CN113515638B - Student clustering-oriented research interest mining method and device and storage medium - Google Patents

Abstract

The invention provides a research interest mining method, device and storage medium oriented to scholar clustering. The method includes the following steps: constructing an academic metadata set based on multi-source scholar-related academic data; inputting the academic metadata as input data into a pre-established In the research interest mining model of the The semantics of interest expresses scholar clustering and obtains scholar clustering results; the research interest mining model shares the same topic distribution for the data of scholars from the same data source and belongs to the same professional field. In the research interest mining model, the professional field-topic distribution are modeled as Dirichlet distributions, topic-English word distribution, topic-Chinese word distribution, and topic-scholar distribution are modeled as multinomial distributions.

Description

Research interest mining method, device and storage medium for scholar clustering

technical field

The invention relates to the technical field of big data, in particular to a research interest mining method, device and storage medium oriented to scholar clustering.

Background technique

Academic data such as scholars, research projects, and dissertations all have their own areas of expertise. For example, some data belongs to software engineering and some data belongs to artificial intelligence. The research contents of different professional fields are different, and the data of the same professional field often have a common topic distribution. Targeting scholar academic data, discovering scholars' research interests from academic data, and clustering scholars according to their research interests is important for many tasks, such as selecting collaborators for scholars, selecting reviewers for journals, and selecting for governments expert. Unlike general data, academic data related to scholars' interests has its own unique properties. On the one hand, it is multi-source, and the funded projects and published papers of scholars can reflect their research interests. On the other hand, it is multilingual and essays and grant applications are usually written in the native language or English. Also, academic data increases over time.

Although there are many methods to deal with the user clustering problem, most of them are only applicable to a single source of data. For example, users can be represented as vectors using User Clustering Topic Models (UCT) or deep learning methods, and then users can be clustered using typical clustering algorithms. As an example, UCT uses information from a single data source to model user interests, and obtains semantic representations of user interests in different time slices by slicing time. Then, the K-means clustering algorithm is used to cluster the semantic representation results of users' interests, and finally the clustering results of users are obtained. This method only uses data from a single source, and only uses a single language. When it is used in the semantic representation of scholar user interests, its representation quality needs to be improved. In addition, K-Means has certain limitations, especially the need to set the K value in advance, which makes the robustness of the user clustering effect relatively limited.

Replacing single-source data with multi-source data can mine and represent user interests more comprehensively and accurately. The effectiveness of clustering depends on the quality of semantic representation of user interests. However, most scholars' related research only uses papers as materials, ignoring scientific research project data. Furthermore, multi-source data cannot be directly merged together. This is because there is a big difference in the amount of data from different sources. For example, the number of funded projects hosted by scholars is often small, and the number of published papers is dozens or even hundreds of times the number of funded projects by scholars. If you simply mix data from different sources, the impact of small sample sources will be overwhelmed by large sample sources. Therefore, information integration of multi-source data is a challenge.

Another challenge facing cluster scholars is how to fully exploit the rich semantics of multilingual data. For large-scale multilingual corpora, translating multiple languages into one language is impractical because the translation process introduces errors and consumes time and effort. Typical multilingual data processing methods are based on probabilistic topic models or word embedding methods. But these methods all require strong associations between different languages, such as sentence pairs, word pairs, or document pairs with translation correspondences. However, academic data presented by scholars in different languages lacks a clear translation correspondence. Therefore, if multilingual academic data is to be fully utilized, it is necessary to be able to find bridges between different languages.

For example, the User Collaborative Interest Tracking Model (UCIT) uses data from different sources to track user interests. The model uses both the user's own data and the user's fan data to mine user interests, which has a greater effect than only using a single source. Although this model utilizes multi-source data, it only utilizes a single language and does not use different languages to model users at the same time. And in the user clustering stage, K-means is also used, so it is also affected by the limitations of K-means.

In addition, similar to the transfer of daily interests, the research interests of scholars also change dynamically over time, so when clustering scholars, changes in scholars’ interests also need to be considered. Existing dynamic representation models can be divided into two categories, one treats time as continuous and processes data online, and the other treats time as discrete and processes data by dividing time slices, that is, batch processing. . However, none of these dynamic representation models can track the research interests of scholars from academic data. On the one hand, the publication time of the paper and the project application time are not continuous, which is not suitable for processing academic data in an online manner. On the other hand, the existing batch processing methods are only applicable to single-source data or monolingual data, so they cannot be directly used to obtain dynamic semantic representations of scholars' research interests.

How to effectively use multi-source and multi-language academic data to achieve accurate and dynamic scholar clustering, so as to realize research interest mining, is a problem to be solved.

SUMMARY OF THE INVENTION

In view of this, the purpose of the present invention is to provide a research interest mining method, device and storage medium for scholar clustering, so as to solve one or more problems existing in the prior art.

One aspect of the present invention provides a research interest mining method for scholar clustering, the method comprising the following steps:

An academic metadata set is constructed based on academic data related to scholars from multiple sources, and each piece of academic metadata in the academic metadata set includes the following information: professional field information, data source information, scholar information and text content information;

Input at least part of the academic metadata in the constructed academic metadata set as input data into the pre-established research interest mining model, and obtain the scholar's interest semantic representation by sampling the topic model, and the scholar's interest semantic representation includes the professional field- topic distribution, topic-English word distribution, topic-Chinese word distribution, and topic-scholar distribution; wherein the research interest mining model shares the same topic distribution for data from scholars from the same data source and belonging to the same professional field, and , in the research interest mining model, the professional field-topic distribution is modeled as Dirichlet distribution, and the topic-English word distribution, topic-Chinese word distribution and topic-scholar distribution are modeled as multinomial distributions;

Based on the obtained semantic representation of scholars' interest, scholars are clustered, and the results of scholars' clustering are obtained.

In some embodiments of the present invention, the academic data includes fund project data and dissertation data; the professional field-topic distribution includes professional field-topic distribution of scientific research projects and professional field-topic distribution of dissertation data.

In some embodiments of the present invention, the method further includes the step of constructing a research interest mining model, the step comprising: determining the professional field to which the academic metadata belongs; The topics are sampled collaboratively in the topic distribution of papers and belong to a certain professional field to obtain academic metadata topics; English words, Chinese words and scholars are generated according to topic-English word distribution, topic-Chinese word distribution, and topic-scholar distribution .

In some embodiments of the present invention, constructing an academic metadata set includes: judging the source of each piece of academic metadata, and determining whether it is funded project data or dissertation data; determining the scholar to which the current academic metadata belongs; extracting each funded project Or the text information of the paper; determine whether the text information of the fund project or paper is Chinese or English; determine the time period to which the current academic metadata belongs, and classify the current academic metadata into the academic metadata subset corresponding to the time period, by This construct includes a collection of academic metadata for the subset of academic metadata.

