A Multimedia Information Retrieval Method Based on Semantic Similarity

Abstract

The present invention belongs to the technical field of information retrieval, and particularly relates to a multimedia information retrieval method based on semantic similarity. During the preprocessing operation of documents, the present invention learns the method of the classical vector space model and replaces the dictionary containing keyword entries with an ontology library, uses the semantic vector containing the concepts described in the document and their features to replace the document and meaningfully extracts and absorbs the content of the document. To achieve the high efficiency of the retrieval completion, semantic classification is performed on the document to prepare for a more effective use of the abduction plan and techniques of the retrieved semantic vectors. When calculating the concept similarity and attribute similarity, the existence relationship between each concept instance and the attribute power is determined, and the influencing factors are analyzed according to the concept characteristics to complete the entire process of semantic retrieval. Finally, multimedia information retrieval is completed based on semantic similarity.

Description

Technical Field

[0001] The present invention belongs to the technical field of information retrieval, and particularly relates to a multimedia information retrieval method based on semantic similarity. Background Art

[0002] With the rapid development of computer technology and the Internet, information of various media including images is developing at a speed beyond human imagination. At the same time, the problem people are facing is how to effectively obtain necessary information in the large multimedia world, rather than being troubled by the lack of multimedia content in daily life. One of the reasons is that traditional database searches use keyword-based search methods. Therefore, it is usually very difficult to comprehensively describe multimedia content with just a few keywords, and it is also very subjective to select image attributes as keywords. Secondly, it is very difficult for users to translate these informative annotations in the form of a certain code. In the field of digital libraries, professional librarians organize important vocabulary in the professional field according to the semantic level and use it to search and retrieve context media data. In order to improve the accuracy of the search, other methods of semantic reporting and analyzing the theme and content of documents or pages are also used. However, since the current information retrieval mainly conducts searches by using an interactive system during the retrieval process without considering other information such as the description of concept features, and this concept semantic search method often cannot meet the actual needs. There are problems such as complex operations, errors in information queries, and low accuracy in multimedia information retrieval using conventional methods. Summary of the Invention

[0003] The purpose of the present invention is to propose a multimedia information retrieval method based on semantic similarity in order to solve the above technical problems.

[0004] The present invention adopts the following technical solutions to achieve the above invention purpose:

[0005] A multimedia information retrieval method based on semantic similarity includes the following steps:

[0006] Step 1, query and extraction of semantic vectors: During the preprocessing operation of the document, learn the method of the classical vector space model and replace the dictionary containing keyword entries with an ontology library, and use the semantic vector containing the concepts described in the document and their features to replace the document and meaningfully extract and absorb the content of the document;

[0007] Step 2, Document Semantic Classification and Indexing: Based on the result of the recorded semantic vector of the record, obtain information from the semantic vectors of the semantics based on the private ontology tree for classification; establish the semantic index status of the classified documents. First, insert the ontology concepts into the index file and arrange them in lexicographical order, then create an ordered linked list. This list first references the document semantic vector features under this concept, and then a pointer with a pointing link will be created in the index file, and the design of the list will be carried out on the pointer, and then the semantic attribute vector of the document will be linked to the corresponding document;

[0008] Step 3, Calculation of Concept Similarity: Determine the existence relationship between each concept instance and the property force, and analyze its influencing factors according to the concept characteristics to complete the entire process of semantic retrieval;

[0009] Step 4, Completing Multimedia Information Retrieval Based on Semantic Similarity: After calculating the concept similarity and the similarity of concept case features of the semantic vectors, obtain the complete semantic similarity magnitude between the semantic vectors, and complete the information retrieval according to the complete semantic similarity magnitude.

[0010] As a preferred technical solution of the present invention, the specific steps of the said Step 1 are as follows: In the traditional vector model of the document, first create words, roots or phrases as dictionary keywords, then represent each document as a multi-dimensional vector, and finally represent the document in various symbols such as the binary vector, frequency or inverse frequency of the document. Replace the dictionary containing keyword entries through the ontology library, and use the meaning vector containing the concepts and their features described in the document to replace the document and meaningfully extract and absorb the content of the document.

