Method and system for generating implicit knowledge link, and storage medium

By analyzing and matching the semantic associations of conceptual words in text data, multiple implicit knowledge links are generated, which solves the problem of difficult to discover multipath multi-group implicit knowledge links in the existing technology, and achieves more refined knowledge Graph construction and knowledge discovery.

CN118982068APending Publication Date: 2024-11-19SHANGHAI BANPO NETWORK TECH
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411012845.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-26
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

The prior art is difficult to discover and mine the multipath multi-group implicit knowledge links in text information, which limits the knowledge discovery of text content and the construction of knowledge graphs.

Method used

By obtaining target text data, analyzing conceptual words, extracting semantic association relationships, using conceptual word weight recommendation algorithm to generate implicit knowledge links of triples, quads and 5 tuples, dynamically adjusting semantic association thresholds to match potential Knowledge path.

Benefits of technology

It realizes a more complete and more refined knowledge graph construction, improves the accuracy and relevance of search results, reveals unknown related information in text information, and enhances the comprehensiveness and depth of the knowledge link.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118982068A_ABST
    Figure CN118982068A_ABST
Patent Text Reader

Abstract

The invention relates to the field of textual literature information processing, and discloses a generation method and system of an implicit knowledge link and a storage medium, and the method comprises the following steps: obtaining target text data; the concept words and semantic association thereof are analyzed; any concept word is anchored from the word set to serve as a starting point A, and a semantic association subset A {} corresponding to the current starting concept word A is obtained; any other concept word is anchored from the word set to serve as a current target B, and a semantic association subset B {} corresponding to the current target concept word B is obtained; based on the initial semantic association set A {} and the target semantic association set B {}, matching all common improved words x [i]; generating and outputting a triple implicit knowledge link set Alt of n paths derived by a single factor (x); = gt; x [i] lt; = gt; b. In the same way, a quintuple implicit knowledge link set Alt of even three factors (x, z, y) between concept words A and B can be deduced; = gt; x [i] lt; = gt; z [k] lt; = gt; y [j] lt; = gt; b. The disclosure of the multi-path multi-tuple implicit knowledge link is beneficial to discovery of new knowledge in text content.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of text document information processing, and in particular to a method, system and storage medium for generating implicit knowledge links. Background Art

[0002] There has never been so much massive text information in human history, so how to find various information association clues from text information and sort out clear knowledge veins is of great significance to the inheritance and development of text knowledge content. Traditional text information reading is an intellectual process of sequential browsing and discovering knowledge associations in the browsing process.

[0003] Text information can be a single document or an arbitrary set of documents. The content of a document has information block characteristics such as sentences and paragraphs. Among them, a paragraph is composed of a number of sentences in sequence; a sentence is composed of a number of conceptual words in sequence. In the same sentence of a document, when a conceptual word (x) and another conceptual word (y) exist at the same time, we say that (x) and (y) form an explicit knowledge link (x<->y) with the same sentence co-occurrence attribute.

[0004] The explicit knowledge link reflects a direct semantic association relationship. In the text analysis process of the current text set, all directly related concept word clusters (y{}, y can be 1-n concept words) are derived and revealed from a concept word (x) to realize the heuristic divergent thinking disclosure of the concept word (x).

[0005] However, among the existing methods for text document processing, no method for mining implicit knowledge links of multi-path tuples has been found.

[0006] The implicit knowledge link is any two conceptual terms (A, B), trying to deduce their potential third conceptual term (x) and find the implicit knowledge link of the triple A<->x<->B.

[0007] A multi-path tacit knowledge link means that x[i] in the tacit knowledge link A<->x[i]<->B can be 1-n recommended conceptual terms, forming 1-n tacit knowledge link paths.

[0008] A multi-tuple implicit knowledge link refers to a set of two-factor four-tuple implicit knowledge links A<=>x[i]<=>y[j]<=>B between A and B. Even a set of three-factor five-tuple implicit knowledge links A<=>x[i]<=>z[k]<=>y[j]<=>B.

[0009] The disclosure of implicit knowledge links of multi-path multi-tuples helps to discover new knowledge in text content.

[0010] The purpose of the present invention is to find a traceable multi-path multi-group implicit knowledge link discovery reasoning method. Summary of the invention

[0011] In order to calculate implicit knowledge links based on existing text documents, the present application provides a method, system and storage medium for generating implicit knowledge links.

