A legal document and project management field correlation analysis method based on NLP

Abstract

The application provides a legal document and project management field correlation analysis method based on NLP, and belongs to the technical field of project management. The method extracts the dispute focus of the legal document through a large language model, obtains a dispute focus table, and maps the keywords in the dispute focus table to the project management knowledge field to obtain a case dispute reason mapping table containing a first project management field. Then, the entropy weight method is used to calculate the weight of the frequency, cosine distance and mutual information between the first project management field and the dispute focus, and the weight is sorted, the second project management field causing the dispute is screened out, and the Pearson correlation between the sub-fields in the first project management field is calculated to obtain the project management sub-field correlation, and the sub-fields strongly related to the second project management field are screened out according to the project management sub-field correlation. The application can analyze the causes of project disputes from various project management field related texts, and reduce the subjectivity of project dispute analysis.

Description

Technical Field

[0001] This invention relates to the field of project management technology, and in particular to a method for analyzing the correlation between legal documents and project management based on NLP. Background Technology

[0002] Construction projects cover a wide range of areas, including infrastructure construction, real estate development, and municipal engineering. According to a global report on construction project disputes, the average value and resolution time of disputes worldwide have both increased significantly in recent years. Rather than addressing disputes after they arise, identifying the root cause of the dispute is the first step in resolving it.

[0003] Currently, when analyzing the causes of project disputes, the common methods are to identify the root causes through internal project documents, expert meetings, or interviews. These methods have achieved relatively good results, but they also have the following drawbacks:

[0004] 1. Internal project data can only display information specific to that project. The unique nature of each project indicates that different projects have vastly different characteristics, and the reasons for disputes also vary. Therefore, results derived from one project are difficult to apply to other projects, and the analytical methods used have poor universality.

[0005] 2. The interview results for project disputes mainly rely on manual or expert analysis, and the reasons for the disputes are highly subjective. Summary of the Invention

[0006] In view of this, the present invention provides a method for correlation analysis between legal documents and project management based on NLP, in order to solve the problems of excessive subjectivity and poor universality of current project dispute cause analysis techniques.

[0007] The technical solution adopted in this invention is:

[0008] This invention provides a method for NLP-based correlation analysis of legal documents and project management, including:

[0009] Legal documents are obtained and input into a large language model to extract the points of contention, resulting in a table of points of contention.

[0010] Mapping the causal keywords in the dispute focus table to the project management knowledge domain, we obtain a case dispute cause mapping table that includes the first project management domain;

[0011] The frequency of each sub-domain in the first project management domain is obtained by counting the number of times it appears in the table of points of contention.

[0012] A joint thesaurus is constructed based on the disputed texts in the dispute focus table and the case dispute cause mapping table, and the mutual information between the disputed texts and the first project management domain is calculated based on the joint thesaurus.

[0013] Convert the disputed text in the dispute focus table into dispute vectors, and convert the first project management domain in the case dispute cause mapping table into management domain vectors. Calculate the cosine distance between each dispute vector and each management domain vector.

[0014] The weighted importance values ​​of frequency, cosine distance and mutual information are calculated using the entropy weight method, and the weights are ranked according to importance. The second project management domain is then selected from the first project management domain according to a preset screening threshold.

[0015] The causal keywords corresponding to the subdomains in the first project management domain are mapped to a set of project domain vectors, and the Pearson correlation between every two sets of project domain vectors is calculated to obtain the project management subdomain correlation. At the same time, based on the project management subdomain correlation, subdomains that are strongly correlated with the second project management domain are selected from the first project management domain.

[0016] Furthermore, the step of acquiring and inputting legal documents into a large language model to extract the points of contention, resulting in a point-of-contention table, includes:

[0017] Obtain legal documents related to project management and input them into a large language model. Use the large language model to extract the case number, plaintiff's claim text, and defendant's response text from each legal document to form the focus of the dispute.

[0018] The TF-IDF algorithm is used to extract causal keywords from the points of contention, and the causal keywords of multiple points of contention are combined to construct a point of contention table for project management.

