A terminology management service system and method based on standard archives

Abstract

The present invention discloses a terminology management service system and method based on standard archives, and relates to the technical field of archive management. The terminology management service method based on standard archives constructs an initial association network between known terms; updates the initial association network between terms based on the outputted relationship strength of known relationships, potential relationships and the relationship strength of potential relationships to obtain a main association network; updates the main association network based on the relationship between new and old terms and the relationship strength of the relationship between new and old terms to obtain an updated association network; aggregates and recommends query terms based on the updated association network, and outputs a term recommendation list. By comprehensively using a term relationship recognition model and a recommendation system, the speed and accuracy of extracting and retrieving information from a large number of documents are significantly improved. Combining multi-dimensional factors such as the relationship strength of terms, user historical behavior and term popularity, the recommendation system can provide more accurate and personalized content.

Description

Technical Field

[0001] The present invention relates to the technical field of archive management, and in particular to a terminology management service system and method based on standard archives. Background Art

[0002] At present, in the process of digitalization of archives, it is necessary to establish a perfect index library to provide retrieval for the future electronic use of archives and improve the efficiency of the use of electronic archives, which is often accompanied by a large amount of archive recording work. The existing technology generally adopts two methods: one is to record in Excel form, and the other is to use archive digital processing software to let employees record; from these two working methods, it still requires a large number of staff and the cost is very high.

[0003] In addition to the above-mentioned digital storage of archives, a large number of staff are also required to conduct quality inspections on the digitally scanned images during subsequent archive scanning, including resolution, page number continuity, content integrity, cropped content integrity, etc. However, the existing archive management methods have slow retrieval speeds and low retrieval accuracy. Summary of the invention

[0004] In view of the deficiencies of the prior art, the present invention provides a terminology management service system and method based on standard archives, which solves the problems of slow retrieval speed and low retrieval accuracy of the existing archive management methods.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a terminology management service method based on standard archives, comprising the following steps: extracting text from standard documents to obtain known terms, obtaining known term attributes, performing association analysis on known terms, and constructing an initial association network between known terms; training and learning the initial association network between known terms based on a term relationship recognition model, outputting the relationship strength of known relationships between known terms, the relationship strength of potential relationships and potential relationships, updating the initial association network between terms based on the output relationship strength of known relationships, the relationship strength of potential relationships and potential relationships, and obtaining a main association network; Other new terms are compared with known terms in the main association network. If the new terms are different from all the recorded known terms in the main association network, the new terms and the analyzed new term attributes are fused with the main association network to form a fused association network; the fused association network is input into the term relationship recognition model to predict and output the new and old term relationships and the relationship strength between the new terms and the known terms, and the main association network is updated based on the new and old term relationships and the relationship strength between the new and old terms to obtain an updated association network; after obtaining the query terms input by the user, the query terms are aggregated and recommended based on the updated association network, and a term recommendation list is output.

[0006] Furthermore, the text extraction of the standard document to obtain known terms, obtain the known term attributes, perform association analysis on the known terms, and construct an initial association network between the known terms, includes: extracting original text data from the electronic document of the standard document, and performing text cleaning and segmentation on the extracted original text data; performing term recognition on the original text data after text cleaning and segmentation, marking the known terms, and extracting the known term attributes, wherein the known term attributes include the known term definitions, known term examples, and the known term usage frequency; storing the acquired known terms and known term attributes in a term library; identifying and mapping the known relationships between known subjects based on statistical methods and speech analysis methods, wherein the known relationships include synonym relationships and hyponymy relationships; integrating and storing the identified known relationships, known terms, and known term attributes in the term library to construct an initial association network.

[0007] Furthermore, the forming of the initial association network includes: defining known terms as nodes, locating known relationships between known subjects as edges, and converting known term attributes into known term feature vectors of nodes.

[0008] Furthermore, the term relationship recognition model is based on which the initial association network between known terms is trained and learned, and the relationship strength of known relationships between known terms, potential relationships and the relationship strength of potential relationships are output, including: the term relationship recognition model takes the initial association network as input, learns the features of the known terms and the known relationships between the known terms according to the feature vectors converted from the known term attributes of the nodes, and outputs the potential relationships between the known terms based on the inference of the known relationships between the known terms; obtains the performance evaluation parameters output after the term relationship recognition model is trained, and obtains a performance evaluation index based on the performance evaluation parameters, wherein the performance evaluation index is used to represent the training result of the term relationship recognition model, and if the performance evaluation index is greater than the set evaluation threshold, the training is completed, and if the performance evaluation index is less than or equal to the set evaluation threshold, the model parameters of the term relationship recognition model are optimized and retrained; the trained term relationship recognition model determines the relationship strength of the known relationships and the relationship strength of the potential relationships based on the relationship determination parameters between the known terms.

[0009] Furthermore, the performance evaluation parameters include the prediction accuracy, recall rate and F1 score of each term relationship. The calculation formula for obtaining the performance evaluation index based on the performance evaluation parameters is as follows:

[0010]

[0011] In the formula, OS is the performance evaluation index, i is the number of the term relationship, i = 1, 2, 3, ..., n, n is the total number of term relationships, w i is the weight factor of the i-th term relationship, Ac iis the prediction accuracy of the i-th term relationship, α is Ac i The weight coefficient, Re i is the recall rate of the i-th term relationship, β is Re i The weight coefficient, F1 i is the F1 score of the i-th term relationship, and γ is F1 i The weight coefficient of .