[0012] In the first aspect, the present application provides a method for generating an implicit knowledge link, which adopts the following technical solution:

[0013] A method for generating a tacit knowledge link comprises the following steps:

[0014] Obtain target text data; the target text data may be a single text document or multiple text documents;

[0015] Parsing the concept words in the target text data to form a word set; and extracting the semantic association relationship between the concept words;

[0016] Anchoring the current starting concept word A from the word set;

[0017] Anchoring the current target concept word B from the word set;

[0018] Extract words that are semantically related to the current starting concept word A from the word set to form a starting semantic association set A{};

[0019] Extracting words that are semantically related to the current target concept word B from the word set to form a target semantic association set B{};

[0020] Excluding the direct association with B from the starting semantic association set A{}; excluding the current starting concept word A from the target semantic association set B{};

[0021] Based on the starting semantic association set A{} and the target semantic association set B{}, a common improved word set x[i] is matched, where (i=1…n), if and only if A<=>x[i] and x[i]<=>B exists;

[0022] According to the concept word weight recommendation algorithm, a triple implicit knowledge link set A<=>x[i]<=>B of a single factor (x) is generated and output.

[0023] By adopting the above technical solution, through text data analysis, the potential semantically related knowledge paths between different concepts are mined and constructed. Through one or more sets of implicit knowledge links, these links reveal the potential relationship between A and B through x[i]; it is conducive to building a more complete and detailed knowledge graph; improving the accuracy and relevance of search results, and discovering more unknown related information through implicit links.

[0024] Optionally, the concept word weight recommendation algorithm includes the following steps:

[0025] Get the word frequency (TF) corresponding to concept word A and concept word B, where word frequency (TF) = the number of times the current concept word (W) appears in the document set / the total number of words in the document set;

[0026] Get the inverse document frequency (IDF) corresponding to concept word A and concept word B; inverse document frequency (IDF) = log (total number of documents in the current document set / number of documents containing the current concept word (W) + 1);

[0027] Calculate the concept word weight, concept word weight Weight (W) = word frequency (TF) × inverse document frequency (IDF);

[0028] Normalize the concept word weight Weight (W) and map it to the range of [0,1]:

[0029] normalized_Weight(W)=(Weight(W)–min_Weight(W) /

[0030] (max_Weight(W)-min_Weight(W);

[0031] Among them, min_Weight(W) and max_Weight(W) are the minimum and maximum values ​​of Weight(W) of all words in the text respectively.

[0032] By adopting the above technical solution, the semantic association relationship of the text content can simultaneously have relationship weight analysis based on different levels of sentences, paragraphs, and articles (chapters).

[0033] Optionally, the step of forming the initial semantic association set A{} further includes the following sub-steps:

[0034] Calculating the semantic relevance between each word element in the word set and the current starting concept word A;

[0035] The word element whose semantic association is greater than a preset first semantic reference value is regarded as a word having a semantic association with the current starting concept word A;

[0036] The first semantic reference value is adjusted in a positive correlation according to the proportion of the weight of the current starting concept word A in the target text data.

[0037] By adopting the above technical solution, the first semantic reference value is adjusted in a positive correlation according to the weight proportion of A. Specifically, if A appears in a large number of texts, the first semantic reference value can be reduced to capture more words potentially related to A; conversely, if A appears only in a few texts, the first semantic reference value should be maintained or increased to ensure the strong relevance of the selected words.

[0038] Optionally, the step of forming the target semantic association set B{} further includes the following sub-steps:

[0039] Calculate the semantic relevance between each word element in the word set and the current target concept word B;

[0040] The word element whose semantic association is greater than a preset second semantic reference value is regarded as a word having a semantic association with the current target concept word B;

[0041] The second semantic reference value is adjusted according to the anti-correlation between the weight ratio of the current target concept word B and the weight ratio of the current starting concept word A.

[0042] By adopting the above technical solution, a semantic association set closely related to the target concept word B can be formed more accurately, providing strong support for subsequent knowledge link construction and analysis.

[0043] Optionally, the step of matching the common improved word set x[i] further includes the following sub-steps:

[0044] Calculate the first semantic association degree a of each word element in the starting semantic association set A{} and the current starting concept word A;

[0045] Calculate the second semantic association degree b between each word element in the target semantic association set B{} and the current target concept word B;

[0046] A bidirectional correlation c is calculated according to the first semantic correlation a and the second semantic correlation b, and the calculation formula is: c=a-first semantic reference value+b-second semantic reference value;

[0047] Then the bidirectional association degree c is compared with a preset bidirectional reference value, and the set of word elements whose bidirectional association degree c is greater than the bidirectional reference value is the word set x[i].

[0048] By adopting the above technical solution, the jointly improved word set x[i] can be matched more accurately, providing strong support for subsequent text analysis or natural language processing tasks.

[0049] Optionally, the method further comprises the following sub-steps:

[0050] According to the ratio between the number of elements in the starting semantic association set A{} and the number of elements in the target semantic association set B{}, the first semantic reference value is adjusted in a positive correlation, and the second semantic reference value is adjusted in an anti-correlation manner;

[0051] Alternatively, the first semantic reference value is adjusted in a positive correlation and the second semantic reference value is adjusted in an anti-correlation according to the ratio of the average semantic association degree of the elements in the starting semantic association set A{} to the average semantic association degree of the elements in the target semantic association set B{};

[0052] The bidirectional reference value is adjusted in positive correlation according to the absolute value of the difference between the first semantic reference value and the second semantic reference value.