[0019] Furthermore, the process of mapping the causal keywords in the dispute focus table to the project management knowledge domain yields a case dispute cause mapping table that includes the first project management domain, including:

[0020] Use the Jieba word segmentation tool to segment the cause keywords in the dispute focus table, remove stop words, and obtain a spare keyword table;

[0021] A stop word list (WL) was created by combining expert opinions on project management and the field of invention.

[0022] Based on the stop word list, the alternative keyword list is segmented again using the TF-IDF algorithm to obtain the reason keyword list;

[0023] Extract subdomains from the project management process and their corresponding major project management knowledge domains, and use these subdomains and major project management knowledge domains as the first project management domain;

[0024] The semantics of the cause keywords in the cause keyword table are compared with the text of the first project management domain. The cause keywords are classified into the first project management domain that is semantically similar to the cause keywords, forming a case dispute cause mapping table; each first project management domain has a project domain identifier.

[0025] Furthermore, the frequency of occurrence of each sub-domain within the first project management domain in the dispute focus table is obtained by counting the number of times the sub-domains appear in the first project management domain, including:

[0026] Based on the project domain identifier of the first project management domain, a dispute label is set for each dispute point in the dispute point table, and each dispute label represents the project management domain that caused the dispute;

[0027] The frequency of each first project management area is obtained by counting the number of times it appears in the dispute focus table based on the disputed tags.

[0028] Furthermore, the construction of a joint thesaurus based on the disputed text in the dispute focus table and the case dispute cause mapping table, and the calculation of mutual information between the disputed text and the first project management domain based on the joint thesaurus, includes:

[0029] Using the Lawformer pre-trained model, a joint thesaurus is constructed between each disputed text in the dispute focus table and each first project management domain in the case-dispute cause mapping table.

[0030] The first and second frequencies of each causal keyword in the joint thesaurus were counted in the disputed text and the first project management domain, respectively, and the third frequency of each causal keyword appearing simultaneously in the disputed text and the first project management domain was also counted.

[0031] Based on the first, second, and third frequencies of each causal keyword in the joint thesaurus, the marginal and joint probabilities of the disputed text with the first project management domain are calculated. The mutual information of multiple subdomains in each first project management domain is obtained, and the mutual information of multiple subdomains is summed and averaged to obtain the average mutual information of each first project management domain.

[0032] Further, the process of converting the disputed text in the dispute focus table into dispute vectors, and converting the first project management domain in the case dispute cause mapping table into management domain vectors, and calculating the cosine distance between each dispute vector and each management domain vector, includes:

[0033] The Lawformer pre-trained model is used to map each point of contention in the contention focus table to a contention vector. At the same time, the Lawformer pre-trained model is used to map each subdomain of the first project management domain in the case dispute cause mapping table to a subdomain management domain vector.

[0034] The Lawformer pre-trained model is used to compute the vector dot product of each dispute vector and each subdomain management domain vector, as well as the vector length of each dispute vector and the vector length of each subdomain management domain vector.

[0035] The cosine similarity between each dispute vector and each subdomain management domain vector is calculated based on the vector length of each dispute vector, the vector length of each subdomain management domain vector, and the corresponding vector dot product. The cosine distance between each subdomain is obtained by subtracting the cosine similarity from 1.

[0036] The average cosine distance of the first project management domain is obtained by summing and averaging the cosine distances of all subdomains.

[0037] Further, the step of calculating the weighted importance values ​​of frequency, cosine distance, and mutual information using the entropy weight method, ranking the weighted importance values, and selecting the second project management domain from the first project management domain according to a preset screening threshold includes:

[0038] Obtain the frequency index value, cosine distance index value, and mutual information index value of each subdomain in the first project management domain, and form an index value matrix;

[0039] Based on the index value matrix, the frequency weight, cosine distance weight, and mutual information weight are calculated using the entropy weight method.