[0011] As a preferred technical solution of the present invention, the specific steps of the said Step 3 are as follows: When calculating, in the ontology concepts of the tree structure, if there is such a hierarchical relationship between concept A and concept B that concept A and concept B are parent-child levels to each other, then A and B have a co-branch concept relationship. Calculate the distance d(A, B) between concept A and concept B through the following formula:

[0012] d(A, B) = dep(B) - dep(A) (1)

[0013] Wherein, dep(A) and dep(B) are respectively the depth values of concept A and concept B in the hierarchical structure;

[0014] When neither concept A nor concept B is an ancestor of the other in the ontology concepts of the tree - like hierarchy, it is in a different - branch state; and in this case, there is no other formal association between concept A and concept B; assume that concept R is the common ancestor level of concept A and concept B in the tree - like structure hierarchy without any association and has the maximum distance; take concept R as the minimum concept of concept A and concept B, denoted as R(A, B), and calculate the distance between them through the following formula:

[0015] d(A, B)=d(A, R)+d(B, R)(2)

[0016] There are only three possibilities for the relationship between any two concepts in the ontology concept tree, namely same - branch concepts, different - branch concepts, and identical concepts;

[0017] When calculating concept similarity, according to whether they are same - branch concepts or different - branch concepts, when the relationship between concept A and concept B gradually becomes distant as the distance between the two concept values increases, on this basis, the concept similarity between A and B will also decrease accordingly. Expressed as a functional relationship, as the independent variable distance x increases, the dependent variable similarity y gradually decreases; when the depth dep value between concept A and concept B is larger, the higher the probability that there are the same attribute values between A and B. Expressed as a functional relationship, as the independent variable depth x increases, the dependent variable similarity y gradually increases;

[0018] There is a connection between the number of concepts of A and B themselves and the number of semantic concepts related to them; when A and B are in the same - branch concept structure, the sub - concepts of A(R) are composed of two parts, where the two parts are the relevance of B and the semantic concept relevance of A and B respectively; and on this basis, when the weight value of the semantic - related concepts of A and B is the minimum, the relevance value of A and B is the maximum;

[0019] When A and B are in a different - branch concept structure, the sub - concepts of R are composed of the sub - concepts of A, the sub - concepts of B, and the semantic concepts related to A and B. If the proportion of the latter is low, it indicates a higher relevance between A and B;

[0020] After calculation, if the semantic similarity value is 1, there is no relationship between the related concepts and their depth. At this time, the distance between the two concepts is zero; define the concept similarity between concept A and concept B as follows:

[0021]

[0022] In formula (3), dep(R(A,B)) is the depth of the nearest root concept of concepts A and B; d(A,B) is the distance between concepts A and B; son(A) and son(B) are the numbers of all nodes in the subtrees rooted at A and B in the concept tree, respectively; α and β adjust the weight values of dep(R(A,B)) and d(A,B), respectively, and the value range is set as 0 ≤ Sim(A,B) ≤ 1.

[0023] As a preferred technical solution of the present invention, step 4 specifically includes the following steps:

[0024] Let the semantic vector V1 be A1[P1], A2[P2], …, A m [P m , and the semantic vector V2 be B1[Q1], B2[Q2], …, B m [Q m ;

[0025] Compare the similarities between the concepts and the semantic vectors of the document features, and set the maximum value among them as Max;

[0026]

[0027] In formula (4), V1 is the semantic vector value in the user query process, V2 is the semantic vector value in the document features, w is the relationship size of the weight value between the concept similarity and the attribute similarity, Sim v (V1,V2) is the similarity value between V1 and V2, Sim c (A i ,B j ) is the similarity of the nearest distance C between the semantic vectors i and j of concepts A and B, and m is the total number.

[0028] For the multimedia information retrieval method based on semantic similarity according to the present invention, compared with the prior art by adopting the above technical solution, it has the following technical effects:

[0029] The method of the present invention has high accuracy based on the semantic similarity method and is usable. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic diagram of the semantic vector process of the document and the user query in the embodiment of the present invention;

[0031] Figure 2 It is a schematic diagram of the semantic classification of the document under the ontology concept tree in the embodiment of the present invention;

[0032] Figure 3 It is a schematic diagram of establishing a document semantic index in the embodiment of the present invention;

[0033] Figure 4 This is the structural diagram of concepts A and B being concepts of the same branch in the embodiments of the present invention;

[0034] Figure 5 This is the structural diagram of concepts A and B being concepts of different branches in the embodiments of the present invention;

[0035] Figure 6 This is the schematic diagram of the information retrieval quantity of three methods in the embodiments of the present invention under different weight values. Detailed implementation manners

[0036] The following further explains the present invention in detail with reference to the accompanying drawings, so that those skilled in the art can understand the present invention more deeply and be able to implement it. However, the following reference examples are only used to explain the present invention and do not limit the present invention.