[0012] Furthermore, the new terms and the new term attributes obtained by analysis are fused with the main association network to form a fused association network, including: extracting features from the new terms to obtain new term attributes, wherein the new term attributes include new term definitions, new term examples and new term usage frequencies; performing feature transformation on the new term attributes to obtain new term feature vectors; merging the new term feature vectors with feature vectors of known terms in the main association network; performing contextual data analysis on the new terms to initially determine the initial term relationship between the new terms and the known terms in the main association network based on the co-occurrence frequency between the new terms and the known terms in the main association network; creating nodes for the new terms in the main association network, and creating edges between the new terms and the known terms in the main association network based on the initial term relationship between the new terms and the known terms in the main association network.

[0013] Furthermore, the prediction outputs the new and old term relationships and the relationship strength of the new and old term relationships between the new terms and the known terms, including: obtaining the new term feature vector and the initial term relationship obtained by analyzing the context data to obtain the new and old term relationship; and determining the parameter analysis based on the new and old term relationship to obtain the relationship strength of the new and old term relationship.

[0014] Furthermore, the method of performing aggregated recommendation on query terms based on the updated associated network and outputting a term recommendation list includes: obtaining relationship parameters between the query terms and known terms in the updated associated network; determining a recommendation index based on the relationship parameters; arranging the obtained recommendation indexes in descending order, and loading the known terms corresponding to the first G recommendation indexes into the term recommendation list.

[0015] Furthermore, the calculation formula of the recommendation index is:

[0016] S(T j ,K)=δ*R(T j ,K)+ε*Hs(T j ,U)+θ*P(T j );

[0017] Where j is the number of the known terms in the updated association network, j = 1, 2, 3, ..., m, m is the total number of known terms in the updated association network, K is the query term, T j is the jth known term in the updated association network, S(T j,K) represents the recommendation index of the query term and the jth known term in the updated association network, R(T j ,K) represents the relationship strength between the jth known term and the query term K, and δ is R(T j ,K) weight factor, Hs(T j ,U) represents the frequency of occurrence of the jth known term in the historical behavior of user U, and ε is Hs(T j ,U) weight factor, P(T j ) represents the popularity of the jth known term, θ is P(T j ) is the weight factor of .

[0018] A terminology management service system based on standard archives, used for the above-mentioned terminology management service method based on standard archives, includes an initial association network construction subsystem, a main association network construction subsystem, a fusion association network formation subsystem, an updated association network formation subsystem and an aggregate recommendation subsystem, wherein: the initial association network construction subsystem is used to perform text extraction on standard documents to obtain known terms, obtain known term attributes, perform association analysis on known terms, and construct an initial association network between known terms; the main association network construction subsystem is used to train and learn the initial association network between known terms based on a term relationship recognition model, output the relationship strength of known relationships between known terms, the relationship strength of potential relationships and potential relationships, and based on the output known relationship strength, the relationship strength of potential relationships and potential relationships, train the initial association network between terms. The initial association network is updated to obtain the main association network; the fused association network forming subsystem is used to compare the identified new terms with the known terms in the main association network. If the new term is different from all the known terms recorded in the main association network, the new term and the analyzed new term attributes are fused with the main association network to form a fused association network; the updated association network forming subsystem is used to input the fused association network into the term relationship recognition model, predict and output the new term and the old term relationship and the relationship strength between the new term and the known term, and update the main association network based on the new term relationship and the relationship strength between the new term and the old term to obtain the updated association network; the aggregate recommendation subsystem is used to aggregate and recommend the query terms based on the updated association network after obtaining the query terms input by the user, and output a term recommendation list.

[0019] The present invention has the following beneficial effects:

[0020] This terminology management service method based on standard archives significantly improves the speed and accuracy of extracting and retrieving information from large amounts of documents by combining term relationship recognition models and recommendation systems. Combining multi-dimensional factors such as term relationship strength, user historical behavior, and term popularity, the recommendation system can provide more accurate and personalized content.

[0021] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The figure is a flow chart of the terminology management service method based on standard archives of the present invention.

[0023] Figure 2 An index image is recommended for the terminology management service method based on standard archives of the present invention.

[0024] Figure 3 This is a flowchart of the terminology management service system based on standard archives of the present invention. DETAILED DESCRIPTION

[0025] The embodiment of the present application provides a comprehensive, dynamic and efficient solution for managing and utilizing large-scale terminology data through a terminology management service method and system based on standard archives.

[0026] The overall idea of ​​the problem in the embodiment of this application is as follows:

[0027] First, by automating text extraction and attribute acquisition, the efficiency and accuracy of quickly identifying and classifying terms from a large number of standard documents are improved. Secondly, using a term relationship recognition model to analyze and learn the relationships between terms can not only deepen the understanding of known relationships, but also discover potential new relationships and their strengths, thereby enriching and updating the association network between terms. In addition, by fusing new and old term data and continuously updating the association network, this method can continuously adapt to the changes and expansion of terms and keep the database current and relevant. Finally, based on this dynamically updated association network, aggregated recommendations can more accurately meet users' query needs for related terms, improving user experience and information retrieval. Overall, this method provides a comprehensive, dynamic and efficient solution for managing and utilizing large-scale term data, especially for knowledge-intensive fields.

[0028] See also Figure 1,The embodiment of the present invention provides a technical solution: a terminology management service method based on standard archives, including the following steps: clarifying from which types of standard archives data need to be extracted, such as technical standards, industry specifications, etc., using text extraction tools to standard documents to obtain known terms, obtaining the attributes of the known terms, performing association analysis on the known terms, and constructing an initial association network among the known terms; training and learning the initial association network among the known terms based on a term relationship recognition model, outputting the relationship strength of the known terms, the relationship strength of the potential relationships, and the relationship strength of the potential relationships among the known terms; The network is updated to obtain the main association network; the identified new terms are compared with the known terms in the main association network. If the new terms are different from all the recorded known terms in the main association network, the new terms and the analyzed new term attributes are fused with the main association network to form a fused association network; the fused association network is input into the term relationship recognition model to predict and output the new and old term relationships and the relationship strength between the new terms and the known terms, and the main association network is updated based on the new and old term relationships and the relationship strength between the new and old terms to obtain an updated association network; after obtaining the query terms input by the user, the query terms are aggregated and recommended based on the updated association network, and a term recommendation list is output.