[0040] The frequency weight is multiplied by the frequency index value, the cosine distance weight is multiplied by the cosine distance index value, and the mutual information weight is multiplied by the mutual information index value, and the three products are added together to obtain the comprehensive importance value for each specific domain.

[0041] The overall importance values ​​of all project management domains in the first project management domain are sorted in descending order. Then, based on a preset screening threshold, subdomains whose overall importance values ​​exceed the preset screening threshold are selected as the second project management domains.

[0042] Further, the step of mapping the causal keywords corresponding to the subdomains in the first project management domain to a set of project domain vectors, and calculating the Pearson correlation between every two sets of project domain vectors to obtain the project management subdomain correlation, and simultaneously filtering out subdomains strongly correlated with the second project management domain from the first project management domain based on the project management subdomain correlation, including:

[0043] The Lawformer pre-trained model is used to map the cause keywords of the subdomains in the first project management domain into a set of project domain vectors in the case dispute cause mapping table.

[0044] Calculate the Pearson correlation between the project domain vectors of every two subdomains, determine the project management subdomain correlation based on the Pearson correlation calculation results, and select subdomains that are strongly correlated with the second project management domain from the first project management domain based on the project management subdomain correlation.

[0045] In summary, the beneficial effects of the present invention are as follows:

[0046] This invention provides a method for NLP-based correlation analysis between legal documents and project management domains. The method involves acquiring legal documents and inputting them into a large language model to extract the points of contention, resulting in a point-of-contention table. The causal keywords in this table are then mapped to project management knowledge domains, creating a mapping table of case dispute causes within a first project management domain. Next, the entropy weight method is used to calculate the frequency, cosine distance, and mutual information weights between the first project management domain and the points of contention, and these weights are ranked. A second project management domain leading to the dispute is then selected from the first project management domain according to a preset screening threshold. Finally, the causal keywords corresponding to the subdomains within the first project management domain are mapped to a set of project domain vectors, and the Pearson correlation between every two sets of project domain vectors is calculated to obtain the project management subdomain correlation. Simultaneously, based on the project management subdomain correlation, subdomains strongly correlated with the second project management domain are selected from the first project management domain, thus identifying the second project management domain requiring focused attention, as well as its strongly correlated subdomains. By utilizing objective legal documents and project management domains for correlation analysis, the causes of project disputes can be analyzed from various project management domain-related texts, reducing the subjectivity of project dispute analysis. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments of the present invention will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, and these are all within the protection scope of the present invention.

[0048] Figure 1 This is a flowchart of a method for analyzing the correlation between legal documents and project management based on NLP, as described in this invention.

[0049] Figure 2 This is a schematic diagram of the legal dispute-project management knowledge domain correlation analysis framework of the present invention;

[0050] Figure 3 This is a flowchart of the keyword table extraction process of the present invention;

[0051] Figure 4 This is a flowchart of the domain mutual information calculation process of the present invention;

[0052] Figure 5 This is a flowchart of the field cosine distance calculation for the present invention;

[0053] Figure 6 This is a flowchart of the entropy weight method calculation of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Unless otherwise specified, the present invention and the various features in the embodiments can be combined with each other, all of which are within the protection scope of the present invention.

[0055] The detailed implementation process of this invention is shown in the following embodiments.

[0056] Example 1: Refer to Figure 1 and Figure 2 As shown, Figure 1 This is a flowchart of a method for correlation analysis between legal documents and project management based on NLP, according to the present invention. Figure 2 This is a schematic diagram of the legal dispute-project management knowledge domain correlation analysis framework. The method in this embodiment includes:

[0057] S1: Obtain legal documents and input them into the large language model to extract the points of contention, thus obtaining a table of points of contention;

[0058] S2: Map the causal keywords in the dispute focus table to the project management knowledge domain to obtain a case dispute cause mapping table that includes the first project management domain;

[0059] S3: Count the number of times the sub-domains of the first project management domain appear in the dispute focus table to obtain the frequency of each sub-domain in the first project management domain;

[0060] S4: Construct a joint terminology based on the disputed texts in the dispute focus table and the case dispute cause mapping table, and calculate the mutual information between the disputed texts and the first project management domain based on the joint terminology;

[0061] S5: Convert the disputed text in the dispute focus table into dispute vectors, and convert the first project management domain in the case dispute cause mapping table into management domain vectors. Calculate the cosine distance between each dispute vector and each management domain vector.