[0037] A multimedia information retrieval method based on semantic similarity includes the following steps: Step 1, query and extraction of semantic vectors: In the process of preprocessing the document, learn the method of the classical vector space model and replace the dictionary containing keyword entries with an ontology library, and use the meaning vectors containing the concepts and their features described in the document to replace the document and meaningfully extract and absorb the content of the document; Step 2, document semantic classification and indexing: According to the result of recording semantic vectors, obtain information from the semantic vectors based on the private ontology tree for classification; establish the semantic index state of the classified documents. First, insert the ontology concepts into the index file and arrange them in lexicographical order, then create an ordered linked list. This list first references the document semantic vector features under this concept, and then a pointer with a pointing link will be created in the index file, and the list will be designed on the pointer, and then the semantic attribute vector of the document will be linked to the corresponding document; Step 3, calculation of concept similarity: Determine the existence relationship between each concept instance and the attribute force, and analyze its influencing factors according to the concept characteristics to complete the entire process of semantic retrieval; Step 4, complete multimedia information retrieval based on semantic similarity: After calculating the concept similarity and the similarity of concept case features of the semantic vectors, obtain the complete semantic similarity magnitude between the semantic vectors, and complete information retrieval according to the complete semantic similarity magnitude.

[0038] Step 1 specifically includes the following steps: In the traditional vector model of the document, first create words, roots or phrases as dictionary keywords, then represent each document as a multi-dimensional vector, and finally represent the document in various symbols such as the binary vector, frequency or inverse frequency of the document. Replace the dictionary containing keyword entries with an ontology library, and use the meaning vectors containing the concepts and their features described in the document to replace the document and meaningfully extract and absorb the content of the document.

[0039] During the preprocessing operation of the document, the method of learning the classical vector space model is adopted. In the traditional vector model of the document, first, words, word roots, or phrases are created as dictionary keywords, then each document is represented as a multi-dimensional vector, and finally, the document is represented by various symbols such as the binary vector, frequency, or inverse frequency of the document. The dictionary containing keyword entries is replaced by an ontology library, and the meaning vector containing the concepts and their features described in the document is used to replace the document and meaningfully extract and absorb the content of the document.

[0040] According to how to handle usage problems, each piece of information contains a theme and content, and each record also describes a view system, which captures keywords by combining statistical and semantic methods and uses ideas and thoughts in the ontology to record and summarize, which has something in common with the method of user queries.

[0041] However, from the perspective of each concept example, the semantic existence attribute features should be extracted from the document. The specific process is as Figure 1 shown. To achieve the efficiency of the retrieval completion, the document is semantically classified to prepare for a more effective outreach plan and techniques for using the retrieved semantic vectors. According to the result of recording the semantic vectors, information is obtained from the semantic vectors based on the private ontology tree. As Figure 2 shown. In general, the semantic feature vector of a text involves various concept ontologies and their features, locates the appearance of the text in the categories corresponding to each concept, and logically establishes a hierarchical control structure corresponding to the text library itself, thus laying the foundation for the semantic indexing function of the text. To facilitate the retrieval operation, the classified documents are first set in the semantic indexing state. In the first step, the ontology concepts are inserted into the index file and arranged in lexicographical order, and then an ordered linked list is created. This list first references the document semantic vector features under this concept, and then a pointer with a link will be created in the index file, and the list is designed on the pointer. The purpose is to facilitate querying the loaded object through the pointer during the buffering process without having to scroll through the entire document set to reduce the time required for retrieval. Then the semantic attribute vector of the document is linked to the corresponding document, and the specific operation method is as Figure 3 shown.

[0042] Before performing multimedia information retrieval based on semantic similarity, the "partial" model strategy of the vector space is borrowed, and tools meaningful for the application problem and the document are borrowed, and ideas are put forward on the vector. The existence relationship between each concept instance and the attribute force is determined, and the influencing factors of its existence are analyzed according to the concept characteristics to complete the entire process of semantic retrieval.