In some embodiments of the present invention, at least part of the academic metadata in the constructed academic metadata set is input into a pre-established research interest mining model as input data, and the semantic representation of scholar interest is obtained by sampling the topic model, Including: sampling the academic metadata from scientific research projects and belonging to a predetermined professional field to obtain the professional field-topic distribution of scientific research projects; sampling academic metadata from papers and belonging to a predetermined professional field to obtain the professional field-topic distribution of paper data; Each predetermined topic is sampled to obtain the corresponding topic-Chinese word distribution, topic-scholar distribution and topic-English word distribution; determine whether it is Chinese or English for the text content in each academic metadata; in the case that the text content is Chinese , extract the double words and take the following actions for each double word: sample the topic of the Chinese double word, independently sample the two words, and sample the author of the double word; in the case of the text content in English, sample the following for each word Action: Sample English topics, sample each English word, and sample the author of the English word.

In some embodiments of the present invention, at least part of the academic metadata in the constructed academic metadata set is input into a pre-established research interest mining model as input data, and scholars are obtained by sampling the topic model multiple times iteratively. Interest semantic representation, further comprising: inputting academic metadata in academic metadata subsets corresponding to different time periods into the research interest mining model as input data to obtain scholar interest semantic representation at different times; The clustering of representations includes: using a single-pass clustering algorithm to cluster the acquired semantic representations of scholars' interests at different times.

In some embodiments of the present invention, the parameters to be estimated in the research interest mining model are estimated by using a Gibbs sampling method;

The sampling of Chinese double words adopts the following formula:

The following formula is used for the sampling of each English word:

表示中文主题Z ^b 下的中文词i，

表示除了第i个单词外的所有中文主题的集合；

表示的英文主题Z ^b 下的英文单词i，

表示除了第i个单词外的所有英文主题的集合；

表示第p个专业领域下，除了词i之外，科研项目数据里属于主题z的所有中英文词总数；

表示第p个专业领域下，除了词i之外，论文数据里属于主题z的中英文词总数；

分别表示主题z的中文主题Z ^b 下除中文词i外，所有科研项目和论文数据里中文单词

出现的次数；

表示主题z下除了词i之外，所有科研项目和论文数据里属于学者s _i 的中英文词出现的次数；

表示主题z的英文主题Z ^e 下除英文单词i外，所有科研项目和论文数据里英文单词

出现的次数；

,

分别为科研项目数据的主题分布以及论文数据的主题分布；

为符合狄利克雷分布的主题-英文单词分布，

为符合狄利克雷分布的主题-学者分布，

为符合狄利克雷分布的主题-中文单词分布；R表示不同学者的个数，B表示双词的集合，S表示学者的集合，W^b表示不同中文单词的个数，W^e表示不同英文单词的个数，

为平衡因子；where Z represents the total number of topics,

represents the Chinese word i under the Chinese topic Z ^b ,

Represents the set of all Chinese topics except the ith word;

represents the English word i under the English topic Z ^b ,

Represents the set of all English topics except the ith word;

Indicates the total number of all Chinese and English words belonging to topic z in the scientific research project data except word i in the pth professional field;

Indicates the total number of Chinese and English words belonging to topic z in the paper data except word i in the pth professional field;

In addition to the Chinese word i , the Chinese words in all scientific research projects and paper data under the Chinese topic Z ^b respectively represent the topic z .

the number of occurrences;

Indicates the number of _{occurrences of Chinese and English words belonging to scholar si} in all scientific research projects and paper data except word i under topic z ;

Except for the English word i , the English words in all scientific research projects and thesis data under the English topic Z ^e representing the topic z

the number of occurrences;

,

are the subject distribution of scientific research project data and the subject distribution of paper data;

For the topic-English word distribution that conforms to the Dirichlet distribution,

To fit the subject-scholar distribution of the Dirichlet distribution,

It is the topic-Chinese word distribution that conforms to the Dirichlet distribution; R represents the number of different scholars, B represents the set of double words, S represents the set of scholars, W ^b represents the number of different Chinese words, and We ^e represents different English words number of ,

is the balance factor;

After the iterative operation of the research interest mining model, the following estimated parameters are obtained:

；

;

；

;

；

;

；

;

；

;

表示科研项目的专业领域为p主题为z的专业领域-主题分布；

表示论文数据的专业领域为p主题为z的专业领域-主题分布；

表示第p个专业领域下，科研项目数据里属于主题z的所有中英文词总数；

表示第p个专业领域下，论文数据里属于主题z的所有中英文词总数；

表示第z^b个主题下，所有科研项目和论文数据里中文单词出现的次数；

表示主题为z的中文双词的主题-中文单词分布；

表示第z^e个主题下，所有科研项目和论文数据里英文单词出现的次数；

表示主题为z的英文单词w ^e 的主题-英文单词分布；

表示主题为z、学者为s的主题-学者分布；

表示主题z下所有科研项目和论文数据里属于学者s的中英文词出现的次数。in,

Indicates that the professional field of the scientific research project is the professional field of p and the topic is z - topic distribution;

Indicates that the professional field of the paper data is p and the topic is the professional field of z - topic distribution;

Indicates the total number of all Chinese and English words belonging to topic z in the scientific research project data under the pth professional field;

Indicates the total number of all Chinese and English words belonging to topic z in the paper data under the pth professional field;

Indicates the number of occurrences of Chinese words in the data of all scientific research projects and papers under the z- ^th topic;

Represents the topic-Chinese word distribution of Chinese double words with topic z;

Represents the number of occurrences of English words in all scientific research projects and paper data under the z ^e -th topic;

Represents the topic-English word distribution of the English word we with topic ^z ;

Represents the topic-scholar distribution with topic z and scholar s;

Indicates the number of occurrences of Chinese and English words belonging to scholar s in all scientific research projects and paper data under topic z.

