CN107622047A - A Method of Extracting and Representing Design Decision Knowledge - Google Patents

Abstract

The invention relates to a method for extracting and expressing design decision-making knowledge. Aiming at the requirement of acquiring design decision-making knowledge and completing model construction, through the template operation of design documents, the design process is realized from three perspectives: design intent, design scheme, and design basis. Discovery and extraction of design decision-making knowledge in design documents, the present invention uses the design decision-making knowledge hierarchical model to reconstruct the extracted design decision-making knowledge, can realize the extraction and expression of design decision-making knowledge in historical design documents, and form a design decision-making knowledge base , thereby improving design efficiency and design quality.

Description

A Method of Extracting and Representing Design Decision Knowledge

technical field

The invention belongs to the research content of knowledge resource management and application method in the field of computer application technology, more specifically, the invention relates to a method for extracting and expressing design decision knowledge.

Background technique

Design decision-making knowledge reflects the dynamic psychological behavior process of designers using various professional knowledge and experience to define design problems, generate design concepts, and make design decisions. Generally speaking, design decision-making knowledge is an explanation of why a product is designed in this way. Effective acquisition, organization and application of design decision-making knowledge play a decisive role in improving design quality and design efficiency. Design decision-making knowledge is usually divided into explicit decision-making knowledge and implicit decision-making knowledge according to the existing form. Explicit decision-making knowledge is the historical knowledge that designers save the decision-making knowledge generated in the design process in different formats of documents, such as design instructions, operation instructions, technical reports, patent documents, etc. Invisible decision-making knowledge is the decision-making knowledge that designers keep in their minds but not written in the design process. In order to better manage and apply design decision knowledge, many studies are devoted to building design decision knowledge base.

At present, there are two sources of design decision knowledge acquisition, one is to capture design decision knowledge (implicit decision knowledge) in the design process, and the other is to capture design decision knowledge from already formed design documents (explicit decision knowledge). In the past 20 years, the research on the capture method of design decision-making knowledge mainly focuses on capturing the design decision-making knowledge generated during the design process. This method can record the explicit knowledge and implicit knowledge generated by engineering designers in the design process in detail, but it needs It consumes a lot of time and energy of designers. Design documents are not regarded as an important source of design rationality because of their complex structure, cumbersome language, and natural language is not easy to be recognized by computers. However, design documents contain a large amount of design decision-making knowledge, which is very important for design reuse, design reasoning, and design evaluation. Therefore, it is necessary to carry out necessary knowledge processing on design documents, effectively extract design decision knowledge from them, construct a design decision knowledge model, and realize efficient reuse of design decision knowledge.

Contents of the invention

The purpose of the present invention is to overcome the deficiencies of the prior art and provide a method for extracting and expressing design decision knowledge.

A method for extracting and expressing design decision knowledge of the present invention includes the following steps:

The steps of design intention layer generation: Use Apache POI to change the design document to standardize its title; then use the Apache POI method to obtain all the titles in the document and the containment relationship between the titles, and store all the titles according to their level into the design intention candidate set P; and then construct a word set D, the vocabulary in D is the vocabulary that cannot be used as the design intention after a large number of experiments, and retrieve similar words in the dictionary and store them in D; the design intention candidate set P Match all the titles with all the vocabulary in the vocabulary D, if the title contains the vocabulary in the vocabulary D, delete it, get the design intent set after the matching is complete, and then generate the design intent layer according to the design intent set;

The steps of design plan layer generation: Divide the design document into sentences, perform word segmentation and part-of-speech tagging for each sentence; use the syntax analysis method based on phrase structure to obtain all the verb-object phrases under the paragraph corresponding to each design intention and the center of the design intention Words, store the verb-object phrases in the design proposal candidate set V corresponding to the design intention; use the correlation calculation formula to calculate the correlation between the vocabulary contained in the content between adjacent titles and the central word of the design intention, and set the threshold according to actual needs σ, extract the word set C composed of words whose correlation degree is greater than the threshold σ, match the verb-object phrase set with the word set C, and the verb-object phrase containing any word in the word set C corresponds to the design intention Design scheme; delete the verb-object phrases that are not design schemes in the design scheme candidate set V; obtain the design scheme set corresponding to each design intention in turn, and connect the verb-object phrases in the design scheme set with their corresponding design intentions to form design intentions - design scheme hierarchy;

The formula for calculating the correlation degree is:

s(i,j) in the formula represents the correlation between vocabulary i and j; tf(i,j) represents the number of times vocabulary i and j appear in a sentence at the same time; w(i,j)=R/A , represents the relative distance between i and j; R represents the number of characters in the content between i and j, and A represents the number of all characters contained in the current content to be extracted;

Design basis layer generation steps: establish a design basis vocabulary library O, and input all words expressing explanations and reasons in the Chinese dictionary; extract sentences containing any vocabulary in O between adjacent titles as the design basis candidate set L; design The design scheme between the corresponding adjacent titles obtained by the scheme extraction algorithm and the sentences in the design basis candidate set L are used as nodes to construct a semantic sentence graph G(B,X);