[0029] Automated text extraction and term recognition significantly improve the speed and efficiency of extracting information from large amounts of documents, reducing the need for manual operations. Through model training, the relationship between terms is continuously learned and verified, which improves the accuracy and reliability of the data. The model can identify subtle relationship changes and new potential relationships, enhancing the richness of the data.

[0030] As new terms are constantly identified and integrated, the network is dynamically updated, able to adapt to changes in terminology and knowledge, and maintain the timeliness and relevance of the database. Based on complex network analysis, the aggregate recommendation system provides accurate terminology recommendations, improves user query experience, and improves the efficiency and effectiveness of information retrieval. The updated association network provides users with a structured knowledge system that facilitates knowledge sharing, learning, and management, especially for professional fields and academic research.

[0031] Specifically, text extraction is performed on the standard document to obtain known terms, obtain the properties of the known terms, perform association analysis on the known terms, and construct an initial association network between the known terms, including: extracting raw text data from the electronic document of the standard document, cleaning and segmenting the extracted raw text data, text cleaning includes removing useless symbols, correcting spelling errors, and unifying the expression of terms (such as abbreviation expansion), text segmentation: segmenting the cleaned text according to logic and structure, such as chapters, clauses, etc., to facilitate subsequent more accurate term extraction; performing term recognition on the raw text data after text cleaning and segmentation, applying the named entity recognition (NER) algorithm in natural language processing technology to perform term recognition on the preprocessed text, and marking the professional terms and their context in the text. The key attributes of the identified known terms are further extracted, such as definition, related examples, frequency of use and other information. The known term attributes include known term definition, known term examples and known term frequency of use; the acquired known terms and known term attributes are stored in the term base; based on statistical methods and speech analysis methods, the known relationships between known subjects are identified and mapped. The known relationships include synonym relationships, hyponymy relationships and lower-upper-down relationships; the identified known relationships, known terms and known term attributes are integrated and stored in the term base to form an initial association network.

[0032] The initial association network is constructed, including: defining known terms as nodes, locating known relationships between known subjects as edges, and converting known term attributes into known term feature vectors of nodes.

[0033] In this implementation scheme, known terms are obtained from standard documents through automated text extraction tools, which not only improves the speed of extracting information from a large number of documents, but also greatly improves the efficiency of information processing. The accuracy of term identification is further enhanced by using natural language processing technology (such as named entity recognition algorithm). The initial association network constructed based on known terms and their attributes is dynamically updated as new terms are continuously identified and integrated. This continuous updating not only adapts to changes in terminology and knowledge, but also maintains the timeliness and relevance of the database, helps to accurately map the relationship between terms (such as synonym relationships, hyponymy relationships, etc.), and enhances the richness and structured nature of the data.

[0034] Through the term relationship recognition model, the relationship between terms is continuously learned and verified to improve the accuracy and reliability of data. The model can identify subtle relationship changes and new potential relationships. In addition, the aggregate recommendation system provides accurate term recommendations based on complex network analysis, improves user query experience, and improves the efficiency and effectiveness of information retrieval. The updated association network provides users with a structured knowledge system that facilitates knowledge sharing, learning, and management. This structured information is very suitable for professional fields and academic research, helping users quickly locate information and understand complex term relationships.

[0035] Specifically, the initial association network between known terms is trained and learned based on the term relationship recognition model, and the relationship strength of the known relationship between the known terms, the potential relationship and the relationship strength of the potential relationship are output, including: the term relationship recognition model takes the initial association network as input, learns the features of the known terms and the known relationships between the known terms according to the feature vectors converted from the known term attributes of the nodes, and outputs the potential relationships between the known terms based on the known relationships between the known terms; obtains the performance evaluation parameters output after the term relationship recognition model is trained, and obtains the performance evaluation index based on the performance evaluation parameters. The performance evaluation index is used to represent the training result of the term relationship recognition model. If the performance evaluation index is greater than the set evaluation threshold, the training is completed. If the performance evaluation index is less than or equal to the set evaluation threshold, the model parameters of the term relationship recognition model are optimized and then retrained; the trained term relationship recognition model determines the relationship strength of the known relationship and the relationship strength of the potential relationship based on the relationship determination parameters between the known terms.

[0036] The term relationship recognition model is a graph neural network model or a relationship extraction model. It accepts the initial association network as input and uses the attributes of known terms (such as definition, frequency of use, etc.) to convert them into feature vectors. These feature vectors help the model learn and understand the uniqueness of each term and the relationship between them. By converting term attributes into feature vectors, the model can process language elements in a mathematical form, thereby more accurately identifying and predicting complex relationships between terms, such as synonyms, hyponyms, and hyponyms.

[0037] The effect of the model is evaluated through performance evaluation indices such as precision, recall, and F1 score. If the performance evaluation index does not reach the preset threshold, the model parameters are optimized and retrained. This iterative training and evaluation process ensures the reliability and accuracy of the model in practical applications. Through continuous optimization, the model can better adapt to the changes and complexity of the data and provide more accurate term relationship recognition.

[0038] The evaluation threshold is usually set based on the results of experiments and previous studies. For example, different thresholds are tested on the model in the early stage of training through the K-Fold Cross-Validation method, and the threshold that makes the model perform best on the validation set is selected as the set evaluation threshold.

