED), an active-matrix organic light-emitting diode (AMOLED) or an active-34 diode, a flexible light-emitting diode (FLED), miniled, microLed, micro-oLed, a quantum dot light-emitting diode (quantum dot lightemitting diodes, QLED), or the like. In some embodiments, the electronic device 300 may include 1 or N display screens 305, N being a positive integer greater than 1. The display screen 305 may be used to display information entered by a user or provided to a user as well as various graphical user interfaces (GRAPHICAL USER INTERFACE, GUI). For example, the display screen 305 may display photographs, videos, web pages, or files, etc.

The electronic device provided by the embodiment of the present application is suitable for any embodiment of the above method, so the beneficial effects that can be achieved by the electronic device can refer to the beneficial effects in the corresponding method provided above, and will not be described herein.

It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a system for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks. While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiments and all such alterations and modifications as fall within the scope of the invention.

The foregoing is only a preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions easily contemplated by those skilled in the art within the scope of the present invention should be included in the scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A product warehouse-holding combination analysis method based on a large model agent is characterized by comprising the following steps:

2. The method of claim 1, wherein the training of the large model agent comprises:

3. The method of claim 1, wherein the acquiring and preprocessing the taken data and the transaction data with the large model agent comprises:

collecting data through an API interface or a data export function;

4. The method of claim 1, wherein inputting the preprocessed out-of-stock data and transaction data into a large model agent for analysis comprises:

5. The method of claim 4, wherein the acquiring asset characteristic data in the taken-off-warehouse data, performing identification classification and duty cycle calculation on the asset characteristic data by a large model agent, comprises:

Acquiring detailed information of funds held by a user;

the large model agent evaluates risk and benefit features of holding funds.

6. The method of claim 1, wherein the method further comprises:

Collecting new data and feedback data;

training and fine-tuning a large model agent;

Updating the knowledge base and the analysis model.

7. Product warehouse-holding combination analysis system based on large model agent, which is characterized by comprising:

8. A computer readable storage medium, characterized in that the storage medium has stored thereon a computer program which, when executed by a processor, implements the method of any of claims 1 to 6.

9. An electronic device comprising a processor and a memory;

The memory is used for storing program instructions;

The processor being adapted to perform the method of any one of claims 1 to 6 by invoking program instructions.

10. A computer program product comprising a computer program and/or instructions which, when executed by a processor, implement the steps of the method of any one of claims 1 to 6.