[0053] By adopting the above technical solution, the relative relationship between the starting semantic association set A{} and the target semantic association set B{} is considered to dynamically adjust the first and second semantic reference values, and a bidirectional reference value is introduced to balance the importance of the two sets.

[0054] Optionally, the method further comprises the following steps:

[0055] Acquire words that are semantically associated with the starting semantic association set A{} from A{} to form an intermediate semantic association set x{};

[0056] Acquire words that are semantically associated with the target semantic association set B{} from B{} to form an intermediate semantic association set y{};

[0057] Excluding the direct association with B{} from the intermediate semantic association set x{} of A, and excluding the direct association with A{} from the intermediate semantic association set y{} of B;

[0058] Match the common improved word path set x[i]<=>y[j], where (i=1…m, j=1…n), if and only if A<=>x[i] and y[j]<=>B path exists;

[0059] Thus, a set of two-factor four-tuple implicit knowledge links is generated and output.

[0060] A<=>x[i]<=>y[j]<=>B.

[0061] By adopting the above technical solution, we can construct a path from A to x[i] based on the direct association path of each semantic segment.

[0062] The two-factor implicit knowledge link A<=>x[i]<=>y[j]<=>B (A<=>x[i]); from x[i] to y[j](x[i]<=>y[j]); and then from y[j] to B(y[j]<=>B) directly shows the potential relationship path between A and B through specific words or word combinations.

[0063] Optionally, the method further comprises the following steps:

[0064] Obtaining words that are semantically associated with the starting semantic association set A{} from A{} to form an intermediate semantic association set x{}; obtaining words that are semantically associated with x{} from x{} to form an intermediate semantic association set z1{};

[0065] Acquire words that are semantically associated with the target semantic association set B{} to form an intermediate semantic association set y{}; acquire words that are semantically associated with y{} from y{} to form an intermediate semantic association set z2{};

[0066] Excluding the direct association with B{} and excluding the direct association with y{} from z1{}; excluding the direct association with A{} and excluding the direct association with x{} from z2{};

[0067] Match z1{} and z2{} to obtain a common improved word set z[k], where (k=1…o), if and only if A<=>x[i], x[i]<=>z[k], and z[k]<=>y[j], y[j]<=>B exists;

[0068] Thus, a three-factor five-tuple implicit knowledge link set A<=>x[i]<=>z[k]<=>y[j]<=>B is generated and output.

[0069] By adopting the above technical solution, the generated three-factor five-tuple implicit knowledge link set provides a new relational view for understanding and explaining the complex relationship between A and B, which helps to reveal the deep and complex knowledge structure and its interactive correlation pattern hidden behind the text information.

[0070] In the second aspect, the present application provides a system for generating implicit knowledge links, which adopts the following technical solutions:

[0071] A system for generating a tacit knowledge link comprises a processor, wherein a program of any one of the above-mentioned methods for generating a tacit knowledge link is running on the processor.

[0072] In a third aspect, the present application provides a storage medium, which adopts the following technical solution:

[0073] A storage medium stores a program for the method for generating a tacit knowledge link as described in any one of the above.

[0074] In summary, the present application includes at least one of the following beneficial technical effects:

[0075] 1. By dynamically calculating and adjusting the threshold of semantic relevance, i.e., the second semantic reference value, the present application can more accurately identify the word elements that have actual semantic relevance to the current target concept word B. This dynamic adjustment mechanism takes into account the relative proportion of the number of occurrences of the target concept word B and the starting concept word A in the text or data set, thereby avoiding the deviation that may be caused by a fixed threshold and improving the accuracy and reliability of semantic relevance.

[0076] 2. Based on the semantic association set A{} of the starting concept word A and the semantic association set B{} of the target concept word B, the present application further expands the construction of up to three-factor pairs ((x[i]<=>z[k])<=>y[i]) by matching the elements in the association set x<=>z<=>y, thereby revealing more complex and in-depth knowledge links. These links not only connect the starting concept word A and the target concept word B, but also show the multiple relationships and interaction patterns between them through intermediate elements (such as x, z, y), enhancing the comprehensiveness and depth of knowledge link construction.

[0077] 3. The method of the present application effectively mines and extracts implicit knowledge hidden in text or data through systematic steps, including excluding unrelated sets, excluding overlapping elements, matching related elements, etc. from the starting semantic association set and the target semantic association set. This method not only improves the efficiency of implicit knowledge discovery, but also makes the discovered knowledge more practical and can provide strong support for decision support, knowledge reasoning and other fields.

[0078] 4. The method of this application has certain flexibility and scalability, and can be adjusted and optimized according to different application scenarios and requirements. For example, according to the complexity of the specific problem and the characteristics of the data, a suitable semantic relevance calculation method and threshold adjustment strategy can be designed; the association set hierarchy can also be expanded according to the needs, and the construction method of four factors (x<=>s<=>t<=>y) or even expanded to five factors (x<=>s<=>z<=>t<=>y) can be derived to adapt to different analysis goals and knowledge representation requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 It is a step diagram of a method for generating implicit knowledge links.