[0039] The formula for calculating the relationship strength of a known relationship or potential relationship is as follows:

[0040]

[0041] In the formula, A and B represent two known terms with a known relationship, a is the number of the known relationship or potential relationship between the known terms A and B, a = 1, 2, 3, ..., Y, Y is the total number of known relationships or potential relationships between the known terms A and B, is the strength of the ath known relationship or potential relationship between known terms A and B, C AB is the number of times the known terms A and B co-occur in the same document. The number of times the two terms appear in the same document is counted using text analysis tools. The number of co-occurrences is a basic indicator for measuring the relevance of two terms. Frequent co-occurrences usually mean high relevance. 1 C preset in the database AB The weight factor, S AB To score the semantic similarity of the text around known terms A and B, use natural language processing tools, such as word vector models (Word2Vec, BERT, etc.), to calculate the semantic similarity of the text context of the two terms. Semantic similarity can capture the potential connection between terms, even if they do not often appear in the same document. 2 The preset S in the database AB The weight factor of The relationship weights between terms are given by domain experts preset in the database based on the semantics and practical applications of known terms A and B. The relationship weights between terms are obtained through expert surveys or domain knowledge feedback. Expert insights can provide details in term relationships that are difficult to capture through automated means, especially in new or not widely documented knowledge fields. 3 Preset in the database The weight factor, F AB Given the rarity of terms A and B, we calculate the inverse of the ratio of the number of documents containing the two terms to the total number of documents. A rarer term combination may represent specific expertise or an emerging research field, and therefore has a higher information value, defined as N is the total number of documents, df AB is the number of documents containing known terms A and B, The adjustment factor for the ath known relationship or potential relationship between known terms A and B preset in the database can be obtained by statistical analysis, analyzing factors such as the context frequency, position (such as title, abstract, text), time (the publication date of the document relative to the current time) of the terms, and defining the adjustment factor based on these statistical results.

[0042] For each weight factor corresponding to each known relationship or potential relationship, the relationship weight and adjustment factor can be directly obtained from the database, and the weight factors stored in the database are set by analyzing the influence of different factors on the relationship strength. For example, if the data shows that the correlation between the number of co-occurrences and the relationship strength is higher than the semantic similarity, ω 1 will be set higher.

[0043] In order to manage and call the various weight factors, relationship weights and adjustment factors corresponding to each known relationship or potential relationship, a database is designed to store the various weight factors, relationship weights and adjustment factors corresponding to each known relationship or potential relationship. Through the dictionary structure in Python, a mapping set is created to store the known relationships or potential relationships and the various weight factors, relationship weights and adjustment factors corresponding to the known relationships or potential relationships. An interface is provided for real-time updating and calling of these data, so that when calculating the relationship strength of known relationships or potential relationships, the various weight factors, relationship weights and adjustment factors corresponding to the known relationships or potential relationships can be extracted from the database as needed, thereby improving the accuracy and operability of the evaluation.

[0044] These parameters work together in a comprehensive model, complementing each other to provide a comprehensive assessment of the strength of a relationship. For example, co-occurrence counts provide intuitive evidence of relevance, semantic similarity adds insight into the underlying meaning of the text, and expert evaluation adds depth of expertise. Rarity and adjustment factors further refine the model, ensuring that reasonable relationship assessments are obtained even in special cases.

[0045] The performance evaluation parameters include the prediction accuracy, recall rate, and F1 score of each term relationship. The calculation formula for the performance evaluation index based on the performance evaluation parameters is as follows:

[0046]

[0047] In the formula, OS is the performance evaluation index, i is the number of the term relationship, i = 1, 2, 3, ..., n, n is the total number of term relationships, w i is the weight factor of the i-th term relationship preset in the database, Ac i is the prediction accuracy of the i-th term relationship, the number of correctly predicted term relationships divided by the total number of predicted relationships, and α is the Ac preset in the database i The weight coefficient, Re i is the recall rate of the ith term relationship, which is the number of correctly predicted term relationships divided by the total number of actual term relationships. β is the Re i The weight coefficient, F1 iis the F1 score of the i-th term relationship, the harmonic mean of precision and recall, and is a balance indicator between the two. γ is the preset F1 score in the database. i The weight coefficient of .

[0048] For each weight coefficient a, β and γ and the weight factor w i , can be obtained directly from the database, and the weight coefficients in the database can be determined by an optimization algorithm, such as cross-validation in the training phase to find the best weight configuration to maximize the overall performance indicator. The weight factor can be set based on the importance or frequency of each term relationship. For example, some term relationships may be more critical in a specific field, or some relationships are more common, so they can be given higher weights.

[0049] In order to manage and call the corresponding weight coefficients and weight factors w when calculating performance evaluation parameters i , design a database to store the corresponding weight coefficients and weight factors w when calculating performance evaluation parameters i Through the dictionary structure in Python, create a mapping set to store the corresponding weight coefficients and weight factors w when calculating the performance evaluation parameters i , providing an interface for real-time updating and calling of these data, so that when calculating the performance evaluation index, the corresponding weight coefficients and weight factors w of the performance evaluation parameters can be extracted from the database as needed. i , thereby improving the accuracy and operability of the evaluation. The calculation formula of this performance evaluation index is used to comprehensively consider multiple performance indicators to obtain an overall performance evaluation result, making the evaluation of the model more comprehensive and better reflecting the effect of the model on the prediction of different term relationships. The formula weights the accuracy, recall and F1 scores according to their respective importance. In this way, the prediction performance of the model on different term relationships can be comprehensively evaluated. The contribution of each performance indicator to the total evaluation index is determined by the corresponding weight coefficient.

[0050] Instead of relying on any one indicator alone, multiple indicators are integrated to more comprehensively evaluate the performance of the model. By adjusting the weight coefficient, the focus of performance evaluation can be customized according to different application requirements and evaluation criteria.