Iteratively calculate the weight value between the design scheme and the design basis, the calculation formula of the rth iteration is f(r+1)=θMf(r)+(1-θ)y, until the calculation result converges; according to the weight value, we can get For the design basis most relevant to the design scheme, take the design basis with the top k% weight value for each design scheme, establish connections, and finally form a three-layer structure of design intent-design scheme-design basis;

Node B is defined as a vertex in the G(B, X), and the connection weight is represented by a matrix X representing node semantic correlation, which represents the correlation between two sentences b _i and b _j , where x _ii =0; Said M is to perform symmetrical normalization on the similarity matrix X as in the formula The resulting, the Q is a diagonal matrix, where The sub-vector f of the sentence = [f ₀ , f ₁ ...... f _n ] ^T , the initial value is set to The y is an n-dimensional vector y=[1,1,...1] ^T ; the r represents the rth calculation, and the θ is a coefficient, and the value range is 0≤θ≤1 ; The n and K are determined by the designer according to actual needs.

The invention has the beneficial effects of realizing the extraction and expression of design decision knowledge in historical design documents, forming a design knowledge base, thereby improving design efficiency and design quality, and avoiding a lot of energy and time for designers.

Description of drawings

Fig. 1 is a hierarchical model diagram of design decision-making knowledge in the present invention;

Figure 2 is a flow chart of design intent layer generation;

Fig. 3 is a flow chart of design scheme layer generation;

Fig. 4 is a flow chart of design basis layer generation;

detailed description

This embodiment is a method for extracting and expressing design decision-making knowledge, and establishes a model composed of three levels: design intent, design scheme, and design basis, as shown in FIG. 1 . The specific method is divided into three steps:

1. Steps for generating the design intent layer:

1) Input the design document W;

2) Check whether there is a standard title format in the design document W, if there is no standard title format, go to step 3), if there is, go to step 4)

3) Change the design document W using Apache POI to standardize its title;

4) Use the Apache POI method to obtain all titles in the document and the inclusion relationship between them;

5) Store all the acquired titles into the design intent candidate set P according to their levels;

6) Construct a word set D, the words in D are words that cannot be used as the design intention after a large number of experiments, and retrieve similar words in the dictionary and store them in D;

7) Match the title in the design intent candidate set P with the vocabulary in the word set D, if the title contains the vocabulary in the word set D, delete it to obtain the design intent set Q;

8) Generate the design intent layer according to Q.

2. Steps for generating the design scheme layer:

1) Input the design document W;

2) Divide the design document W into n sentences;

3) Carry out word segmentation and part-of-speech tagging for each sentence in the design document;

4) Use the Apache POI method to determine the scope of the extracted design scheme. It is mentioned in the design intention layer extraction algorithm that each design intention corresponds to a title, and the content between adjacent titles is the scheme and basis for realizing the design intention;

5) Using the syntax analysis method based on the phrase structure, obtain all the verb-object phrases under the paragraph corresponding to a certain design intention and the central word N of the design intention, store the verb-object phrases in V, and record it as the design scheme candidate set V;

6) Considering that not all the verb-object phrases in the design proposal candidate set V obtained in 5) can be used as design proposals, so the verb-object phrases that are not design proposals should be deleted from V. This paper uses the correlation calculation method to calculate the correlation between the vocabulary contained in the content between adjacent titles and the central word N of the design intention, set the threshold σ, extract the word set C with a correlation greater than the threshold σ, and combine the verb-object phrase set V with the The word set C is matched, and the verb-object phrase containing any word in the word set C is the design scheme corresponding to the design intention. Relevance calculation formula:

s(i,j) in the formula represents the correlation between vocabulary i and j; tf(i,j) represents the number of times vocabulary i and j appear in a sentence at the same time; w(i,j)=R/A , represents the relative distance between i and j; R represents the number of characters in the content between i and j, and A represents the number of all characters contained in the current content to be extracted;

7) After the calculation and matching in 6), the verb-object phrases that are not design schemes in the design scheme candidate set are deleted;

8) Update the design scheme candidate set V, which is now the design scheme set;

9) Connect the verb-object phrases in the design scheme set with their corresponding design intentions to form a design intention-design scheme hierarchy.