[0080] Figure 2 It is a word display diagram between the starting concept word "immune response" and the current interest target word "glioma".

[0081] Figure 3 It is a potential knowledge connection link diagram such as "T cell" and "neoantigen" derived by the system between the starting concept word "immune response" and the current target word of interest "glioma".

[0082] Figure 4 It is a graph of sub-steps to form the starting semantic association set A{}.

[0083] Figure 5 It is a graph of sub-steps for forming the target semantic association set B{}.

[0084] Figure 6 It is a sub-step graph for matching the common improved word set x[i].

[0085] Figure 7 It is a step diagram for outputting a set of ternary implicit knowledge links. DETAILED DESCRIPTION

[0086] Embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings.

[0087] In the description of this specification, the description with reference to the terms "certain embodiments", "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiments or examples are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0088] The present application embodiment discloses a method for generating an implicit knowledge link, referring to Figure 1 and Figure 2 and Figure 3 , including the following steps:

[0089] Obtain target text data; the target text data may be a document library, an article package, an electronic book, a web page, etc. The target text data may be a single text document or multiple text documents.

[0090] Parse the concept words in the target text data to form a word set; and extract the semantic association relationship between the concept words; the word set can use natural language processing technology, such as word segmentation, part-of-speech tagging and other methods to decompose the text data into words.

[0091] Anchor the current starting concept word A from the word set; the current starting concept word A can be a search word or any current interest entry word. For example, in the text document parsing result, the current starting concept word A is selected as "immune response" (the first column on the left in the semantic association matrix).

[0092] Words that are semantically associated with the current starting concept word A are extracted from the word set to form a starting semantic association set A{}; semantic analysis techniques, such as dictionary-based similarity calculation, word vector model, etc., can be used to find words that are semantically associated with A.

[0093] Anchor the current target concept word B of interest from the word set; for example, in this case, the current target concept word B is selected as "glioma" (black node in the semantic association matrix).

[0094] Words that are semantically associated with the current target concept word B are extracted from the word set to form a target semantic association set B{}. Semantic analysis techniques, such as dictionary-based similarity calculation and word vector models, can be used to find words that are semantically associated with B.

[0095] The current target concept word B is excluded from the starting semantic association set A{}. Furthermore, the current starting concept word A is excluded from the target semantic association set B{}. This excludes the direct association between A and B.

[0096] Based on the starting semantic association set A{} and the target semantic association set B{}, a common improved word set x[i] is matched, where (i=1...n). Similarity thresholds, clustering algorithms and other methods can be used to compare A{} and B{} to find words that are common or semantically similar in the two sets to form a common improved word set x[i].

[0097] According to the concept word weight recommendation algorithm, an implicit knowledge link A<=>x[i]<=>B from A to x[i] and then to B is constructed, that is, a triple implicit knowledge link set A<=>x[i]<=>B of a single factor (x) is generated and output.

[0098] For example, in the case, there is an implicit knowledge link of “immune response” – “T cell” – “glioma”.

[0099] The current target concept word B is the target concept selected arbitrarily in the word set. If and only if there is a concept word x between the current starting concept word A and the current target concept word B, the concept nodes of the concept word x constitute a common improved word set, that is, the potential knowledge connection node A<=>x<=>B between the current starting concept word A and the current target concept word B. This type of x concept node is also the potential extended knowledge node based on the current starting concept word A and the current target concept word B that people try to discover in the process of literature reading. A<=>x<=>B is also called the intermediate implicit knowledge link. Through text data analysis, the potential semantically related knowledge paths between different concepts are mined and constructed. Through one or more implicit knowledge link sets, these links reveal the potential relationship between A and B through x[i]; it is conducive to building a more complete and refined knowledge graph; improving the accuracy and relevance of search results, and discovering more unknown related information through implicit links.

[0100] The concept word weight recommendation algorithm includes the following steps:

[0101] Get the word frequency (TF) corresponding to concept word A and concept word B, where word frequency (TF) = the number of times the current concept word (W) appears in the document set / the total number of words in the document set;

[0102] Get the inverse document frequency (IDF) corresponding to concept word A and concept word B; inverse document frequency (IDF) = log (total number of documents in the current document set / number of documents containing the current concept word (W) + 1);

[0103] Calculate the concept word weight, concept word weight Weight (W) = word frequency (TF) × inverse document frequency (IDF);

[0104] Normalize the concept word weight Weight (W) and map it to the range of [0,1]:

[0105] normalized_Weight(W)=(Weight(W)–min_Weight(W) /

[0106] (max_Weight(W)-min_Weight(W);

[0107] Among them, min_Weight(W) and max_Weight(W) are the minimum and maximum values ​​of Weight(W) of all words in the text respectively.