[0043] The tree structure is used to express the hierarchical features reflected in the ontology. The semantic tree structure provides a theoretical basis for the retrieval of multimedia information. Since there are some similarities and connections between different ideas in the search process, it is necessary to check the relationship between different types of tree structures. For example, the relationship between nodes of the same level in the concept tree, or between nodes at the grandparent and grandchild level, on this basis, the concept similarity is calculated and the value is taken and quantified to achieve high-precision information retrieval.

[0044] Step 3 specifically includes the following steps: When calculating, when there is such a hierarchical relationship between concept A and concept B in the ontology concept of the tree structure that concept A and concept B are at the parent-child level, A and B are considered to have a same-branch concept relationship, and the distance d(A, B) between concept A and concept B is calculated by the following formula:

[0045] d(A, B) = dep(B) - dep(A) (1)

[0046] Among them, dep(A) and dep(B) are the depth values ​​of concept A and concept B in the hierarchy respectively; Figure 4 As shown;

[0047] When concept A and concept B are not mutually ancestral in the ontology concepts of the tree hierarchy, they are in a different branch state; and in this case, there is no other form of association between concept A and concept B; assuming that concept R is the common ancestral level of concept A and concept B in the tree structure hierarchy without any association, and has the farthest distance; concept R is the minimum value of concept A and concept B, expressed as R(A, B), and the distance between them is calculated by the following formula:

[0048] d(A,B)=d(A,R)+d(B,R) (2)

[0049] There are only three possibilities to obtain the relationship between any two concepts in the ontology concept tree, namely, concepts in the same branch, concepts in different branches, and the same concept; Figure 5 As shown;

[0050] When calculating the concept similarity, according to the same branch concept or different branch concept, when the relationship between concept A and concept B becomes increasingly distant as the distance between the two concept values ​​increases, the concept similarity between A and B will also decrease accordingly. It can be expressed as a function relationship that as the independent variable distance x increases, the dependent variable similarity y gradually decreases; when the depth dep value between concept A and concept B is larger, the more likely A and B have the same attribute value before. It can be expressed as a function relationship that as the independent variable depth x increases, the dependent variable similarity y gradually increases.

[0051] There is a connection between the number of conceptual numbers of A and B themselves and the number of semantic concepts related to them; when A and B are of the same branch concept structure, the sub - concepts of A(R) are composed of two parts, where the two parts are the relevance of B and the semantic concept relevance of A and B respectively; and on this basis, when the weight value of the semantic related concepts of A and B is the minimum value, the relevance value of A and B is the maximum;

[0052] When A and B are of different - branch concept structures, the sub - concepts of R are composed of three parts: the sub - concepts of A, the sub - concepts of B, and the semantic concepts related to A and B. If the proportion of the latter is low, it indicates a higher relevance between A and B;

[0053] After calculation, if the semantic similarity value is 1, there is no relationship between the related concepts and their depths. At this time, the distance between the two concepts is zero; the concept similarity between concept A and concept B is defined as follows:

[0054]

[0055] In formula (3), dep(R(A,B)) is the depth of the nearest root concept of concept A and concept B; d(A,B) is the distance between concept A and concept B; son(A) and son(B) are the number of all nodes of the sub - trees rooted at A and B in the concept tree respectively; α and β adjust the weight values of dep(R(A,B)) and d(A,B) respectively, and the value range is set as 0 ≤ Sim(A,B) ≤ 1.

[0056] Any ontology concept will have instances in various situations, and the criterion for distinguishing them is the difference in attribute values. There may be the same attributes among the instances with different concepts. Based on this, after calculating the concept similarity between instances, the implementation of information retrieval is completed. After calculating the concept similarity of semantic vectors and the similarity of concept case features, the complete semantic similarity between semantic vectors can be obtained. When applying semantic expansion technology to information retrieval, the similarity is mainly manifested in the ability to expand concepts to meet the user's query requirements.

[0057] Step 4 specifically includes the following steps:

[0058] Let the semantic vector V1 be A1[P1], A2[P2], …, A m [P m , and the semantic vector V2 be B1[Q1], B2[Q2], …, B m [Q m ;

[0059] Compare the similarity between the concept and the semantic vector of the document features, and set the maximum value among them as Max;

[0060]

[0061] In formula (4), V1 is the semantic vector value during the user's query process, V2 is the semantic vector value in the document features, w is the magnitude of the weight value relationship between the concept similarity and the attribute similarity, and Sim v (V1, V2) is the similarity value between V1 and V2, and Sim c (A i , B j ) is the similarity of the closest distance C between the semantic vectors i and j of concept A and concept B, and m is the total quantity.