[0051] Taking synonymous relationship, hyponymous relationship and part-whole relationship as examples, the performance evaluation index data are as follows (α=β=γ=0.3, ):

[0052] Table 1 Data of performance evaluation index

[0053] Iterations Term Relationship Prediction accuracy Recall F1 score Performance Evaluation Index 1 Synonym relationship 70 65 67.5 67.8 1 Synonym relationship 60 70 64.7 65.2 1 Synonym relationship 50 55 52.5 53.0 2 Hierarchical relationship 75 70 72.5 73.0 2 Hierarchical relationship 65 75 69.8 70.5 2 Hierarchical relationship 55 60 57.5 58.2 3 Part-whole relationship 80 75 77.5 78.2 3 Part-whole relationship 70 80 74.9 75.8 3 Part-whole relationship 60 65 62.5 63.5

[0054] Iteration number: indicates the iteration stage of model training. The model will be adjusted and optimized after each iteration. Relation type: specifies the type of term relationship to be evaluated. Precision, recall, F1 score: These indicators measure the accuracy, coverage, and balance indicators of the model in predicting each relationship. Precision is the proportion of predicted relationships that are actually the relationship; recall is the proportion of predicted relationships that are actually the relationship; F1 score is the harmonic mean of precision and recall. The table shows the trend that the performance indicators of each relationship type gradually improve with the increase of iteration number. This shows that as the model is trained and optimized, it can more accurately predict different types of term relationships. The comprehensive score gradually improves with the increase of iteration number, which reflects the improvement of the overall performance of the model. The improvement in score means that the model has become more accurate and reliable in predicting different relationship types.

[0055] Specifically, the new terms and the new term attributes obtained by analysis are fused with the main association network to form a fused association network, including: extracting features of the new terms to obtain new term attributes, which include new term definitions, new term examples and new term usage frequencies; performing feature transformation on the new term attributes to obtain new term feature vectors; merging the new term feature vectors with feature vectors of known terms in the main association network; performing contextual data analysis on the new terms, and initially determining the initial term relationship between the new terms and the known terms in the main association network based on the co-occurrence frequency between the new terms and the known terms in the main association network; creating nodes for the new terms in the main association network, and creating edges between the new terms and the known terms in the main association network based on the initial term relationship between the new terms and the known terms in the main association network.

[0056] The prediction output is the new and old term relationship between the new term and the known term and the relationship strength of the new and old term relationship, including: obtaining the new term feature vector and the initial term relationship obtained by context data analysis to obtain the new and old term relationship; based on the new and old term relationship, determining the parameter analysis to obtain the relationship strength of the new and old term relationship.

[0057] The properties of the new term (such as definition, example sentences, and frequency of use) are first extracted. These properties are then converted into feature vectors through a feature transformation process, which allows the term's information to be quantified and used in the computer model. The feature vector of the new term is merged with the feature vectors of known terms in the main association network. This involves creating nodes for the new term in the network and determining the initial term relationships based on its co-occurrence frequency with known terms and contextual data analysis, thereby creating corresponding edges. The feature vector of the new term and the initial relationship data are used to predict the detailed relationship between the new and old terms. Based on these relationships, the relationship strength is further analyzed and calculated, which usually involves the optimization and adjustment of model parameters.

[0058] The formula for calculating the relationship strength of the new and old term relationships is as follows:

[0059]

[0060] In the formula, C represents the new term, b is the number of the new and old term relationship between the known term A and the new term C, b = 1, 2, 3, ..., X, X is the total number of the new and old term relationships between the known term A and the new term C, is the relationship strength of the bth new-old term relationship between the known term A and the new term C, T AC is the term text similarity score between the known term A and the new term C, calculated by natural language processing techniques such as TF-IDF weighted cosine similarity, σ 1 T preset in the database AC The weight factor, V AC The importance of known term A and new term C in their fields is calculated by the frequency of occurrence and citation in the document, and the mean of the frequency of occurrence and citation is directly used as V AC , σ 2 V preset in the database AC The weight factor, L AC is the link strength between the known term A and the new term C, which is evaluated by the number of common neighbors, that is, the number of neighbors is directly used as the link strength between the known term A and the new term C, σ 3 L is preset in the database AC The weight factor, N AC is the global network importance factor of the known term A and the new term C, reflecting the centrality and influence of the two terms in the entire term network, which can be calculated using PageRank or other network centrality indicators.

[0061] Each weight factor corresponding to each new and old term relationship can be directly obtained from the database, and each weight factor corresponding to each new and old term relationship stored in the database is set by analyzing the influence of different factors on the relationship strength of the new and old term relationship. For example, if the data shows that T AC Higher correlation with relationship strength than V AC Correlation with tie strength, σ 1 will be set higher.

[0062] In order to manage and call the various weight factors corresponding to each new and old term relationship, a database is designed to store the various weight factors corresponding to each new and old term relationship. Through the dictionary structure in Python, a mapping set is created to store the various weight factors corresponding to each new and old term relationship. An interface is provided for real-time updating and calling of these data, so that when calculating the relationship strength of the new and old term relationship, the various weight factors, relationship weights and adjustment factors corresponding to the new and old term relationship can be extracted from the database as needed, thereby improving the accuracy and operability of the evaluation.

[0063] In this implementation, the calculation formula combines multiple factors such as text similarity, term importance, link strength and network global importance to comprehensively evaluate the strength of the relationship between known terms and new terms. The multi-dimensional evaluation method can reflect the correlation and interaction between terms from different angles.

[0064] Taking into account the position and global importance of terms in the network, this approach can dynamically adapt to changes in the term network and reflect the changing interrelationships between new and old terms as they develop over time.

[0065] These parameters complement each other and work together in the evaluation of relationship strength. For example, high text similarity may mean that two terms are closely related in content; at the same time, if the two terms also have high link strength and global importance, the strength of the relationship between them may be further strengthened. In addition, the importance of a term may also strengthen the strength of its relationship with other important terms, especially when they co-appear in many high-impact documents. This comprehensive evaluation method ensures accurate identification of term relationships and is very suitable for complex and dynamically changing knowledge network environments.