[0108] The semantic association relationship of text content also has relationship weight analysis based on different levels of sentences, paragraphs, and articles (chapters).

[0109] Reference Figure 4 , the step of forming the initial semantic association set A{} also includes the following sub-steps:

[0110] Calculate the semantic relevance between each word element in the word set and the current starting concept word A; semantic analysis techniques, such as cosine similarity based on word vectors, path length in semantic networks, statistical information based on knowledge bases or corpora, etc., can be used to calculate the semantic relevance between each word element in the word set and the current starting concept word A. A relevance score list is obtained, in which each word element corresponds to a semantic relevance value with A.

[0111] Word elements with a semantic relevance greater than a preset first semantic reference value are regarded as words that are semantically associated with the current starting concept word A. A preset first semantic reference value is set as a threshold. Only word elements with a semantic relevance greater than this threshold are considered to be semantically associated with A. Traverse the list of relevance scores and add word elements greater than the first semantic reference value to the starting semantic association set A{}.

[0112] The distribution of the current starting concept word A in the target text data (such as the proportion of articles in which it appears) will affect the judgment of its semantic relevance. If A appears in multiple texts, then its semantic environment may be wider, and the first semantic reference value needs to be appropriately adjusted to include more possible related words.

[0113] The first semantic reference value is adjusted in a positive correlation according to the weight ratio of the current starting concept word A in the target text data. Specifically, if A appears in a large number of texts, the first semantic reference value can be lowered to capture more words potentially related to A; conversely, if A only appears in a few texts, the first semantic reference value should be maintained or increased to ensure the strong relevance of the selected words. Before or after calculating the semantic relevance, the first semantic reference value is dynamically adjusted according to the weight ratio of A, and then the screening operation is performed. Through the above method, the starting semantic association set A{} can be constructed more accurately, providing a solid foundation for the subsequent generation of implicit knowledge links.

[0114] Reference Figure 5 , the step of forming the target semantic association set B{} also includes the following sub-steps:

[0115] Calculate the semantic association between each word element in the word set and the current target concept word B; use the same or similar semantic analysis technology as when forming A{} to calculate the semantic association between each word element in the word set and the current target concept word B; obtain a list of association scores related to B.

[0116] Word elements whose semantic association is greater than a preset second semantic reference value are regarded as words that are semantically associated with the current target concept word B; a preset second semantic reference value is set as a threshold, and only word elements whose semantic association is greater than this threshold are considered to be semantically associated with B; the association score list is traversed, and word elements whose semantic association is greater than the second semantic reference value are added to the target semantic association set B{}.

[0117] The weight ratio of the current target concept word B to the current starting concept word A will affect the judgment of the semantic relevance of B. If B appears less frequently than A, then the second semantic reference value may need to be lowered to capture more words potentially related to B; conversely, if B appears more frequently, the second semantic reference value should be maintained or increased to ensure strong relevance of the selected words.

[0118] The second semantic reference value is adjusted in an anti-correlation based on the weight ratio of the current target concept word B and the weight ratio of the current starting concept word A. If the weight ratio of B is much lower than that of A, the second semantic reference value should be lowered; if the weight ratio of B is close to or higher than that of A, the second semantic reference value should be maintained or appropriately increased. The second semantic reference value is adjusted in an anti-correlation based on the ratio. If the weight ratio of B is very high relative to A, it means that B is more important or appears frequently in the context, so the second semantic reference value can be appropriately lowered to capture more words related to B; conversely, if the weight ratio of B is low, the second semantic reference value is increased to ensure that only those words that are highly related to B are selected. Through the above adjustment method, a semantic association set closely related to the target concept word B can be formed more accurately, providing strong support for the subsequent knowledge link construction and analysis.

[0119] In this way, the target semantic association set B{} is constructed more flexibly, ensuring that it contains words that are strongly related to B and takes into account the relative importance of B and A in the text data. The method of dynamically adjusting the reference value helps to improve the accuracy and adaptability of implicit knowledge link generation.

[0120] Reference Figure 6 , the step of matching the common improved word set x[i] also includes the following sub-steps:

[0121] Calculate the first semantic association a between each word element in the starting semantic association set A{} and the current starting concept word A; evaluate by calculating the similarity between word vectors, such as cosine similarity.

[0122] Calculate the second semantic association b between each word element in the target semantic association set B{} and the current target concept word B; similarly, evaluate by calculating the similarity between word vectors, such as cosine similarity.

[0123] The bidirectional association degree c is calculated according to the first semantic association degree a and the second semantic association degree b, and the calculation formula is: c=a-first semantic reference value+b-second semantic reference value.

[0124] For example, in the starting semantic association set A{}, all words have a degree of association with A greater than 50 (assuming that the degree of association is in the range of 0-100 for easy calculation), that is, a is greater than 50. Similarly, in the target semantic association set B{}, all words have a degree of association with B greater than 50 (assuming that the degree of association is in the range of 0-100 for easy calculation), that is, b is greater than 50. Assuming that the first semantic reference value and the second semantic reference value are both 50, then c is always a number greater than 0.