[0062] S6: Calculate the weight importance values ​​of frequency, cosine distance and mutual information using the entropy weight method, sort the weight importance values, and select the second project management domain from the first project management domain according to the preset screening threshold;

[0063] S7: Map the causal keywords corresponding to the subdomains in the first project management domain to a set of project domain vectors, and calculate the Pearson correlation between every two sets of project domain vectors to obtain the project management subdomain correlation. At the same time, based on the project management subdomain correlation, select subdomains that are strongly correlated with the second project management domain from the first project management domain.

[0064] This embodiment extracts dispute focus by acquiring legal documents and inputting them into a large language model, resulting in a dispute focus table. The causal keywords in this table are then mapped to project management knowledge domains, creating a case dispute cause mapping table encompassing the first project management domain. Next, the entropy weight method is used to calculate the frequency, cosine distance, and mutual information weights between the first project management domain and the dispute focus, and these weights are ranked. A second project management domain leading to the dispute is then selected from the first project management domain according to a preset screening threshold. Finally, the causal keywords corresponding to the subdomains within the first project management domain are mapped to a set of project domain vectors, and the Pearson correlation between every two sets of project domain vectors is calculated to obtain the project management subdomain correlation. Simultaneously, based on the project management subdomain correlation, subdomains strongly correlated with the second project management domain are selected from the first project management domain. By using objective legal documents and project management domains for correlation analysis, the causes of project disputes can be analyzed from various project management domain-related texts, reducing the subjectivity of project dispute analysis.

[0065] Specifically, this embodiment describes the detailed implementation process of each step of the above method as follows:

[0066] In this embodiment, step S1 specifically includes the following sub-steps:

[0067] S11: Obtain legal documents related to project management and input them into a large language model. Use the large language model to extract the case number and plaintiff's claim text (e.g., ...) from each legal document. ) and the defendant's response text (such as This forms the focus of the dispute. Legal documents refer to legal documents such as legal rulings related to project management disputes.

[0068] S12: Use the TF-IDF algorithm to extract causal keywords from the points of contention, and obtain the corresponding causal keyword table. KL :

[0069] ;

[0070] in, Representing the n One reason keyword, Indicates the first n The weight of each reason's keywords.

[0071] Finally, combine the keyword list of reasons for multiple points of contention. KL Obtain the list of points of contention in project management. FL :

[0072] ;

[0073] Each ruling corresponds to one row in the table of disputed issues.

[0074] In this embodiment, refer to Figure 3 The keyword extraction process shown in step S2 includes the following sub-steps:

[0075] S21: Use the Jieba word segmentation tool to segment the cause keywords in the dispute focus table, remove stop words, and obtain a spare keyword table;

[0076] S22: Create a stop word list by combining expert opinions on project management and the field of invention. WL :

[0077] ;

[0078] in, This indicates the nth stop word.

[0079] S23: Based on the stop word list, the backup keyword list is segmented a second time using the TF-IDF algorithm to obtain the reason keyword list. The specific calculation formula for secondary segmentation using the TF-IDF algorithm is as follows:

[0080] ;

[0081] ;

[0082] Where t represents the target word, d represents the current document, and D represents the document set. This indicates the number of times the target word t appears in document d. Let t represent the total number of words in document set D, TF represent term frequency, IDF(t,D) represent inverse document frequency, reflecting the general importance of word t in the entire document set D, TF(t,d) represent the term frequency of target word t in document d, and TF-IDF(t,d,D) is term frequency-inverse document frequency, used to quantify the frequency of a word. In the document The importance of [the subject / method].

[0083] ;

[0084] Where N represents the total number of documents in document set D, This indicates the number of documents containing the target word t.