[0066] Specifically, query terms are aggregated and recommended based on the updated associated network, and a term recommendation list is output, including: obtaining relationship parameters between the query terms and known terms in the updated associated network; (the relationship strength can be obtained through synonym relationships, up-down relationships, etc. If the query terms do not match any known terms in the updated associated network, there is no term in the recommendation list) determining a recommendation index based on the relationship parameters; arranging the obtained recommendation indexes in descending order, and loading the known terms corresponding to the first G recommendation indexes into the term recommendation list.

[0067] First, the relationship parameters of the known terms related to the query terms are retrieved from the updated association network. These relationship parameters include synonym relationships, hyponymy relationships, etc. These parameters can be directly used to calculate the relationship strength. According to the relationship parameters between the query terms and the known terms, a recommendation index is calculated. This index reflects the strength of the relevance of each known term to the query term. The recommendation index is arranged in order from high to low. Then, the top G terms with the highest relevance are selected and loaded into the term recommendation list.

[0068] Through the aggregate recommendation system, users can quickly find the terms most relevant to the query terms, which greatly improves the efficiency of information retrieval. The recommendation method based on relationship strength and network analysis makes the term recommendation system not just a simple search tool, but an advanced analysis tool that can provide in-depth insights and help understand complex knowledge structures.

[0069] The formula for calculating the recommendation index is:

[0070] S(T j ,K)=δ*R(T j ,K)+ε*Hs(T j ,U)+θ*P(T j );

[0071] Where j is the number of the known terms in the updated association network, j = 1, 2, 3, ..., m, m is the total number of known terms in the updated association network, K is the query term, T j is the jth known term in the updated association network, S(T j ,K) represents the recommendation index of the query term and the jth known term in the updated association network, R(T j ,K) represents the relationship strength between the jth known term and the query term K. By analyzing the relationship between the query term K and the known term T j The text similarity between them can be obtained by Jaccard similarity, which is based on the intersection and union of the words in the description of the two terms. δ is the R(T j ,K) weight factor, Hs(T j ,U) represents the frequency of occurrence of the jth known term in the historical behavior of user U, which is calculated through the user's historical query records or operation records related to the term, that is, the data of a specific time period (such as the most recent year) is selected for analysis, and the ratio of the number of occurrences of the jth known term in the historical behavior of user U to the time is counted. ε is the Hs(T j ,U) weight factor, P(T j) represents the popularity of the jth known term, which can be evaluated by the number of citations or frequency of occurrence of the term in all documents or databases. The number of citations or frequency of occurrence of the jth known term is divided by the total number of documents or total number of citations to obtain a ratio, which is used as the popularity. θ is the P(T j ) is the weight factor of .

[0072] Each weight factor corresponding to the recommendation index can be directly obtained from the database, and the weight factors corresponding to the recommendation index stored in the database are used to learn the optimal weight configuration through a machine learning model. For example, an optimization goal can be set (such as maximizing the click-through rate or satisfaction of the recommendation system), and then an algorithm (such as gradient descent) can be used to adjust the weights to optimize the objective function value.

[0073] In order to manage and call the weight factors corresponding to the recommendation index, a database is designed to store the weight factors, and the weight values ​​are determined by analyzing historical data in combination with the machine learning model. Through the dictionary structure in Python, a mapping set is created to store the weight factors corresponding to the recommendation index, and an interface is provided for real-time updating and calling of these data, so that the weight factors corresponding to the recommendation index can be extracted from the database as needed when calculating the recommendation index, thereby improving the accuracy and operability of the evaluation.

[0074] The formula is designed to calculate the recommendation index by combining three key dimensions: the strength of the relationship between terms, the frequency of term appearance in the user's historical behavior, and the popularity of the terms. This approach ensures that recommendations are not only based on the objective association between terms, but also takes into account the user's personal historical preferences and the overall importance of the terms. By considering the frequency of term use in the user's historical behavior, terms that are more in line with personal preferences are recommended, which improves user satisfaction and the applicability of recommendations. Combining the popularity of terms and the specific relationship strength helps balance the recommendation results and avoid only recommending terms that are too popular or too unpopular.

[0075] These parameters are selected to ensure that the recommendation system does not rely solely on content similarity, but provides a personalized and balanced recommendation that integrates user historical behavior and term popularity. This approach improves the relevance, personalization, and usefulness of recommendations, allowing users to get more satisfying search results and exploration experience.

[0076] The table of recommendation indexes of query terms and known terms in the updated association network is as follows:

[0077] Table 2 Data table of the recommendation index of query terms and known terms in the updated association network

[0078] Term Number <![CDATA[S(T j ,K)]]> <![CDATA[R(T j ,K)]]> <![CDATA[Hs(T j ,U)]]> <![CDATA[S(T j ,K)]]> 1 94 87 86 89.5 6 66 78 82 74.4 5 95 62 54 72.8 2 52 90 72 69.4 4 74 65 62 67.7 7 50 57 72 58.7 3 55 50 66 56.8

[0079] Set G to 5. From the data in the table, it can be seen that the known terms with term numbers 1, 6, 5 and 3 have the largest recommendation index with respect to the query terms. When G is 5, the known terms corresponding to term numbers 1, 6, 5 and 3 are loaded into the term recommendation list. When the user enters a query term, the known terms in the term recommendation list are automatically pushed.

[0080] Figure 2 It is a bar chart of the recommendation index corresponding to each term number. The top terms are 1, 6, 5 and 3, and the two at the bottom (term numbers 2 and 4) are represented by different slashes, which clearly and intuitively shows the recommendation index of known terms with different numbers.