[0125] The bidirectional correlation c is then compared with a preset bidirectional reference value, and the set of word elements whose bidirectional correlation c is greater than the bidirectional reference value is the word set x[i]. The threshold comparison method is used to determine whether the word element is "relevant" enough to be included in the final word set x[i].

[0126] Through the above method, the jointly improved word set x[i] can be matched more accurately, providing strong support for subsequent text analysis or natural language processing tasks.

[0127] The method further comprises the following sub-steps:

[0128] The first semantic reference value is adjusted positively and the second semantic reference value is adjusted anti-correlatedly according to the ratio between the number of elements in the starting semantic association set A{} and the number of elements in the target semantic association set B{}. The first semantic reference value is adjusted positively and the second semantic reference value is adjusted anti-correlatedly according to the ratio between the number of elements in A{} and the number of elements in B{}, that is, if A{} has more elements than B{}, the first semantic reference value is increased to capture more words related to A, and the second semantic reference value is adjusted anti-correlatedly, that is, if A{} has more elements than B{}, the second semantic reference value is decreased to relax the selection criteria for words related to B.

[0129] Alternatively, the first semantic reference value is adjusted positively and the second semantic reference value is adjusted anti-correlatedly according to the ratio of the average semantic association of the elements in the starting semantic association set A{} to the average semantic association of the elements in the target semantic association set B{}. The first semantic reference value is adjusted positively and the second semantic reference value is adjusted anti-correlatedly according to the ratio of the two average semantic associations of the elements in A{} and B{}, that is, if the average association of A{} is higher than that of B{}, the first semantic reference value is increased, and the second semantic reference value is adjusted anti-correlatedly, that is, if the average association of A{} is higher than that of B{}, the second semantic reference value is decreased.

[0130] A bidirectional reference value is adjusted according to the positive correlation of the absolute value of the difference between the first semantic reference value and the second semantic reference value. A bidirectional reference value is adjusted according to the positive correlation of the absolute value of the difference between the first semantic reference value and the second semantic reference value. The bidirectional reference value will be used to give different weights or priorities to words from A{} and B{} when matching the common improved word set x[i] to balance the contributions of the two sets.

[0131] In actual implementation, the element quantity adjustment method, the semantic relevance average adjustment method, or a combination of the two can be used to more accurately adjust the first and second semantic reference values ​​according to the specific situation. The introduction of bidirectional reference values ​​provides additional flexibility for the generation of implicit knowledge links, and its influence in the matching process can be adjusted as needed. The purpose of these adjustment steps is to ensure that the generation of implicit knowledge links takes into account both the individual characteristics of concept terms A and B and the relative relationship between them, thereby generating more accurate and meaningful links.

[0132] The method further comprises the steps of:

[0133] From the starting semantic association set A{}, words that are semantically associated with A{} are obtained to form an intermediate semantic association set x{}; from the target semantic association set B{}, words that are semantically associated with B{} are obtained to form an intermediate semantic association set y{}; and paths x[i]<=>y[j] are matched, where (i=1…n), (j=1…n). More complex semantic analysis techniques, such as semantic models based on deep learning, can be used to calculate the semantic similarity between elements in A{} and B{}. Pairs of elements whose similarity exceeds a certain threshold are considered to be directly associated and added to x[i]<=>y[j]. Combinations of elements in A{} and B{} are considered, especially phrases or word groups that do not belong to each other's sets individually, but have strong semantic associations with both A and B when combined. One or more semantically directly associated path sets x[i]<=>y[j] are obtained, each of which contains words or word combinations that are directly semantically associated with both A and B.

[0134] Generate and output a set of two-factor four-tuple implicit knowledge links A<=>x[i]<=>y[j]<=>B, which directly show the potential relationship path between A and B connected by specific words or word combinations.

[0135] Based on each semantic direct association path x[i]<=>y[j], a two-factor implicit knowledge link A<=>x[i]<=>y[j]<=>B is constructed from A to x[i]; from x[i]<=>y[j] and then from y[j]<=>B. The above link directly displays the potential relationship path between A and B connected by specific words or word combinations.

[0136] For example, in the case, there is an implicit knowledge link of “immune response” – “immunotherapy” – “malignant tumor” – “glioma”.

[0137] The determination of the semantic direct association path set x[i]<=>y[j] needs to be adjusted according to the specific application scenario and requirements. For example, in some cases, it may only be necessary to find a single word as a direct association, while in other cases, more complex phrases or word groups may need to be considered. The generated binary implicit knowledge link set provides a new perspective for understanding and explaining the relationship between A and B, which helps to discover new knowledge points and insights.

[0138] Reference Figure 7 , the method further comprises the following steps:

[0139] From the initial semantic association set A{}, words that are semantically associated with A{} are obtained to form an intermediate semantic association set x{}; from x{}, words that are semantically associated with x{} are obtained to form an intermediate semantic association set z1{}.