In some embodiments of the present invention, the method further includes: using the topic distribution of the previous period as the prior distribution of the current period to calculate the prior parameters of the current period based on the following formula:

为衰减因子，

表示在t-1时刻，第p个专业领域下，科研项目数据里属于主题

的所有中英文词总数；

表示在t-1时刻，科研项目数据的专业领域主题分布；

表示在

时刻，科研项目数据的专业领域主题分布；

表示在

时刻，第

个专业领域下，论文数据里属于主题

的所有中英文词总数；

表示在

时刻，论文数据的专业领域主题分布；

表示在

时刻，论文数据的专业领域主题分布；

表示在

时刻，第

个主题下，所有科研项目和论文数据里中文单词

出现的次数；

表示在

时刻，所有科研项目和论文数据的主题中文单词分布；

表示在

时刻，所有科研项目和论文数据的主题中文单词分布；

表示在

时刻，第

个主题下，所有科研项目和论文数据里英文单词

出现的次数；

表示在

时刻，所有科研项目和论文数据的主题英文单词分布；

表示在

时刻，所有科研项目和论文数据的主题英文单词分布；

表示在

时刻，第

个主题下，所有科研项目和论文数据里学者

出现的次数；

表示在

时刻，所有科研项目和论文数据的主题学者分布；

表示在

时刻，所有科研项目和论文数据的主题学者分布。in,

is the attenuation factor,

Indicates that at time t-1 , under the pth professional field, the scientific research project data belongs to the theme

The total number of all Chinese and English words of ;

Represents the subject distribution of scientific research project data in professional fields at time t-1 ;

expressed in

Time, subject distribution of scientific research project data in professional fields;

expressed in

time, the

Under a professional field, the thesis data belongs to the topic

The total number of all Chinese and English words of ;

expressed in

Moment, the subject distribution of the professional field of the paper data;

expressed in

Moment, the subject distribution of the professional field of the paper data;

expressed in

time, the

Under this topic, Chinese words in all scientific research projects and paper data

the number of occurrences;

expressed in

Time, topic Chinese word distribution of all scientific research projects and dissertation data;

expressed in

Time, topic Chinese word distribution of all scientific research projects and dissertation data;

expressed in

time, the

Under this topic, English words in all scientific research projects and paper data

the number of occurrences;

expressed in

Time, the subject English word distribution of all scientific research projects and paper data;

expressed in

Time, the subject English word distribution of all scientific research projects and paper data;

expressed in

time, the

Under this topic, scholars in all scientific research projects and dissertation data

the number of occurrences;

expressed in

Time, the distribution of subject scholars for all scientific research projects and dissertation data;

expressed in

Moment, subject scholar distribution of all research projects and dissertation data.

Another aspect of the present invention provides a research interest mining device for scholar clustering, the device includes a processor and a memory, where computer instructions are stored in the memory, and the processor is configured to execute the data stored in the memory. Computer instructions which, when executed by a processor, implement the steps of the method as previously described.

In yet another aspect of the present invention, there is also provided a computer storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of the aforementioned method are implemented.

The research interest mining method and device for scholar clustering provided by the embodiments of the present invention can effectively utilize multi-source and multilingual academic data to obtain high-quality semantic representation of scholars' interests through the research interest mining model, improve the clustering accuracy, and achieve precise Dynamic clustering of scholars, so as to realize research interest mining.

Further, the present invention may be able to obtain scholar clustering results in different time periods according to changes in time.

Additional advantages, objects, and features of the present invention will be set forth in part in the description that follows, and in part will become apparent to those of ordinary skill in the art upon study of the following, or may be learned from practice of the invention. The objectives and other advantages of the invention may be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

Those skilled in the art will appreciate that the objects and advantages that can be achieved with the present invention are not limited to those specifically described above, and that the above and other objects that can be achieved by the present invention will be more clearly understood from the following detailed description.

Description of drawings

The accompanying drawings described herein are used to provide a further understanding of the present invention, and constitute a part of the present application, and do not constitute a limitation to the present invention. In the attached image:

FIG. 1 is a schematic flowchart of a research interest mining method for scholar clustering according to an embodiment of the present invention.

FIG. 2 is a schematic diagram of a research interest mining model in another embodiment of the present invention.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but not to limit the present invention.

Here, it should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, only the structures and/or processing steps closely related to the solution according to the present invention are shown in the drawings, and the related structures and/or processing steps are omitted. Other details not relevant to the invention.

It should be emphasized that the term "comprising/comprising" when used herein refers to the presence of a feature, element, step or component, but does not exclude the presence or addition of one or more other features, elements, steps or components.

Aiming at the problem that the existing user interest mining methods cannot effectively utilize multi-source and multi-language data, so that accurate and dynamic scholar clustering cannot be realized, the present invention uses the multi-source and multi-language information of scholars to propose a new research interest mining model. (Research Interest Mining Module, RIMM), can obtain high-quality scholars' research interest semantic representation, so as to achieve accurate scholar clustering.

The RIMM proposed by the present invention can dynamically cluster scholars from academic data. In RIMM, it is assumed that data from different sources belong to different topic distributions, and in order to obtain the semantic representation of scholars' interests by using data from different sources at the same time, the present invention designs a balance factor to synergistically extract topics from topic distributions. In order to bridge the lexical gap of different languages, the present invention also sets different languages belonging to the same professional field and from the same data source to share the same topic distribution, thereby obtaining the common semantics of different languages and generating the topic distributions of the model. Through the above operations, the present invention can make full use of multi-source and multi-language data to achieve accurate scholar interest mining. In addition, the present invention also takes the topic distribution of historical moments as prior knowledge, executes RIMM on different time slices, and can obtain the semantic representation of scholars' interests at different moments, and uses a single-pass according to the semantic representation of interest. Clustering algorithms cluster scholars.

Research interest can be reflected in many ways. From the perspective of data sources, both scientific research project information (such as applications and final reports) and papers can reflect the interests of scholars. In addition, academic data is multilingual. Therefore, the present invention maps users, project information, papers, Chinese and English to the same topic space, and expresses the scholar's interest in a multi-dimensional (eg K-dimensional) topic vector, where each K value can reflect the scholar's interest in the topic The probability that the topic is of interest.

FIG. 1 is a schematic flowchart of a research interest mining method for scholar clustering in an embodiment of the present invention. As shown in FIG. 1 , the method includes the following steps S110-S130:

Step S110 , constructing an academic metadata set based on academic data related to multi-source scholars.

Each piece of academic metadata in the academic metadata set may include the following information: professional field information, data source information, scholar information and text content information.

Scientific research projects and papers belong to the research materials of the present invention. Although it is difficult to obtain the full text of a research project application or a research project final report, their titles, keywords, and abstracts can be easily obtained. For the paper, although the full text of the paper is easy to obtain, it requires a lot of resources to obtain the full text of the paper, and the title and keyword abstract of the paper are enough to tap the interest of scholars. Therefore, in this embodiment of the present invention, the data of scientific research projects (such as funded projects) and thesis data can be used as multi-source scholar-related academic data, that is, academic metadata may come from both scientific research projects and papers, and further, Based on academic data related to scholars from multiple sources, the concept of academic metadata is defined. The defined academic metadata can include the following information: professional field information, data source information, scholar information and text content information. As an example, the present invention can use a quadruple to represent academic metadata m = ( p _m , d _m , s _m , t _m ), where p _m represents a professional field, d _m represents a data source, s _m represents a scholar, t _m represents the text content, and m represents the mth academic metadata. The text content of the academic metadata includes titles, keywords and/or abstracts of funded projects or papers, but is not limited thereto.