[0085] S24: Extract subdomains and corresponding major project management knowledge domains from the project management process, and use these subdomains and major project management knowledge domains as the first project management domain. There are 49 subdomains and 10 major project management knowledge domains.

[0086] S25: Semantically compare the cause keywords in the cause keyword table with the text in the first project management domain, and categorize the cause keywords into the first project management domain that is semantically similar to the cause keywords, forming a case dispute cause mapping table. Each first project management domain has a project domain identifier. The case dispute cause mapping table is specifically represented as follows:

[0087] ;

[0088] in, This represents the nth index.

[0089] In this embodiment, step S3 specifically includes the following sub-steps:

[0090] S31: Based on the project area identifier of the first project management area, assign a dispute label to each dispute point in the dispute point table. Each dispute label represents the project management area that caused the dispute. Dispute Labels FL_Label It can be represented as:

[0091] ;

[0092] in, The label representing the nth dispute.

[0093] S32: Based on the disputed tags, count the number of times each sub-domain in the first project management domain appears in the disputed focus table to obtain the frequency of each sub-domain in each first project management domain. The formula for calculating the frequency is as follows:

[0094] ;

[0095] Where N represents the number of documents, and L represents the Lth tag. This indicates that there exists a document with label K.

[0096] In this embodiment, refer to Figure 4 The mutual information calculation process shown includes the following sub-steps in step S4:

[0097] S41: Construct a joint thesaurus of each disputed text in the dispute focus table and each first project management domain in the case-dispute cause mapping table using the Lawformer pre-trained model;

[0098] S42: Count the first and second frequencies of each causal keyword in the joint thesaurus in the disputed text and the first project management domain, respectively, and count the third frequency of each causal keyword in both the disputed text and the first project management domain;

[0099] S43: Based on the first, second, and third frequencies of each causal keyword in the joint thesaurus, calculate the marginal and joint probabilities between the disputed text and the first project management domain, obtain the mutual information of multiple subdomains within each first project management domain, and sum and average the mutual information of multiple subdomains to obtain the average mutual information of each first project management domain. The specific formula for calculating mutual information is as follows:

[0100] ;

[0101] ;

[0102] in, x , y They are two discrete random variables. yes , The joint probability. , yes , The marginal probability. It is usually based on the natural logarithm or with a base of 2. Let j be the mutual information value of the i-th subdomain, and n be the number of documents. Represents mutual information within subdomains. This represents the average mutual information in the first project management domain. r represents the number of subdomains, and c represents the number of mutual information values.

[0103] In this embodiment, refer to Figure 5 The illustrated process for calculating the cosine distance within the domain, specifically step S5, includes the following sub-steps:

[0104] S51: Using the Lawformer pre-trained model, each point of contention in the dispute focus table is mapped to a dispute vector. At the same time, the Lawformer pre-trained model is used to map each subdomain of the first project management domain in the case dispute cause mapping table to a subdomain management domain vector, thereby obtaining the text vector of each dispute and each management domain.

[0105] Among them, the Lawformer pre-trained model is a pre-trained language model for long legal texts.

[0106] S52: Calculate the vector dot product of each dispute vector and each subdomain management domain vector using the Lawformer pre-trained model, as well as the vector length of each dispute vector and the vector length of each subdomain management domain vector.

[0107] S53: Calculate the cosine similarity between each dispute vector and each subdomain management vector based on the vector length of each dispute vector, the vector length of each subdomain management vector, and the corresponding vector dot product, and subtract the cosine similarity from 1 to obtain the cosine distance of each subdomain.

[0108] S54: The average cosine distance of the first project management domain is obtained by summing and averaging the cosine distances of all subdomains of the first project management domain.

[0109] The formula for calculating the cosine distance is as follows:

[0110] ;

[0111] ;

[0112] ;

[0113] Where A and B are two n-dimensional vectors, representing the dispute vector and the subdomain management domain vector, respectively; It is the angle between A and B; symbol Represents the dot product operation of vectors; The norm of a vector. , The first and second vectors are the first and second vectors, respectively. Each component. Let be the j-th cosine value of the i-th management domain, and n be the number of documents. The cosine similarity between the dispute vector and the management domain vector of each subdomain is represented; the cosine distance represents the cosine distance between the subdomains. This represents the mean cosine distance in the first project management domain.