[0140] From the target semantic association set B{}, words that are semantically associated with B{} are obtained to form an intermediate semantic association set y{}; from y{}, words that are semantically associated with y{} are obtained to form an intermediate semantic association set z2{}.

[0141] Exclude the direct connection with B{} and exclude the direct connection with y{} from z1{}; exclude the direct connection with A{} and exclude the direct connection with x{} from z2{};

[0142] A common improved word set z[k] is matched by z1{} and z2{}, where (k=1...n).

[0143] Thus, a three-factor five-tuple implicit knowledge link set A<=>x[i]<=>z[k]<=>y[j]<=>B is generated and output.

[0144] The generated three-factor five-tuple implicit knowledge link set provides a new relational view for understanding and explaining the complex relationship between A and B, and helps to reveal the deep and complex knowledge structure and its interactive association pattern hidden behind the text information.

[0145] The method further comprises the steps of:

[0146] From the initial semantic association set A{}, words that are semantically associated with A{} are obtained to form an intermediate semantic association set x{}; from x{}, words that are semantically associated with x{} are obtained to form an intermediate semantic association set s{}.

[0147] From the target semantic association set B{}, words that are semantically associated with B{} are obtained to form an intermediate semantic association set y{}; from y{}, words that are semantically associated with y{} are obtained to form an intermediate semantic association set t{}.

[0148] Exclude the direct connection with B{} and exclude the direct connection with y{} from s{}; exclude the direct connection with A{} and exclude the direct connection with x{} from t{};

[0149] Match s{} and t{} to obtain a common improved word path set s[k]<=>t[l], where (k=1…o, l=1…p), if and only if A<=>x[i]<=>s[l], and B<=>y[j]<=>t[l] path exists;

[0150] Match the association set s{} with the association set t{} to obtain the jointly improved four-factor tuple word path set ((x<=>s<=>t)<=>y):

[0151] Find those s[k]<=>t[l] paths in the association sets s{} and t{} (they appear as closely related pairs in the original text or knowledge base. They can also be called "bridge" element pairs (s, t) here) so that these element pairs can serve as key nodes connecting multiple paths between A and B. For example, traverse each element in s{} and check each "bridge" path of s[k]<=>t[l] in t{}. If and only if A<=>x[i]<=>s[k] path exists, and B<=>y[j]<=>t[l] path also exists;

[0152] Thus, a quadruple pair (x<=>s<=>t<=>y) is formed that represents a complex relationship.

[0153] Generate and output the four-factor six-tuple implicit knowledge link set A<=>x[i]<=>s[k]

[0154] <=>t[l]<=>y[j]<=>B. Among them (i=1...m, j=1...n, k=1...o, l=1...p).

[0155] Based on all constructed four-tuple pairs, a four-tuple implicit knowledge link is formed from A to x, through the bridge element pair (s, t), to y, and finally to B. The generated four-tuple implicit knowledge link set can provide a deeper and more comprehensive view of the relationship between A and B, helping to reveal the complex structure and interaction patterns hidden behind the data.

[0156] The embodiment of the present application further discloses a system for generating an implicit knowledge link, including a processor, in which a program of any of the above-mentioned methods for generating an implicit knowledge link is executed.

[0157] The embodiment of the present application also discloses a storage medium storing a program of any one of the above-mentioned methods for generating an implicit knowledge link.

[0158] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in the field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for generating a tacit knowledge link, characterized in that: The steps include: Acquire target text data; the target text data may be a single text document or multiple text documents; Parsing the concept words in the target text data to form a word set; and extracting the semantic association relationship between the concept words; Anchoring the current starting concept word A from the word set; Extract words that are semantically related to the current starting concept word A from the word set to form a starting semantic association set A{}, Anchoring the current target concept word B from the word set; Extracting words that are semantically related to the current target concept word B from the word set to form a target semantic association set B{}; Excluding the direct association with B from the semantic association set A{}; and excluding the current starting concept word A from the target semantic association set B{}; Based on the starting semantic association set A{} and the target semantic association set B{}, a common improved word set x[i] is matched, where (i=1…n), if and only if A<=>x[i] path and B<=>x[i] path exist; According to the concept word weight recommendation algorithm, a triple implicit knowledge link set A<=>x[i]<=>B of a single factor (x) is generated and output.

2. The method for generating a tacit knowledge link according to claim 1, characterized in that: The concept word weight recommendation algorithm includes the following steps: Get the word frequency (TF) corresponding to concept word A and concept word B, where word frequency (TF) = the number of times the current concept word (W) appears in the document set / the total number of words in the document set; Get the inverse document frequency (IDF) corresponding to concept word A and concept word B; inverse document frequency (IDF) = log (total number of documents in the current document set / number of documents containing the current concept word (W) + 1); Calculate the concept word weight, concept word weight Weight (W) = word frequency (TF) × inverse document frequency (IDF); Normalize the concept word weight Weight (W) and map it to the range of [0,1]: normalized_Weight(W)=(Weight(W)–min_Weight(W) / (max_Weight(W)-min_Weight(W); Among them, min_Weight(W) and max_Weight(W) are the minimum and maximum values ​​of Weight(W) of all words in the text respectively.