[0062] In specific implementation, to verify the feasibility of this method, the multimedia information retrieval method based on semantic similarity is compared with the traditional method and the multimedia information retrieval method based on data analysis. First, relevant network documents from university homepages such as www.ustc.edu.cn and www.pku.edu.cn are selected. The test environment in the experiment is a Windows PC, and the parameters are Genuine Intel(R) CPU T2080 @ 2.67GHz CPU, 16GB of memory, and 1TB of hard disk. The operating system is Windows 10, and it is known that the software has three problems in terms of performance.

[0063] The number of information retrieved after multimedia information retrieval by the three methods is used as a comparison index, and the concept similarity weight is used as a variable. As Figure 6 shown, it is the number of information retrieved by the three methods under different weight values. It is known that in this experiment, the complete retrieval for multimedia information retrieval should be 1200; from the data obtained through the above experiments, it can be seen that when using the traditional method for multimedia information retrieval, when the weight value is 0, the number of information retrieved is 200, which is the minimum among the three. When the weight value is 1.0, the number of information retrieved is 600, which is closest to the standard value compared with other data, but the accuracy rate is only 50%, and when the weight value is between 0 - 0.8, the number of information retrieved is between 200 - 400. When using the data analysis-based method for multimedia information retrieval, when the weight value is between 0.6 - 0.7, the number of information retrieved is the highest, which is 800, and its accuracy rate is 67%, but when it is greater than 0.6, the number of information retrieved shows a downward trend, and when the weight value is 1.0, the number of information retrieved is only 200. Generally speaking, when using the data analysis-based method for multimedia information retrieval, the results will show large fluctuations and there is volatility. However, when using the method of the present invention, it is greater than the other two methods at any weight value, and the number of information retrieved is between 750 - 1200, and the highest accuracy rate can reach 100%. Thus, it can be seen that when using the present invention for information retrieval, it can maintain a high accuracy rate, and is more advantageous and applicable.

[0064] The present invention proposes a multimedia information retrieval method based on semantic similarity. During the preprocessing operation of documents, the method of learning the classical vector space model is adopted, and the dictionary containing keyword entries is replaced by an ontology library. The meaning vectors containing the concepts described in the document and their features are used to replace the document, and the content of the document is meaningfully extracted and absorbed. To achieve the high efficiency of the retrieval completion, the documents are semantically classified to prepare for a more effective abduction plan and technique for using the retrieved semantic vectors. When calculating the concept similarity and attribute similarity, the existence relationship between each concept instance and the attribute power is determined, and the influencing factors are analyzed according to the concept characteristics to complete the entire process of semantic retrieval. Finally, the multimedia information retrieval is completed based on the semantic similarity. Through comparative experiments, the multimedia information retrieval method based on semantic similarity is compared with the traditional method and the method based on data analysis, and it is concluded that the method based on semantic similarity has a high accuracy rate and is usable.

[0065] The specific implementation manners described above further elaborate on the purpose, technical solution and beneficial effects of the present invention. It should be understood that the above is only the specific implementation manner of the present invention and is not intended to limit the scope of the present invention. Any equivalent changes and modifications made by those skilled in the art without departing from the concept and principle of the present invention shall fall within the scope of protection of the present invention.