[0114] In this embodiment, refer to Figure 6 The domain cosine distance calculation process shown in step S6 is to first obtain the frequency index value, cosine distance index value and mutual information index value of each subdomain in the first project management domain, and form an index value matrix.

[0115] Based on the index value matrix, the frequency weight, cosine distance weight, and mutual information weight are calculated using the entropy weight method.

[0116] The frequency weight is multiplied by the frequency index value, the cosine distance weight is multiplied by the cosine distance index value, and the mutual information weight is multiplied by the mutual information index value. The three products are then added together to obtain the comprehensive importance value for each specific domain.

[0117] The overall importance values ​​of all project management areas in the first project management area are sorted in descending order. Then, according to a preset screening threshold, sub-areas whose overall importance values ​​exceed the preset screening threshold are selected as the second project management areas, thus obtaining the important management areas that lead to disputes.

[0118] This embodiment calculates the frequency, cosine distance, and mutual information of each project management domain within the first project management domain. It then multiplies the weight of each indicator by its own value and sums the results to obtain the final importance value for each subdomain. These importance values ​​are then sorted from largest to smallest. This allows for the selection of the top n project management domains based on a preset screening threshold; these are the second project management domains. The selected second project management domains are those that require focused attention in practice; the more important they are, the more likely they are to cause disputes.

[0119] In this embodiment, step S7 specifically includes the following sub-steps:

[0120] S71: Using the Lawformer pre-trained model, map the cause keywords of the subdomains in the first project management domain to a set of project domain vectors in the case dispute cause mapping table.

[0121] S72: Calculate the Pearson correlation between the project domain vectors of every two sub-domains. Determine the project management sub-domain correlation based on the Pearson correlation calculation results, and filter out sub-domains strongly correlated with the second project management domain from the first project management domain based on the project management sub-domain correlation. When filtering strongly correlated sub-domains, a strong correlation threshold can be set. Although the sub-domains strongly correlated with the second project management domain are not the project management domains that directly cause disputes, these sub-domains are strongly correlated with project management domains that are prone to causing disputes. Therefore, they also need attention. Thus, this embodiment uses correlation analysis between legal documents and project management domains to filter out the second project management domain that requires attention, as well as the sub-domains strongly correlated with the second project management domain.

[0122] The formula for calculating Pearson correlation is as follows:

[0123] ;

[0124] in, It is a variable and variables The covariance. , They are variables ,variable The standard deviation of . When At that time, there is a perfectly positive linear correlation. When perfectly negative linear correlation, Time is not related to wireless properties.

[0125] This embodiment has the following technical advantages:

[0126] 1. By using Large Language Model (LLM) to automatically extract the points of contention in legal documents and generate a keyword mapping table, the subjectivity and limitations of traditional manual analysis are effectively overcome, ensuring the comprehensiveness and objectivity of the extraction of points of contention and laying an accurate data foundation for subsequent correlation analysis.

[0127] 2. By mapping the key keywords of the dispute to the project management knowledge domain to generate a case dispute cause mapping table, a direct link between legal disputes and the standardized project management domain is realized, and complex legal texts are structured into a quantifiable and analyzable project management framework.

[0128] 3. By using the entropy weight method to comprehensively calculate and rank the project management sub-domains that lead to disputes, the core project management domains that have a significant impact on the occurrence of disputes can be identified scientifically and objectively. This effectively avoids the one-sidedness of single indicator analysis and improves the accuracy and reliability of the selection of key cause domains.

[0129] 4. By mapping the causal keywords in the project management subdomains that lead to disputes to text vectors and calculating the Pearson correlation between the vectors to obtain the project management subdomain correlation, we can deeply reveal the intrinsic correlation strength and synergistic effect patterns between different project subdomains, thereby gaining a more comprehensive understanding of the complex causes of disputes.