In some embodiments of the present invention, academic data of Chinese scholars, such as fund project data and dissertation data, are collected as academic data, and then the National Natural Science Foundation of China application code is used to divide the scholar's professional field to obtain professional field information. Specifically, the present invention first obtains the application code of the scientific research project hosted by the scholar, and sets the code as the professional field of the scholar during the execution of the scientific research project.

In some embodiments of the present invention, the academic metadata set may include multiple academic metadata subsets, and each academic metadata subset may include academic metadata based on different time periods.

Step S120 , input at least part of the academic metadata in the constructed academic metadata set as input data into a pre-established Research Interest Mining Model (RIMM), and obtain a semantic representation of scholar interest by sampling this topic model.

The semantic representation of scholars’ interests can include professional field-topic distribution, topic-English word distribution, topic-Chinese word distribution, and topic-scholar distribution; among them, professional field-topic distribution can include professional field-topic distribution of scientific research projects and the distribution of paper data. Areas of Expertise - Subject distribution.

In the RIMM constructed by the embodiment of the present invention, it is assumed that the data of scholars from the same data source and belonging to the same professional field share a subject distribution, and the data from different sources belong to different subject distributions, for example, belong to the same professional field and come from the same subject distribution Different languages of the data source are set to share the same topic distribution, so as to obtain the common semantics of different languages, and generate the topic distribution of the model, thereby bridging the lexical gap of different languages.

、论文数据的专业领域-主题分布

分别建模为具有参数

,

的狄利克雷分布。主题-英文单词分布

、主题-学者分布

、主题-中文单词分布

分别建模为具有参数

,

,

的多项式分布。由此可知，通过本发明提出的研究兴趣挖掘模型得到的主题分布是科研项目、论文、中文、英文的混合。In addition, in the RIMM constructed by the embodiment of the present invention, the professional field-topic distribution is modeled as a Dirichlet distribution, and the topic-English word distribution, the topic-Chinese word distribution, and the topic-scholar distribution are modeled as polynomials distributed. Fig. 2 is an example of a graph model of the research interest mining model (or a multi-feature topic model of social network) proposed by the present invention. As shown in Fig. 2, the professional field-topic distribution of scientific research project data

, the professional field of the paper data - topic distribution

separately modeled as having parameters

,

The Dirichlet distribution. Topic - English word distribution

, subject-scholar distribution

, subject-Chinese word distribution

separately modeled as having parameters

,

,

the multinomial distribution of . It can be seen that the topic distribution obtained by the research interest mining model proposed by the present invention is a mixture of scientific research projects, papers, Chinese and English.

The symbols and descriptions involved in Figure 2 are shown in Table 1 below.

Table 1. Symbols and their descriptions

In this embodiment of the present invention, all academic metadata in the constructed academic metadata set may be input into a pre-established research interest mining model (RIMM) as input data, or part of the constructed academic metadata set may be Academic metadata is fed into a pre-built Research Interest Mining Model (RIMM) as input data. The portion of academic metadata may be, for example, one or more subsets of academic metadata in a set of academic metadata.

Step S130, perform scholar clustering based on the obtained scholar's interest semantic representation, and obtain a scholar clustering result.

It can be seen from the above that the semantic representation of scholars can be obtained by studying the interest mining model. However, the representation is static and is only a topic representation within a period of time. Therefore, in the present invention, in order to be able to cluster scholars according to the semantic representation of scholars' interests in different time periods, it is necessary to track the dynamic changes of scholars' interests. Although the interests of scholars are dynamic, there is also a certain persistence. Therefore, the present invention does not simply slice the corpus, but uses the historical interests of scholars to help infer the topic distribution of the current period. In order to maintain the continuity of scholars' interests in different periods, the present invention uses the topic distribution of the previous period (eg, the t-1th period) as the prior distribution of the current period (eg, the tth period).

In the embodiment of the present invention, step S130 is to use a single-pass clustering algorithm to cluster scholars.

The single-pass clustering algorithm is usually the classic method for streaming data clustering. For the data streams that arrive in sequence, the aggregation algorithm processes one data at a time in the input order, and judges the data as an existing class or creates a new data class according to the matching degree between the current data and the existing class to realize streaming Incremental and dynamic clustering of data. In the embodiment of the present invention, by using single-pass to cluster the acquired scholar's interest semantic representation at different times, dynamic scholar clustering results can be obtained, which overcomes the traditional K-means clustering algorithm in scholar clustering. limitations.

The present invention is the research interest mining method for scholar clustering as above, which can effectively utilize multi-source and multilingual academic data to obtain high-quality semantic representation of scholars' interests through the research interest mining model, improve the clustering accuracy, and realize accurate and dynamic scholar clustering. class, so as to realize research interest mining.

In some embodiments of the present invention, academic metadata may come from both scientific research projects and papers. For each piece of academic metadata, the process of constructing a RIMM from academic metadata is as follows:

First, obtain the professional field p to which the piece of metadata belongs;

和来自论文且属于专业领域p的主题分布

中协同地对主题进行采样，得到主题z，其中，

和

分别表示科研项目的专业领域p的学术元数据的专业领域-主题分布和论文数据的专业领域-主题分布。Then, from the distribution of topics from research projects that belong to the area of expertise p

and the distribution of topics from papers that belong to the area of expertise p

The topics are sampled collaboratively in , to get the topic z, where,

and

Respectively represent the professional field-topic distribution of academic metadata and the professional field-topic distribution of dissertation data of the professional field p of the scientific research project.

Then, English words, Chinese words, and scholars are generated according to topic-English word distribution, topic-Chinese word distribution, and topic-scholar distribution.

In addition, since Chinese data is relatively sparse, in order to alleviate the sparseness of Chinese data, the present invention also introduces a two-word model, setting two Chinese words in the same academic metadata to share the same topic.