[0081] A terminology management service system based on standard archives, used for the above-mentioned terminology management service method based on standard archives, such as Figure 3 As shown, it includes an initial association network construction subsystem, a main association network construction subsystem, a fusion association network formation subsystem, an updated association network formation subsystem and an aggregate recommendation subsystem, wherein: the initial association network construction subsystem is used to perform text extraction on standard documents to obtain known terms, obtain known term attributes, perform association analysis on known terms, and construct an initial association network between known terms; the main association network construction subsystem is used to train and learn the initial association network between known terms based on a term relationship recognition model, output the relationship strength of known relationships between known terms, the relationship strength of potential relationships and potential relationships, and update the initial association network between terms based on the output relationship strength of known relationships and the relationship strength of potential relationships and potential relationships to obtain the main association network; The fused association network forming subsystem is used to compare the identified new terms with the known terms in the main association network. If the new term is different from all the recorded known terms in the main association network, the new term and the analyzed new term attributes are fused with the main association network to form a fused association network; the updated association network forming subsystem is used to input the fused association network into the term relationship recognition model, predict and output the new term and the old term relationship and the relationship strength between the new term and the known term, and update the main association network based on the new term relationship and the relationship strength between the new term and the old term to obtain the updated association network; the aggregate recommendation subsystem is used to aggregate and recommend the query terms based on the updated association network after obtaining the query terms input by the user, and output a term recommendation list.

[0082] In this implementation, the initial association network construction subsystem is responsible for extracting terms and their attributes from standard documents and performing preliminary association analysis to construct an initial association network between terms, providing a structured foundation for the entire system so that the relationships between terms can be preliminarily mapped based on actual data, laying the foundation for subsequent in-depth analysis and optimization.

[0083] Based on the initial association network, the main association network construction subsystem uses the term relationship recognition model for training and learning, thereby updating and refining the relationship strength between terms, including known and potential relationships, and can more accurately identify and predict complex relationships between terms, thereby improving the accuracy and practicality of the network.

[0084] The fused association network formation subsystem compares the newly identified terms with the known terms in the main association network. If the new term is found not in the existing network, its attributes are merged into the main network and the network structure is updated, ensuring the continuous updating and expansion of the term network, reflecting the latest knowledge and information, and enhancing the dynamic adaptability and comprehensiveness of the system.

[0085] The updated association network formation subsystem further analyzes and predicts the fused association network, updates the relationship strength between new and old terms to form the latest updated association network, so that the term network is always kept up to date, ensuring that all relationship evaluations are based on the latest data and analysis, thereby improving the accuracy of recommendations and searches.

[0086] The aggregate recommendation subsystem is based on the updated association network. When a user queries for a specific term, the system can provide a term recommendation list sorted based on relationship strength, which improves the user's query efficiency and satisfaction, enables users to quickly find the most relevant terms, and supports more effective information retrieval and knowledge exploration.

[0087] An electronic device comprises: a processor and a memory, wherein computer program instructions are stored in the memory, and when the computer program instructions are executed by the processor, the processor executes the terminology management service method based on standard archives as described above.

[0088] A computer-readable storage medium is used to store a program, and when the program is executed by a processor, the terminology management service method based on standard archives as described above is implemented.

[0089] In summary, this application has at least the following effects:

[0090] By combining the terminology relationship recognition model and the recommendation system, the speed and accuracy of extracting and retrieving information from a large number of documents have been significantly improved. Combining multi-dimensional factors such as the relationship strength of terms, user historical behavior, and term popularity, the recommendation system can provide more accurate and personalized content, enhance user experience, improve the efficiency and effectiveness of terminology management, and provide users with more intelligent and accurate services, greatly improving the operational performance of information systems and user satisfaction. This is particularly important for environments that need to process and analyze a large number of professional terms (such as academic research, technical document management, etc.), and helps promote knowledge integration, sharing, and innovation.

[0091] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0092] The present invention is described with reference to flowcharts and / or block diagrams of systems, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0093] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0094] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0095] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0096] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A terminology management service method based on standard archives, characterized in that: The following steps are involved: Extract text from standard documents to obtain known terms, obtain the attributes of known terms, perform association analysis on known terms, and build an initial association network between known terms; Based on the term relationship recognition model, the initial association network between known terms is trained and learned, and the relationship strength of known relationships between known terms, the relationship strength of potential relationships and potential relationships are output; based on the output relationship strength of known relationships, potential relationships and potential relationship strength, the initial association network between terms is updated to obtain the main association network; The identified new terms are compared with the known terms in the main association network. If the new terms are different from all the known terms recorded in the main association network, the new terms and the analyzed new term attributes are merged with the main association network to form a fused association network. Input the fused association network into the term relationship recognition model, predict and output the new term and the old term relationship and the relationship strength between the new term and the known term, and update the main association network based on the new term and the old term relationship and the relationship strength between the new term and the old term to obtain the updated association network; After obtaining the query terms input by the user, the query terms are aggregated and recommended based on the updated association network, and a term recommendation list is output; The formula for calculating the relationship strength of a known relationship or potential relationship is as follows: ； In the formula, and represent two known terms with a known relationship, Known term and The number of known or potential relationships between , Known term and The total number of known or potential relationships between Known term and Between The strength of the relationship between known or potential relationships, Known term and The number of times they co-occur in the same document, Preset in the database The weight factor of Known term and Semantic similarity score of surrounding text, Preset in the database The weight factor of Based on known terms, domain experts are pre-set in the database and The semantics and practical application of the relationship weights given, Preset in the database The weight factor of Known term and The rarity of is calculated as the inverse of the ratio of the number of documents containing the two terms to the total number of documents, defined as , is the total number of documents, To include known terms and The number of documents, Preset known terms in the database and Between adjustment factors for known or potential relationships; The step of performing aggregate recommendation on the query terms based on the updated association network and outputting a term recommendation list includes: Obtaining relationship parameters between the query term and the known terms in the updated association network; determining a recommendation index based on the relationship parameter; Arrange the obtained recommendation indexes in descending order, and load the known terms corresponding to the first G recommendation indexes into the term recommendation list; The calculation formula of the recommendation index is: ； In the formula, is the number of the known term in the updated association network, , is the total number of known terms in the updated association network, is the query term, is the updated associated network known terms, Represents the query term and the updated association network The recommendation index of known terms, Indicates known terms and query terms The strength of the relationship, for The weight factor of Indicates Known terms in user The frequency of occurrence in the historical behavior of for The weight factor of Indicates The popularity of known terms, for The weight factor of .