[0130] 5. Finally, by combining the importance ranking of project management subdomains and their relevance, it can simultaneously help prevent subdomains and their strongly related domains that may lead to disputes in project management practice. It realizes a closed loop of correlation analysis based on objective data (legal document texts) and rigorous algorithms (LLM extraction, entropy weight method screening, vector mapping and relevance calculation), so that the final determined causes of project disputes have a high degree of objectivity, interpretability and pertinence.

[0131] Therefore, by comprehensively utilizing technologies such as natural language processing, domain mapping, multi-index weight calculation, and correlation analysis, this invention can systematically and objectively analyze and extract the root causes and related relationships of project disputes from massive amounts of legal documents, significantly reducing the subjectivity and arbitrariness of the analysis process and providing strong data support and decision-making basis for project risk management.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for analyzing the correlation between legal documents and project management based on NLP, characterized in that, include: Legal documents are obtained and input into a large language model to extract the points of contention, resulting in a table of points of contention. Mapping the causal keywords in the dispute focus table to the project management knowledge domain, we obtain a case dispute cause mapping table that includes the first project management domain; The frequency of each sub-domain in the first project management domain is obtained by counting the number of times it appears in the table of points of contention. A joint thesaurus is constructed based on the disputed texts in the dispute focus table and the case dispute cause mapping table, and the mutual information between the disputed texts and the first project management domain is calculated based on the joint thesaurus. Convert the disputed text in the dispute focus table into dispute vectors, and convert the first project management domain in the case dispute cause mapping table into management domain vectors. Calculate the cosine distance between each dispute vector and each management domain vector. The weighted importance values ​​of frequency, cosine distance and mutual information are calculated using the entropy weight method, and the weights are ranked according to importance. The second project management domain is then selected from the first project management domain according to a preset screening threshold. The causal keywords corresponding to the subdomains in the first project management domain are mapped to a set of project domain vectors, and the Pearson correlation between every two sets of project domain vectors is calculated to obtain the project management subdomain correlation. At the same time, based on the project management subdomain correlation, subdomains that are strongly correlated with the second project management domain are selected from the first project management domain.

2. The method for NLP-based correlation analysis of legal documents and project management as described in claim 1, characterized in that, The process of acquiring legal documents and inputting them into a large language model to extract the points of contention results in a point-of-contention table, including: Obtain legal documents related to project management and input them into a large language model. Use the large language model to extract the case number, plaintiff's claim text, and defendant's response text from each legal document to form the focus of the dispute. The TF-IDF algorithm is used to extract causal keywords from the points of contention, and the causal keywords of multiple points of contention are combined to construct a point of contention table for project management.

3. The method for analyzing the correlation between legal documents and project management based on NLP according to claim 1, characterized in that, The process of mapping the causal keywords in the dispute focus table to the project management knowledge domain yields a case dispute cause mapping table that includes the first project management domain, including: Use the Jieba word segmentation tool to segment the cause keywords in the dispute focus table, remove stop words, and obtain a spare keyword table; A stop word list (WL) was created by combining expert opinions on project management and the field of invention. Based on the stop word list, the alternative keyword list is segmented again using the TF-IDF algorithm to obtain the reason keyword list; Extract subdomains from the project management process and their corresponding major project management knowledge domains, and use these subdomains and major project management knowledge domains as the first project management domain; The semantics of the cause keywords in the cause keyword table are compared with the text of the first project management domain. The cause keywords are classified into the first project management domain that is semantically similar to the cause keywords, forming a case dispute cause mapping table; each first project management domain has a project domain identifier.

4. The method for analyzing the correlation between legal documents and project management based on NLP according to claim 1, characterized in that, The frequency of occurrence of sub-domains within the first project management domain in the dispute focus table is obtained by counting the number of times each sub-domain appears in the first project management domain, including: Based on the project domain identifier of the first project management domain, a dispute label is set for each dispute point in the dispute point table, and each dispute label represents the project management domain that caused the dispute; The frequency of each first project management area is obtained by counting the number of times it appears in the dispute focus table based on the disputed tags.