In some embodiments of the present invention, the step S110 of constructing the academic metadata set may specifically include the following steps:

(1) Determine the professional field of each piece of academic metadata;

(2) Determine the source of each piece of academic metadata, that is, determine whether it is fund project data or dissertation data;

(3) Extract the scholars to which the current academic metadata belongs; the scholars to which the academic metadata belongs may include the project host or the author of the paper, etc. That is, in this step, the moderator of each scientific research project or the author of the paper can be extracted, that is, to determine which scholar the academic metadata belongs to;

(4) Extract the text information (such as title, keywords and/or abstract, etc.) of each funded project or paper;

(5) Determine whether the text information of the funded project or thesis is in Chinese or English;

(6) Determine the time period to which the academic metadata belongs, and classify the current academic metadata into the academic metadata subset corresponding to the time period to which the academic metadata belongs, thereby constructing an academic metadata set including multiple academic metadata subsets.

For example, after crawling a number of fund projects (scientific research projects) information, the academic metadata describing the scientific research project is obtained by preprocessing the scientific research project data, and then the host of the above-mentioned scientific research project is used as the research object, and the Extract the text information and project effective time information of these scientific research projects; further crawl the papers published by the above scholars from Baidu Scholar, and extract the text information (such as titles, abstracts and keywords) of these papers and the publication time of the papers. Based on the obtained information, a collection of academic metadata can be constructed.

In addition, the above preprocessing of academic metadata may include: word segmentation, stop word removal, high frequency word removal and other data cleaning for the text, and then two words appearing simultaneously in each piece of Chinese academic metadata. Two combinations to extract double words. Obtain the scholar id number, code the scholar, etc.

After that, at least part of the data in the constructed academic metadata set is input into the research interest mining model (RIMM), that is, step S120 is executed. More specifically, in step S120, these processed academic metadata sets are input into the proposed research interest mining model RIMM as input data, and the number of topics, the number of iterations, and the hyperparameters (before starting the machine learning process) can be determined. The parameters of setting value) are set according to experience, and the semantic representation of scholar interest related to the topic is finally obtained by repeatedly performing RIMM sampling for multiple iterations, and the semantic representation of scholar interest includes scientific research project professional field-topic distribution, topic-English words Distribution, topic-Chinese word distribution, topic-scholar distribution.

的情况下，可分别对不同时段t的学术元数据子集

通过多次迭代来执行RIMM，由此就可以得到不同时段的学者兴趣表示。As an example, a subset of academic metadata for different time periods t is included in the academic metadata set

In the case of , the subsets of academic metadata for different periods t can be

The RIMM is performed through multiple iterations, so that the interest representations of scholars at different time periods can be obtained.

In addition, in some embodiments of the present invention, a single-pass clustering algorithm may be run on the scholar's interest semantic representation of each time period, and finally the scholar clustering results of multiple time periods (eg, T time periods) are obtained.

In the embodiment of the present invention, the execution process of the research interest mining model may include the following steps:

；In step S1, a topic distribution is sampled for academic metadata from scientific research projects and belonging to the professional field p :

;

；Step S2, sample a topic distribution for academic metadata from papers and belonging to the professional domain p :

;

和一个主题-学者分布

以及主题-英文词分布

；Step S3, for each topic, sample a topic-Chinese word distribution

and a subject-scholar distribution

and topic-English word distribution

;

Step S4, according to the text content in each piece of academic metadata, determine whether it is Chinese or English:

（b _i 表示第i个双词，

表示属于专业领域p的第m条元数据中的双词集合）：(1) If it is Chinese, extract double words, for each double word,

( b _i represents the i -th double word,

represents the set of bigrams in the m-th metadata belonging to the domain of expertise p ):

；i. Sampling a topic

;

表示第i个中文双词的主题，

表示多项式分布，

为平衡因子，用来调节学者的兴趣受科研项目和论文的影响的大小。in,

Represents the subject of the i -th Chinese double word,

represents a multinomial distribution,

It is a balance factor, which is used to adjust the influence of scholars' interests by scientific research projects and papers.

；ii. Independently sample two words in Chinese double words

;

；iii. Scholars who sample Chinese double words

;

：(2) If it is in English, for each word,

:

；i. Sampling a topic

;

表示第i个英文单词的主题。in,

Represents the subject of the i -th English word.

；ii. Sample each English word

;

。iii. Scholars who sample English words

.

，

，

，

和

。本发明实施例中可以采用吉布斯采样方法（如collapsed Gibbs sampling）算法来推断RIMM中待估计的参数。吉布斯采样（Gibbs sampling）是统计学中用于马尔科夫蒙特卡洛（MCMC）的一种算法，用于在难以直接采样时从某一多变量概率分布中近似抽取样本序列。Research Interests There are two latent variables z ^b and z ^e in the mining model RIMM, and five parameters to be estimated

,

,

,

and

. In this embodiment of the present invention, a Gibbs sampling method (eg, collapsed Gibbs sampling) algorithm may be used to infer parameters to be estimated in the RIMM. Gibbs sampling is an algorithm used in statistics for Markov Monte Carlo (MCMC) to approximate a sequence of samples from a multivariate probability distribution when direct sampling is difficult.

In the embodiment of the present invention, the sampling formula of each Chinese two-word topic is as follows:

(1)

(1)

To derive formula (1), first calculate its joint distribution:

（2）

(2)

Wherein, the joint distribution represents an approximate distribution inferred based on the results obtained by the above sampling.

According to formula (1) and formula (2), the sampling formula of double words is obtained as follows:

（3）

(3)

表示中文主题Z ^b 下的中文词i，

表示除了第i个单词外的所有中文主题的集合；

表示的英文主题Z ^b 下除英文单词i，

表示除了第i个单词外的所有英文主题的集合；

表示第p个专业领域下，除了词i之外，科研项目数据里属于主题z的所有中英文词总数；

表示第p个专业领域下，除了词i之外，论文数据里属于主题z的中英文词总数；

分别表示主题z的中文主题Z ^b 下除中文词i外，所有科研项目和论文数据里中文单词

出现的次数；

表示主题z下除了词i之外，所有科研项目和论文数据里属于学者s _i 的中英文词出现的次数；

表示主题z的英文主题Z ^e 下除英文单词i外，所有科研项目和论文数据里英文单词

出现的次数；

,

分别为科研项目数据的主题分布以及论文数据的主题分布；

为符合狄利克雷分布的主题-英文单词分布，

为符合狄利克雷分布的主题-学者分布，

为符合狄利克雷分布的主题-中文单词分布；

为平衡因子，表示利用论文信息的比例，用来调节学者的兴趣受科研项目和论文的影响的大小；R表示不同学者的个数，B表示双词的集合，S表示学者的集合，W^b表示不同中文单词的个数，W^e表示不同英文单词的个数。where Z represents the total number of topics,

represents the Chinese word i under the Chinese topic Z ^b ,

Represents the set of all Chinese topics except the ith word;