3. The method for generating a tacit knowledge link according to claim 1, characterized in that: The step of forming the initial semantic association set A{} further includes the following sub-steps: Calculating the semantic relevance between each word element in the word set and the current starting concept word A; The word element whose semantic association is greater than a preset first semantic reference value is regarded as a word having a semantic association with the current starting concept word A; The first semantic reference value is adjusted in a positive correlation according to the proportion of the weight of the current starting concept word A in the target text data.

4. The method for generating a tacit knowledge link according to claim 3, characterized in that: The step of forming the target semantic association set B{} further includes the following sub-steps: Calculate the semantic relevance between each word element in the word set and the current target concept word B; The word element whose semantic association is greater than a preset second semantic reference value is regarded as a word having a semantic association with the current target concept word B; The second semantic reference value is adjusted according to the anti-correlation between the weight ratio of the current target concept word B and the weight ratio of the current starting concept word A.

5. The method for generating a tacit knowledge link according to claim 4, characterized in that: The step of matching the common improved word set x[i] also includes the following sub-steps: Calculate the first semantic association degree a of each word element in the starting semantic association set A{} and the current starting concept word A; Calculate the second semantic association degree b between each word element in the target semantic association set B{} and the current target concept word B; A bidirectional correlation c is calculated according to the first semantic correlation a and the second semantic correlation b, and the calculation formula is: c=a-first semantic reference value+b-second semantic reference value; Then the bidirectional association degree c is compared with a preset bidirectional reference value, and the set of word elements whose bidirectional association degree c is greater than the bidirectional reference value is the word set x[i].

6. The method for generating a tacit knowledge link according to claim 5, characterized in that: The method further comprises the following sub-steps: According to the ratio between the number of elements in the starting semantic association set A{} and the number of elements in the target semantic association set B{}, the first semantic reference value is adjusted in a positive correlation, and the second semantic reference value is adjusted in an anti-correlation manner; Alternatively, the first semantic reference value is adjusted in a positive correlation and the second semantic reference value is adjusted in an anti-correlation according to the ratio of the average semantic association degree of the elements in the starting semantic association set A{} to the average semantic association degree of the elements in the target semantic association set B{}; The bidirectional reference value is adjusted in positive correlation according to the absolute value of the difference between the first semantic reference value and the second semantic reference value.

7. The method for generating a tacit knowledge link according to claim 1, characterized in that: The method further comprises the steps of: Acquire words that are semantically associated with the starting semantic association set A{} to form an intermediate semantic association set x{}, and acquire words that are semantically associated with the target semantic association set B{} to form an intermediate semantic association set y{}; exclude the direct association with B{} from x{}; exclude the direct association with A{} from y{}; Match the common improved word path set x[i]<=>y[j], where (i=1…m, j=1…n), if and only if A<=>x[i] and B<=>y[j] path exists; Generate and output a two-factor four-tuple implicit knowledge link set A<=>x[i]<=>y[j]<=>B.

8. The method for generating a tacit knowledge link according to claim 7, characterized in that: The method further comprises the steps of: Obtaining words that are semantically associated with the starting semantic association set A{} from A{} to form an intermediate semantic association set x{}; obtaining words that are semantically associated with x{} from x{} to form an intermediate semantic association set z1{}; Acquire words that are semantically associated with the target semantic association set B{} to form an intermediate semantic association set y{}; acquire words that are semantically associated with y{} from y{} to form an intermediate semantic association set z2{}; Exclude the direct connection with B{} and exclude the direct connection with y{} from z1{}; exclude the direct connection with A{} and exclude the direct connection with x{} from z2{}; Match z1{} and z2{} to obtain a common improved word set z[k], where (k=1…o), if and only if A<=>x[i], x[i]<=>z[k], and B<=>y[j], y[j]<=>z[k] exists; Thus, a three-factor five-tuple implicit knowledge link set A<=>x[i]<=>z[k]<=>y[j]<=>B is generated and output.

9. A system for generating implicit knowledge links, characterized in that: It comprises a processor, in which a program of the method for generating a tacit knowledge link as claimed in any one of claims 1 to 8 is run.

10. A storage medium, characterized in that: A program storing the method for generating a tacit knowledge link as claimed in any one of claims 1 to 8.

Citation Information

Patent Citations

  • A multi-relation fusion method and an intelligent system for tacit related knowledge discovery

    CN108959540A

  • Cross-language implicit association knowledge discovery method and device, equipment and storage medium

    CN116595124A

  • Scientific knowledge discovery method and system based on knowledge graph

    CN117786122A