5. The method for NLP-based correlation analysis of legal documents and project management as described in claim 1, characterized in that, The construction of a joint thesaurus based on the disputed text in the dispute focus table and the case dispute cause mapping table, and the calculation of mutual information between the disputed text and the first project management domain based on the joint thesaurus, including: Using the Lawformer pre-trained model, a joint thesaurus is constructed between each disputed text in the dispute focus table and each first project management domain in the case-dispute cause mapping table. The first and second frequencies of each causal keyword in the joint thesaurus were counted in the disputed text and the first project management domain, respectively, and the third frequency of each causal keyword appearing simultaneously in the disputed text and the first project management domain was also counted. Based on the first, second, and third frequencies of each causal keyword in the joint thesaurus, the marginal and joint probabilities of the disputed text with the first project management domain are calculated. The mutual information of multiple subdomains in each first project management domain is obtained, and the mutual information of multiple subdomains is summed and averaged to obtain the average mutual information of each first project management domain.

6. The method for NLP-based correlation analysis of legal documents and project management as described in claim 1, characterized in that, The process of converting disputed text in the dispute focus table into dispute vectors and converting the first project management domain in the case dispute cause mapping table into management domain vectors, and calculating the cosine distance between each dispute vector and each management domain vector, includes: The Lawformer pre-trained model is used to map each point of contention in the contention focus table to a contention vector. At the same time, the Lawformer pre-trained model is used to map each subdomain of the first project management domain in the case dispute cause mapping table to a subdomain management domain vector. The Lawformer pre-trained model is used to compute the vector dot product of each dispute vector and each subdomain management domain vector, as well as the vector length of each dispute vector and the vector length of each subdomain management domain vector. The cosine similarity between each dispute vector and each subdomain management domain vector is calculated based on the vector length of each dispute vector, the vector length of each subdomain management domain vector, and the corresponding vector dot product. The cosine distance between each subdomain is obtained by subtracting the cosine similarity from 1. The average cosine distance of each subdomain of the first project management domain is obtained by summing and averaging the cosine distances of all subdomains of the first project management domain.

7. The method for NLP-based correlation analysis of legal documents and project management as described in claim 1, characterized in that, The process of calculating the weighted importance values ​​of frequency, cosine distance, and mutual information using the entropy weight method, ranking the weighted importance values, and selecting the second project management domain from the first project management domain according to a preset screening threshold includes: Obtain the frequency index value, cosine distance index value, and mutual information index value of each subdomain in the first project management domain, and form an index value matrix; Based on the index value matrix, the frequency weight, cosine distance weight, and mutual information weight are calculated using the entropy weight method. The frequency weight is multiplied by the frequency index value, the cosine distance weight is multiplied by the cosine distance index value, and the mutual information weight is multiplied by the mutual information index value, and the three products are added together to obtain the comprehensive importance value for each specific domain. The overall importance values ​​of all project management domains in the first project management domain are sorted in descending order. Then, based on a preset screening threshold, subdomains whose overall importance values ​​exceed the preset screening threshold are selected as the second project management domains.

8. The method for NLP-based correlation analysis of legal documents and project management as described in claim 1, characterized in that, The process involves mapping the causal keywords corresponding to subdomains within the first project management domain to a set of project domain vectors, calculating the Pearson correlation between every two sets of project domain vectors to obtain the project management subdomain correlation, and simultaneously filtering out subdomains strongly correlated with the second project management domain from the first project management domain based on the project management subdomain correlation, including: The Lawformer pre-trained model is used to map the cause keywords of the subdomains in the first project management domain into a set of project domain vectors in the case dispute cause mapping table. Calculate the Pearson correlation between the project domain vectors of every two subdomains, determine the project management subdomain correlation based on the Pearson correlation calculation results, and select subdomains that are strongly correlated with the second project management domain from the first project management domain based on the project management subdomain correlation.

Images