In addition to the English word i under the English topic Z ^b represented,

Represents the set of all English topics except the ith word;

Indicates the total number of all Chinese and English words belonging to topic z in the scientific research project data except word i in the pth professional field;

Indicates the total number of Chinese and English words belonging to topic z in the paper data except word i in the pth professional field;

In addition to the Chinese word i , the Chinese words in all scientific research projects and paper data under the Chinese topic Z ^b respectively represent the topic z .

the number of occurrences;

Indicates the number of _{occurrences of Chinese and English words belonging to scholar si} in all scientific research projects and paper data except word i under topic z ;

Except for the English word i , the English words in all scientific research projects and thesis data under the English topic Z ^e representing the topic z

the number of occurrences;

,

are the subject distribution of scientific research project data and the subject distribution of paper data;

For the topic-English word distribution that conforms to the Dirichlet distribution,

To fit the subject-scholar distribution of the Dirichlet distribution,

is a topic-Chinese word distribution that conforms to the Dirichlet distribution;

is a balance factor, which indicates the proportion of the information of the papers used to adjust the influence of scholars' interests by scientific research projects and papers; R represents the number of different scholars, B represents the set of double words, S represents the set of scholars, and W ^b Represents the number of different Chinese words, ^We represent the number of different English words.

来协同地从主题分布中抽取主题，以能够同时利用不同来源的数据获取学者的兴趣语义表示，能克服现有技术中小样本源的影响会被大样本源淹没的问题。In the embodiment of the present invention, a balance factor is designed

To synergistically extract topics from the topic distribution, to be able to use data from different sources at the same time to obtain the semantic representation of scholars' interests, and to overcome the problem that the influence of small sample sources in the prior art will be overwhelmed by large sample sources.

Similarly, in the embodiment of the present invention, it is the English word sampling subject, and the sampling formula is as follows:

(4)

(4)

According to the joint formula (2), the sampling formula of English words can be obtained as follows:

（5）

(5)

表示主题z下除英文单词i外，所有基金项目和论文数据里英文单词

出现的次数。in,

Indicates the English words in all fund projects and thesis data except the English word i under topic z

the number of occurrences.

Iteratively execute the above sampling rules until the results converge, and finally the following estimated parameters can be obtained:

； （6-1）

; (6-1)

； （6-2）

; (6-2)

； （6-3）

; (6-3)

； （6-4）

; (6-4)

（6-5）

(6-5)

表示科研项目的专业领域为p主题为z的专业领域-主题分布；

表示论文数据的专业领域为p主题为z的专业领域-主题分布；

表示第p个专业领域下，科研项目数据里属于主题z的所有中英文词总数；

表示第p个专业领域下，论文数据里属于主题z的所有中英文词总数；

表示第z^b个主题下，所有科研项目和论文数据里中文单词出现的次数；

表示主题为z的中文双词的主题-中文单词分布；

表示第z^e个主题下，所有科研项目和论文数据里英文单词出现的次数；

表示主题为z的英文单词w ^e 的主题-英文单词分布；

表示主题为z、学者为s的主题-学者分布；

表示主题z下所有科研项目和论文数据里属于学者s的中英文词出现的次数。in,

Indicates that the professional field of the scientific research project is the professional field of p and the topic is z - topic distribution;

Indicates that the professional field of the paper data is p and the topic is the professional field of z - topic distribution;

Indicates the total number of all Chinese and English words belonging to topic z in the scientific research project data under the pth professional field;

Indicates the total number of all Chinese and English words belonging to topic z in the paper data under the pth professional field;

Indicates the number of occurrences of Chinese words in the data of all scientific research projects and papers under the z- ^th topic;

Represents the topic-Chinese word distribution of Chinese double words with topic z;

Represents the number of occurrences of English words in all scientific research projects and paper data under the z ^e -th topic;

Represents the topic-English word distribution of the English word we with topic ^z ;

Represents the topic-scholar distribution with topic z and scholar s;

Indicates the number of occurrences of Chinese and English words belonging to scholar s in all scientific research projects and paper data under topic z.

It can be seen from the above that the semantic representation of scholars' research interests can be obtained through the research interest mining model, but the representation is static, that is, only a topic representation for a period of time. Therefore, in order to cluster scholars according to the semantic representation of users' interests at different times, it is necessary to reflect the dynamic changes of scholars' interests. Although the interests of scholars are dynamic, there is also a certain persistence. Therefore, the present invention does not simply slice the corpus, but uses the historical interests of scholars to help infer the topic distribution at the current moment.

位学者，对于每一位学者，都收集了他的写书消息。基于这些学术消息，如果利用RIMM来获取他们的兴趣语义表示，并将具有共同兴趣的学者聚集在一起作为一个聚类c。则该聚类的正式陈述如下：Suppose that in time period t there are

A scholar, for each scholar, has collected his writing news. Based on these academic messages, if we utilize RIMM to obtain semantic representations of their interests, and gather scholars with common interests together as a cluster c . Then the formal statement of the cluster is as follows:

；

;

和

分别表示学者和簇在时段t 内的数量。Among them, f represents the mapping relationship between independent scholars to scholar clusters. c represents a group of scholars with common interests.

and

are the numbers of scholars and clusters in time period t , respectively.

In order to model the dynamic changes of scholars' interests, the present invention fully considers how to use the historical topic distribution of multi-source and multilingual data to infer the topic distribution of current scholars.

In order to maintain the continuity of scholars' interests in different periods, the present invention takes the topic distribution of the previous period (eg, the (t-1)th period) as the prior distribution of the current period (eg, the tth period). Therefore, the prior parameters at the current moment are calculated as follows:

； （7-1）

; (7-1)

； （7-2）

; (7-2)

； （7-3）

; (7-3)

； （7-4）

; (7-4)

； （7-5）

; (7-5)

为衰减因子，

值越大，表示历史兴趣对当前主题分布的影响越大；

表示在t-1时刻，第p个专业领域下，科研项目数据里属于主题

的所有中英文词总数；

表示在t-1时刻，科研项目数据的专业领域主题分布；

表示在

时刻，科研项目数据的专业领域主题分布；

表示在

时刻，第

个专业领域下，论文数据里属于主题

的所有中英文词总数；

表示在

时刻，论文数据的专业领域主题分布；

表示在

时刻，论文数据的专业领域主题分布；

表示在

时刻，第

个主题下，所有科研项目和论文数据里中文单词

出现的次数；

表示在

时刻，所有科研项目和论文数据的主题中文单词分布；

表示在

时刻，所有科研项目和论文数据的主题中文单词分布；

表示在

时刻，第

个主题下，所有科研项目和论文数据里英文单词

出现的次数；

表示在

时刻，所有科研项目和论文数据的主题英文单词分布；

表示在

时刻，所有科研项目和论文数据的主题英文单词分布；

表示在

时刻，第

个主题下，所有科研项目和论文数据里学者

出现的次数；

表示在

时刻，所有科研项目和论文数据的主题学者分布；

表示在

时刻，所有科研项目和论文数据的主题学者分布。in,

is the attenuation factor,

The larger the value, the greater the impact of historical interest on the current topic distribution;