2. A terminology management service method based on standard archives according to claim 1, characterized in that: The method of extracting text from standard documents to obtain known terms, obtaining attributes of known terms, performing association analysis on known terms, and constructing an initial association network between known terms includes: Extracting original text data from the electronic document of the standard document, and performing text cleaning and segmentation on the extracted original text data; Performing term recognition on the original text data after text cleaning and segmentation, marking known terms, and extracting known term attributes, wherein the known term attributes include known term definitions, known term example sentences, and known term usage frequencies; The acquired known terms and known term attributes are stored in a term base; Identify and map known relationships between known subjects based on statistical methods and speech analysis methods, wherein the known relationships include synonym relationships and hyponymy relationships; The identified known relationships, known terms and known term attributes are integrated and stored in the term base to form an initial association network.

3. A terminology management service method based on standard archives according to claim 2, characterized in that: The initial association network is constructed, including: defining known terms as nodes, locating known relationships between known subjects as edges, and converting known term attributes into known term feature vectors of nodes.

4. The terminology management service method based on standard archives according to claim 1 is characterized in that: The initial association network between known terms is trained and learned based on the term relationship recognition model, and the relationship strength of known relationships between known terms, potential relationships and the relationship strength of potential relationships are output, including: The term relationship recognition model takes the initial association network as input, learns the features of known terms and the known relationships between known terms based on the feature vectors converted from the known term attributes of the nodes, and infers and outputs the potential relationships between known terms based on the known relationships between known terms. Obtaining performance evaluation parameters output after the term relationship recognition model is trained, and obtaining a performance evaluation index based on the performance evaluation parameters, wherein the performance evaluation index is used to represent the training result of evaluating the term relationship recognition model. If the performance evaluation index is greater than a set evaluation threshold, the training is completed; if the performance evaluation index is less than or equal to the set evaluation threshold, the model parameters of the term relationship recognition model are optimized and retrained; The trained term relationship recognition model determines the relationship strength of known relationships and the relationship strength of potential relationships based on relationship determination parameters between known terms.

5. A terminology management service method based on standard archives according to claim 4, characterized in that: The performance evaluation parameters include the prediction accuracy, recall rate and F1 score of each term relationship. The calculation formula for the performance evaluation index based on the performance evaluation parameters is as follows: ； In the formula, is the performance evaluation index, is the number of the term relationship, , is the total number of term relationships, For the The weight factor of the term relationship, For the The prediction accuracy of term relationships, for The weight coefficient of For the The recall rate of term relations, for The weight coefficient of For the The F1 score of the term relationship, for The weight coefficient of .

6. The terminology management service method based on standard archives according to claim 1, characterized in that: The new terms and the attributes of the new terms obtained through analysis are integrated with the main association network to form an integrated association network, including: Extract features of new terms to obtain new term attributes, where the new term attributes include new term definitions, new term example sentences, and new term usage frequency; Perform feature transformation on the attributes of the new term to obtain the feature vector of the new term; Merge the new term feature vector with the known term feature vectors in the main association network; Conducting context data analysis on new terms, and initially determining the initial term relationship between the new terms and the known terms in the main association network based on the co-occurrence frequency between the new terms and the known terms in the main association network; A node is created for the new term in the main association network, and an edge is created between the new term and the known terms in the main association network based on the initial term relationship between the new term and the known terms in the main association network.

7. A terminology management service method based on standard archives according to claim 6, characterized in that: The prediction output includes the new-to-old term relationship and the relationship strength between the new term and the known term, including: Obtain the new term feature vector and the initial term relationship obtained by context data analysis to obtain the new and old term relationship; Based on the relationship between the new and old terms, the relationship strength of the relationship between the new and old terms is obtained through parameter analysis.

8. A terminology management service system based on standard archives, used in a terminology management service method based on standard archives as claimed in any one of claims 1 to 7, characterized in that: It includes an initial association network construction subsystem, a main association network construction subsystem, a fusion association network formation subsystem, an updated association network formation subsystem and an aggregate recommendation subsystem, among which: The initial association network construction subsystem is used to extract text from standard documents to obtain known terms, obtain attributes of known terms, perform association analysis on known terms, and construct an initial association network between known terms; The main association network construction subsystem is used to train and learn the initial association network between known terms based on the term relationship recognition model, output the relationship strength of known relationships between known terms, potential relationships and the relationship strength of potential relationships, and update the initial association network between terms based on the output relationship strength of known relationships, potential relationships and the relationship strength of potential relationships to obtain the main association network; The fusion association network forming subsystem is used to compare the identified new term with the known terms in the main association network. If the new term is different from all the known terms recorded in the main association network, the new term and the analyzed new term attributes are fused with the main association network to form a fusion association network. The updated association network forming subsystem is used to input the fused association network into the term relationship recognition model, predict and output the new term and the old term relationship and the relationship strength between the new term and the known term, and update the main association network based on the new term relationship and the relationship strength between the new term and the old term to obtain the updated association network; The aggregate recommendation subsystem is used to perform aggregate recommendation on the query terms based on the updated association network after obtaining the query terms input by the user, and output a term recommendation list.

Images