Claims

1. A multimedia information retrieval method based on semantic similarity, characterized in that, The method includes the following steps: Step 1, querying and extracting semantic vectors: In the process of preprocessing the document, learn the method of the classical vector space model and replace the dictionary containing keyword entries with an ontology library, and use the semantic vectors containing the concepts and their features described in the document to replace the document and meaningfully extract and absorb the content of the document; Step 2, document semantic classification and indexing: According to the result of recording the semantic vectors, obtain information from the semantic vectors based on the private ontology tree for classification; establish the semantic index state of the classified documents. First, insert the ontology concepts into the index file and arrange them in lexicographical order, then create an ordered linked list. This linked list first references the document semantic vector features under this concept, and then a pointer with a pointing link will be created in the index file, and the list will be designed on the pointer, and then link the semantic attribute vector of the document to the corresponding document; Step 3, calculating concept similarity: Determine the existence relationship between each concept instance and the attribute force, and analyze its influencing factors according to the concept characteristics to complete the entire process of semantic retrieval; Define the concept similarity between concept A and concept B as follows: In formula (3), dep(R(A, B)) is the depth of the nearest root concept of concept A and concept B; d(A, B) is the distance between concept A and concept B; son(A) and son(B) are the numbers of all nodes in the subtrees rooted at A and B in the concept tree respectively; α and β respectively adjust the weight values of dep(R(A, B)) and d(A, B), and the value range is set as 0 ≤ Sim(A, B) ≤ 1; Step 4, completing multimedia information retrieval based on semantic similarity: After calculating the concept similarity and the similarity of concept case features of the semantic vectors, obtain the complete semantic similarity between the semantic vectors, and complete the information retrieval according to the complete semantic similarity; The specific steps of step 4 include the following steps: Let the semantic vector V1 be A1[P1], A2[P2], …, A m [P m , and the semantic vector V2 be B1[Q1], B2[Q2], …, B m [Q m ; Compare the similarity between the concept and the document feature semantic vector, and set the maximum value among them as Max; In formula (4), V1 is the semantic vector value in the user query process, V2 is the semantic vector value in the document feature, w is the magnitude of the weight value relationship between the concept similarity and the attribute similarity, and Sim v (V1, V2) is the similarity value between V1 and V2, and Sim c (A i , B j ) is the similarity of the closest distance C between the semantic vectors i and j of concept A and concept B, and m is the total quantity.

2. The multimedia information retrieval method based on semantic similarity according to claim 1, characterized in that, The specific steps of step 1 include the following steps: In the traditional vector model of the document, first create words, word roots or phrases as dictionary keywords, then represent each document as a multi-dimensional vector, and finally represent the document in terms of the binary vector, frequency or inverse frequency of the document. Replace the dictionary containing keyword entries with an ontology library, and use the semantic vectors containing the concepts and their features described in the document to replace the document and meaningfully extract and absorb the content of the document.

3. The multimedia information retrieval method based on semantic similarity according to claim 2, characterized in that, The specific steps of step 3 include the following steps: When calculating, in the ontology concepts of the tree structure, if there is such a hierarchical relationship between concept A and concept B that concept A and concept B are parent-child levels to each other, then A and B have a co-branch concept relationship, and the distance d(A, B) between concept A and concept B is calculated by the following formula: d(A, B) = dep(B) - dep(A) (1) Where, dep(A) and dep(B) are the depth values of concept A and concept B in the hierarchical structure respectively; When neither concept A nor concept B is an ancestor of the other in the ontological concepts of the tree - like hierarchy, it is in a different - branch state; and in this case, there is no other formal association between concept A and concept B; assume that concept R is the common ancestor level of concept A and concept B in the tree - like structure hierarchy without any association and has the maximum distance; take concept R as the minimum concept of concept A and concept B, denoted as R(A, B), and calculate the distance between them through the following formula: d(A, B)=d(A, R)+d(B, R)(2) There are only three possibilities for the relationship between any two concepts in the ontological concept tree, namely same - branch concepts, different - branch concepts, and identical concepts; When calculating concept similarity, according to whether the concepts are same - branch or different - branch, when the relationship between concept A and concept B gradually becomes distant as the distance between the two concept values increases, on this basis, the concept similarity between A and B also decreases accordingly. It can be expressed by a functional relationship that as the independent variable distance x increases, the dependent variable similarity y gradually decreases; when the depth dep value between concept A and concept B is larger, the higher the probability that there are the same attribute values between A and B. It can be expressed by a functional relationship that as the independent variable depth x increases, the dependent variable similarity y gradually increases; There is a connection between the number of concepts of A and B themselves and the number of semantic concepts related to them; when A and B are in the same - branch concept structure, the sub - concepts of A(R) are composed of two parts, where the two parts are the relevance of B and the semantic concept relevance of A and B respectively; and on this basis, when the weight value of the semantic related concepts of A and B is the minimum, the relevance value of A and B is the maximum; When A and B are in a different - branch concept structure, the sub - concepts of R are composed of the sub - concepts of A, the sub - concepts of B, and the semantic concepts related to A and B. If the proportion of the latter is low, it indicates a higher relevance between A and B; After calculation, if the semantic similarity value is 1, there is no relationship between the related concepts and their depth. At this time, the distance between the two concepts is zero.

Images