Indicates that at time t-1 , under the pth professional field, the scientific research project data belongs to the theme

The total number of all Chinese and English words of ;

Represents the subject distribution of scientific research project data in professional fields at time t-1 ;

expressed in

Time, subject distribution of scientific research project data in professional fields;

expressed in

time, the

Under a professional field, the thesis data belongs to the topic

The total number of all Chinese and English words of ;

expressed in

Moment, the subject distribution of the professional field of the paper data;

expressed in

Moment, the subject distribution of the professional field of the paper data;

expressed in

time, the

Under this topic, Chinese words in all scientific research projects and paper data

the number of occurrences;

expressed in

Time, topic Chinese word distribution of all scientific research projects and dissertation data;

expressed in

Time, topic Chinese word distribution of all scientific research projects and dissertation data;

expressed in

time, the

Under this topic, English words in all scientific research projects and paper data

the number of occurrences;

expressed in

Time, the subject English word distribution of all scientific research projects and paper data;

expressed in

Time, the subject English word distribution of all scientific research projects and paper data;

expressed in

time, the

Under this topic, scholars in all scientific research projects and dissertation data

the number of occurrences;

expressed in

Time, the distribution of subject scholars for all scientific research projects and dissertation data;

expressed in

Moment, subject scholar distribution of all research projects and dissertation data.

，其中，

。通过在每个子集上运行研究兴趣挖掘模型就可以得到不同时间片内的主题-学者分布，也就是可以获得学者在不同时段t的兴趣语义表示。本发明实施例中利用单遍（single-pass）聚类算法对所获取的不同时刻的学者兴趣语义表示进行聚类，来获得动态的学者聚类结果。If T time slices (periods) are given, the entire corpus (a collection of academic metadata) M can be divided into T corpus subsets according to the period t :

,in,

. By running the research interest mining model on each subset, the topic-scholar distribution in different time slices can be obtained, that is, the semantic representation of scholars' interests in different time periods t can be obtained. In the embodiment of the present invention, a single-pass clustering algorithm is used to cluster the acquired scholar's interest semantic representations at different times, so as to obtain a dynamic scholar clustering result.

As can be seen from the above, the present invention studies the scholar clustering problem in multi-source, multi-language, and growing academic data, and can infer the topic distribution of scholars through the newly proposed RIMM, and realize dynamic clustering according to their research interests. The research interest mining model (RIMM) proposed by the present invention can effectively integrate multi-source data, make up the vocabulary gap of multilingual data, and accurately mine the research interests of scholars. In addition, the present invention also proposes to use the Gibbs sampling method to infer the parameters of the RIMM. Through iterative sampling of topics, the present invention can map all features of academic data into a common semantic space and obtain their topic distribution. The RIMM-based research interest tracking and dynamic scholar clustering method of the present invention can dynamically track and cluster scholars' interests by capturing the continuity and variability of scholars' research interests.

Corresponding to the above method, the present invention also provides a research interest mining device for scholar clustering, comprising a processor and a memory, wherein the memory stores computer instructions, and the processor is used to execute the data stored in the memory. Computer instructions, when the computer instructions are executed by the processor, the apparatus implements the steps of the aforementioned edge computing server deployment method.

Embodiments of the present invention further provide a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, implements the steps of the foregoing method for deploying an edge computing server. The computer-readable storage medium may be a tangible storage medium, such as an optical disk, a USB flash drive, a floppy disk, a hard disk, and the like.

It is to be understood that the present invention is not limited to the specific arrangements and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above-described embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the sequence of steps after comprehending the spirit of the present invention.

It should be understood by those of ordinary skill in the art that the various exemplary components, systems and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software or a combination of the two. Whether it is implemented in hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may implement the described functionality using different methods for each particular application, but such implementations should not be considered beyond the scope of the present invention. When implemented in hardware, it may be, for example, an electronic circuit, an application specific integrated circuit (ASIC), suitable firmware, a plug-in, a function card, or the like. When implemented in software, elements of the invention are programs or code segments used to perform the required tasks. The program or code segments may be stored in a machine-readable medium or transmitted over a transmission medium or communication link by a data signal carried in a carrier wave.

It should also be noted that the exemplary embodiments mentioned in the present invention describe some methods or systems based on a series of steps or devices. However, the present invention is not limited to the order of the above steps, that is, the steps may be performed in the order mentioned in the embodiments, or may be different from the order in the embodiments, or several steps may be performed simultaneously.

In the present invention, features described and/or illustrated with respect to one embodiment may be used in the same or similar manner in one or more other embodiments, and/or in combination with or in place of features of other embodiments Features of the implementation.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. For those skilled in the art, various modifications and changes may be made to the embodiments of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims (8)