Three, the steps of design basis layer generation:

1) Input the design document W;

2) The design document W is divided into n sentences;

3) Establish a design basis vocabulary library O, and input all expressions and reasons in the Chinese dictionary;

4) According to the vocabulary stored in the design basis vocabulary library O, extract sentences containing any vocabulary among adjacent titles as the design basis candidate set L;

5) Use the design schemes between the corresponding adjacent titles obtained by the design scheme extraction algorithm and the sentences in the design basis candidate set L obtained in 4) as nodes to construct a semantic sentence graph G(B,X). Node B in G(B,X) is defined as a vertex, and the connection weight is represented by a matrix X representing the semantic correlation of nodes, which represents the correlation between two sentences b _i and b _j , where x _ii = 0; for similarity The matrix X is symmetric and normalized as follows: Q is a diagonal matrix where The sub-vector f of the sentence = [f ₀ , f ₁ ...... f _n ] ^T , the initial value is set to Then define an n-dimensional vector y=[1,1,...1] ^T , n is determined by the designer according to actual needs;

6) Calculate the weight value between the design scheme and the design basis; cyclically calculate f(r+1)=θMf(r)+(1-θ)y until the calculation result converges. In the formula, r represents the rth calculation, θ is a coefficient, and the value range is 0≤θ≤1;

7) According to the weight value, the design basis most relevant to a certain design scheme can be obtained, and the design basis with the top k% weight value is selected for each design scheme, where K is determined by the designer according to actual needs, and the relationship is established. Finally, a three-layer structure of design intent-design scheme-design basis is formed.

The above is only a specific embodiment of the present invention, and the scope of protection of the present invention is not limited thereto. For those of ordinary skill in the art, improvements or transformations can be made without creativity according to the above description, and all these improvements All changes and transformations should belong to the scope of protection of the appended claims of the present invention.

Claims (1)

1. A method for extracting and expressing design decision-making knowledge, comprising the following steps: The steps of design intention layer generation: Use Apache POI to change the design document to standardize its title; then use the Apache POI method to obtain all the titles in the document and the containment relationship between the titles, and store all the titles according to their level into the design intention candidate set P; and then construct a word set D, the vocabulary in D is the vocabulary that cannot be used as the design intention after a large number of experiments, and retrieve similar words in the dictionary and store them in D; the design intention candidate set P Match all the titles with all the vocabulary in the vocabulary D, if the title contains the vocabulary in the vocabulary D, delete it, get the design intent set after the matching is complete, and then generate the design intent layer according to the design intent set; The steps of design plan layer generation: Divide the design document into sentences, perform word segmentation and part-of-speech tagging for each sentence; use the syntax analysis method based on phrase structure to obtain all the verb-object phrases under the paragraph corresponding to each design intention and the center of the design intention Words, store the verb-object phrases in the design proposal candidate set V corresponding to the design intention; use the correlation calculation formula to calculate the correlation between the vocabulary contained in the content between adjacent titles and the central word of the design intention, and set the threshold according to actual needs σ, extract the word set C composed of words whose correlation degree is greater than the threshold σ, match the verb-object phrase set with the word set C, and the verb-object phrase containing any word in the word set C corresponds to the design intention Design scheme; delete the verb-object phrases that are not design schemes in the design scheme candidate set V; obtain the design scheme set corresponding to each design intention in turn, and connect the verb-object phrases in the design scheme set with their corresponding design intentions to form design intentions - design scheme hierarchy; The formula for calculating the correlation degree is: <mrow><mi>s</mi><mrow><mo>(</mo><mi>i</mi><mo>,</mo><mi>j</mi><mo>)</mo></mrow><mo>=</mo><mfrac><mrow><mi>t</mi><mi>f</mi><mrow><mo>(</mo><mi>i</mi><mo>,</mo><mi>j</mi><mo>)</mo></mrow></mrow><mrow><mi>w</mi><mrow><mo>(</mo><mi>i</mi><mo>,</mo><mi>j</mi><mo>)</mo></mrow></mrow></mfrac></mrow> s(i,j) in the formula represents the correlation between vocabulary i and j; tf(i,j) represents the number of times vocabulary i and j appear in a sentence at the same time; w(i,j)=R/A , represents the relative distance between i and j; R represents the number of characters in the content between i and j, and A represents the number of all characters contained in the current content to be extracted; Design basis layer generation steps: establish a design basis vocabulary library O, and input all words expressing explanations and reasons in the Chinese dictionary; extract sentences containing any vocabulary in O between adjacent titles as the design basis candidate set L; design The design scheme between the corresponding adjacent titles obtained by the scheme extraction algorithm and the sentences in the design basis candidate set L are used as nodes to construct a semantic sentence graph G(B,X); Iteratively calculate the weight value between the design scheme and the design basis, the calculation formula of the rth iteration is f(r+1)=θMf(r)+(1-θ)y, until the calculation result converges; according to the weight value, we can get For the design basis most relevant to the design scheme, take the design basis with the top k% weight value for each design scheme, establish connections, and finally form a three-layer structure of design intent-design scheme-design basis; Node B is defined as a vertex in the G(B, X), and the connection weight is represented by a matrix X representing node semantic correlation, which represents the correlation between two sentences b _i and b _j , where x _ii =0; Said M is to perform symmetrical normalization on the similarity matrix X as in the formula The resulting, the Q is a diagonal matrix, where The sub-vector f of the sentence = [f ₀ , f ₁ ...... f _n ] ^T , the initial value is set to The y is an n-dimensional vector y=[1,1,...1] ^T ; the r represents the rth calculation, and the θ is a coefficient, and the value range is 0≤θ≤1 ; The n and K are determined by the designer according to actual needs.