1. A research interest mining method for scholar clustering, characterized in that the method comprises the following steps: An academic metadata set is constructed based on academic data related to scholars from multiple sources. Each piece of academic metadata in the academic metadata set includes the following information: professional field information, data source information, scholar information and text content information. The academic data includes Research project data and dissertation data; Input at least part of the academic metadata in the constructed academic metadata set as input data into the pre-established research interest mining model, and obtain the scholar interest semantic representation by repeatedly executing the research interest mining model for sampling multiple iterations. The semantic representation of interests includes professional field-topic distribution, topic-English word distribution, topic-Chinese word distribution, and topic-scholar distribution; wherein, the research interest mining model is based on the data of scholars from the same data source and belonging to the same professional field share the same topic distribution, and, in the research interest mining model, the professional field-topic distribution is modeled as a Dirichlet distribution, the topic-English word distribution, topic-Chinese word distribution and topic-scholar distribution is modeled as a polynomial distribution, and the area of expertise-topic distribution includes the area of expertise-topic distribution of scientific research projects and the area of expertise-topic distribution of dissertation data; Perform scholar clustering based on the obtained semantic representation of scholar interest, and obtain scholar clustering results; Wherein, at least part of the academic metadata in the constructed academic metadata set is input into the pre-established research interest mining model as input data, and the scholar's interest semantic representation is obtained by repeatedly executing the research interest mining model for sampling multiple times. ,include: Sampling the academic metadata from scientific research projects and belonging to a predetermined professional field to obtain the professional field-topic distribution of scientific research projects; Sampling academic metadata from papers that belong to a predetermined professional domain to obtain the professional domain-topic distribution of the paper data; Sampling for each predetermined topic to obtain the corresponding topic-Chinese word distribution, topic-scholar distribution and topic-English word distribution; Determine whether it is Chinese or English according to the text content in each piece of academic metadata; in the case that the text content is Chinese, the two words appearing at the same time in each piece of Chinese academic metadata are combined in pairs to extract the double words. And take the following actions for each double word: sample the topic of the Chinese double word, independently sample the two words, and sample the author of the double word; in the case of the text content in English, sample the following operations for each word: sample the English topic , sample each English word, and the author of the sampled English word. 2. The method according to claim 1, wherein the method further comprises the step of constructing a research interest mining model, the step comprising: Identify the area of expertise to which the academic metadata falls; To obtain academic metadata topics by collaboratively sampling topics from a distribution of topics from research projects and belonging to a defined area of expertise and a distribution of topics from papers and belonging to a defined area of expertise; Generate English words, Chinese words, and scholars according to topic-English word distribution, topic-Chinese word distribution, and topic-scholar distribution. 3. The method according to claim 1, wherein the constructing an academic metadata set comprises: Determine the source of each piece of academic metadata, and determine whether it is scientific research project data or paper data; Extract the scholar to which the current academic metadata belongs; Extract the text information of each scientific research project or paper; Determine whether the text information of the scientific research project or paper is in Chinese or English; The time period to which the current academic metadata belongs is determined, and the current academic metadata is classified into the academic metadata subset corresponding to the time period to which the current academic metadata belongs, thereby constructing an academic metadata set including the academic metadata subset. 4. The method of claim 1, wherein Said inputting at least part of the academic metadata in the constructed academic metadata set into the pre-established research interest mining model as input data, and obtaining the scholar's interest semantic representation by iteratively executing the research interest mining model for multiple times for sampling, and further Including: inputting the academic metadata in the academic metadata subsets corresponding to different time periods as input data into the research interest mining model to obtain the semantic representation of scholars' interests at different times; The clustering of the acquired scholar's interest semantic representation includes: using a single-pass clustering algorithm to cluster the acquired scholar's interest semantic representation at different times. 5. The method according to claim 1, wherein the parameters to be estimated in the research interest mining model are calculated by using a Gibbs sampling method; The sampling of Chinese double words adopts the following formula:

The following formula is used for the sampling of each English word:

where Z represents the total number of topics,

represents the Chinese word i under the Chinese topic Z ^b ,

Represents the set of all Chinese topics except the ith word;

represents the English word i under the English topic Z ^b ,

Represents the set of all English topics except the ith word;

Indicates the total number of all Chinese and English words belonging to topic z in the scientific research project data except word i in the pth professional field;

Indicates the total number of Chinese and English words belonging to topic z in the paper data except word i in the pth professional field;

In addition to the Chinese word i , the Chinese words in all scientific research projects and paper data under the Chinese topic Z ^b respectively represent the topic z .

the number of occurrences;

Indicates the number of _{occurrences of Chinese and English words belonging to scholar si} in all scientific research projects and paper data except word i under topic z ;

Except for the English word i , the English words in all scientific research projects and thesis data under the English topic Z ^e representing the topic z

The number of occurrences; a ₁ and a ₂ are the subject distribution of scientific research project data and the subject distribution of paper data respectively;

For the topic-English word distribution that conforms to the Dirichlet distribution,

To fit the subject-scholar distribution of the Dirichlet distribution,

It is the topic-Chinese word distribution that conforms to the Dirichlet distribution; R represents the number of different scholars, B represents the set of double words, S represents the set of scholars, W ^b represents the number of different Chinese words, and We ^e represents different English words number of ,

is the balance factor; After the iterative operation of the research interest mining model, the following estimated parameters are obtained:

;

;

;

;

; in,

Indicates that the professional field of the scientific research project is the professional field of p and the topic is z - topic distribution;

Indicates that the professional field of the paper data is p and the topic is the professional field of z - topic distribution;

Indicates the total number of all Chinese and English words belonging to topic z in the scientific research project data under the pth professional field;

Indicates the total number of all Chinese and English words belonging to topic z in the paper data under the pth professional field;

Indicates the number of occurrences of Chinese words in the data of all scientific research projects and papers under the z- ^th topic;

Represents the topic-Chinese word distribution of Chinese double words with topic z;

Represents the number of occurrences of English words in all scientific research projects and paper data under the z ^e -th topic;

Represents the topic-English word distribution of the English word we with topic ^z ;

Represents the topic-scholar distribution with topic z and scholar s;

Indicates the number of occurrences of Chinese and English words belonging to scholar s in all scientific research projects and paper data under topic z. 6. The method according to claim 3, wherein the method further comprises: Taking the topic distribution of the previous period as the prior distribution of the current period, the prior parameters of the current period are calculated based on the following formula:

in,

is the attenuation factor,

Represents the total number of all Chinese and English words belonging to topic z in the scientific research project data under the pth professional field at time t-1 ;

Represents the subject distribution of scientific research project data in professional fields at time t-1 ;

expressed in

Time, subject distribution of scientific research project data in professional fields;

expressed in

time, the

Under a professional field, the thesis data belongs to the topic

The total number of all Chinese and English words of ;

expressed in

Moment, the subject distribution of the professional field of the paper data;

expressed in

Moment, the subject distribution of the professional field of the paper data;

expressed in

time, the

Under this topic, Chinese words in all scientific research projects and paper data

the number of occurrences;

expressed in

Time, topic Chinese word distribution of all scientific research projects and dissertation data;

expressed in

Time, topic Chinese word distribution of all scientific research projects and dissertation data;

expressed in

time, the

Under this topic, English words in all scientific research projects and paper data

the number of occurrences;

expressed in

Time, the subject English word distribution of all scientific research projects and paper data;

expressed in

Time, the subject English word distribution of all scientific research projects and paper data;

expressed in

time, the

Under this topic, scholars in all scientific research projects and dissertation data

the number of occurrences;

expressed in

Time, the distribution of subject scholars for all scientific research projects and dissertation data;

expressed in

Moment, subject scholar distribution of all research projects and dissertation data. 7. A research interest mining device for scholar clustering, comprising a processor and a memory, characterized in that the memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory, when the memory is stored. The apparatus implements the steps of the method as claimed in any one of claims 1 to 6 when the computer instructions are executed by the processor. 8. A computer-readable storage medium on which a computer program is stored, characterized in that, when the program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images