International commercial law retrieval system and method based on artificial intelligence

Abstract

The invention discloses an international commercial law retrieval system and method based on artificial intelligence, and relates to the technical field of legal retrieval. Feature words are accurately extracted from a retrieval request input by a user through a natural language processing technology; in combination with semantic understanding, performing normalization processing on synonyms and synonyms; performing similarity calculation by adopting a vector space model in combination with a cosine similarity algorithm to obtain feature word similarity of each legal text and the retrieval request in the legal knowledge base; sorting the retrieval results according to the similarity, and preferentially displaying the most relevant legal text; a collaborative filtering algorithm is combined with a content-based recommendation algorithm, historical retrieval records and current retrieval requests of a user are collected, a user interest model is constructed, and legal information which the user may be interested in is recommended. According to the method, retrieval omission caused by insufficient semantic understanding can be effectively avoided, and the accuracy of a retrieval result is greatly improved; and the most relevant legal text is preferentially displayed, so that the retrieval efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of legal retrieval technology, and in particular to an international commercial law retrieval system and method based on artificial intelligence. Background Art

[0002] When dealing with international business affairs, companies, lawyers, legal personnel, and scholars need to frequently consult a large amount of international commercial law materials to obtain accurate legal basis. However, traditional legal search methods have many disadvantages:

[0003] On the one hand, the search method based on keyword matching has difficulty understanding the natural language semantics of user input, and often misses key legal texts related to synonyms and near-synonyms, resulting in incomplete and inaccurate search results;

[0004] On the other hand, search results are usually sorted according to fixed rules, without fully considering users' personalized needs and interest preferences, and are unable to efficiently push the most relevant information to users.

[0005] In addition, faced with the complex and ever-changing international commercial law system, manual retrieval is time-consuming and laborious, and it is difficult to adapt to the fast pace of business.

[0006] Therefore, there is an urgent need for an international commercial law search technology that is intelligent, precise and can meet the personalized needs of users to solve the above problems. Summary of the invention

[0007] The purpose of the present invention is to provide an international commercial law retrieval method and system based on artificial intelligence, which, by integrating advanced technologies such as natural language processing and machine learning, can achieve accurate extraction of retrieval feature words, precise calculation of text similarity, and personalized recommendation of legal information, thereby improving the efficiency and accuracy of international commercial law retrieval and meeting the diverse needs of users.

[0008] In order to solve the above technical problems, the present invention provides a technical solution: an international commercial law search method based on artificial intelligence, comprising the following steps:

[0009] S1. Through lexical analysis and part-of-speech tagging in natural language processing technology, feature words are accurately extracted from the search request input by the user; at the same time, synonyms and near-synonyms are normalized in combination with semantic understanding;

[0010] S2. Use the vector space model combined with the cosine similarity algorithm to calculate the similarity and obtain the similarity between the feature words of each legal text in the legal knowledge base and the search request;

[0011] S3. Sort the search results according to similarity, and give priority to displaying legal texts with high similarity;

[0012] S4. Use collaborative filtering algorithms combined with content-based recommendation algorithms to collect users’ historical search records and current search requests, build user interest models, and recommend legal information that users may be interested in.

[0013] Furthermore, in step S1, a deep learning model is introduced to capture potential key feature words in complex contexts; at the same time, combined with deep semantic understanding, intelligent normalization processing is performed on synonyms, near-synonyms, and polysemous words. The normalization processing method is as follows:

[0014] Assume that the user's search request text is T, and after deep preprocessing, the word set W = w 1 ,w 2 ,…,w n , through the pre-trained deep semantic knowledge base and the dynamically updated semantic mapping function f s Normalize the semantics of each word and get the feature word set F = f s (w 1 ),f s (w 2 ),…f s (w n ), where f s Based on the pre-trained semantic knowledge base, synonym conversion and polysemy analysis operations of words are realized.

[0015] Furthermore, in step S2, the specific method for calculating the similarity between each legal text in the legal knowledge base and the feature word of the search request is as follows:

[0016] Vectorize the feature words of each legal text and search request in the legal knowledge base; use the TF-IDF algorithm to assign weights to each feature word and construct a text vector; then the feature word f in the legal text L i The TF-IDF value is calculated as TF(f i ,L)×IDF(f i ), where TF(f i ,L) is f i The frequency of words in L, N is the total number of legal texts, DF(f i ) is the one containing f i Number of legal texts;

[0017] For the retrieval request feature word vector V q =(v q1 ,v q2 ,…,v qm ) and the legal text vector V l =(v l1 ,v l2 ,…,v lm ), their cosine similarity calculation formula is:

[0018]

[0019] A larger value of the cosine similarity indicates a higher similarity.

[0020] Furthermore, in step S3, the search results are sorted in descending order according to the similarity values ​​calculated in step S2, and the legal texts with high similarity are placed in the front and displayed to the user first.

[0021] Furthermore, in step S4, the collaborative filtering algorithm is combined with the content-based recommendation algorithm to collect the user's historical search records and build a user interest model; the collaborative filtering algorithm finds user groups with similar search behaviors and mines the relevant legal information that they have searched but the current user has not searched; the content-based recommendation analyzes the content association between the feature words of the current search request and the unsearched text in the legal knowledge base; the specific method of building the user interest model is as follows:

[0022] The historical search feature word set of user u is H u =h u1 ,h u2 ,…,h uk , the current search request feature word set is F, for the legal text L j , collaborative filtering similarity calculation is sim cf (u,L j ), calculated based on content similarity as sim cb (F,L j ), comprehensive recommendation score R = λsim cf (u,L j )+(1-λ)sim cb (F,L j ), where λ balances the weights of the two algorithms.

[0023] Furthermore, in the process of feature word extraction, a professional terminology dictionary is established for professional legal terminology and updated in real time, combined with semantic normalization processing to ensure the accurate extraction and standardized conversion of professional terminology; at the same time, knowledge graph technology is used to expand the associations of feature words, and related upstream and downstream legal concepts are mined as supplementary feature words to further enrich the retrieval clues.

[0024] Furthermore, in the similarity calculation stage, hierarchical weighted processing is adopted for the cited clauses, explanatory notes and other ancillary text information in the legal text, and different weights are assigned according to the degree of their connection with the core legal provisions, so that the similarity calculation is more in line with the actual value of the legal text.

[0025] Furthermore, during the intelligent recommendation process, the user interest model is regularly updated and verified, and the model parameters are dynamically adjusted based on factors such as recent changes in user search behavior and dynamic updates of industry regulations, to ensure that the recommended content always meets the user's latest needs. At the same time, a recommendation effect feedback mechanism is established to collect user behavior data such as clicks, collections, and evaluations of recommended content, and reversely optimize the recommendation algorithm.

[0026] The present invention also provides an international commercial law retrieval system based on artificial intelligence, comprising

[0027] User interface: used for interaction between users and the system, providing one-stop functions such as legal search, result display, case analysis, regulatory interpretation, online consultation, etc. The interface design follows the user experience principles and is adapted to multi-terminal access.

[0028] Legal knowledge base: a database containing international commercial law regulations, legal interpretations, and legal cases; it integrates multiple legal resources such as international treaties and cross-border trade practices, and builds a knowledge association network through knowledge graph technology to facilitate rapid retrieval and association expansion;

[0029] Natural language processing module: connects the user interface with the legal knowledge base and is used to convert the text of the legal knowledge base and the search text entered by the user into a format that can be understood by the computer, including text preprocessing, vocabulary embedding, dependency parsing and named entity recognition; it also has functions such as text error correction, semantic repair, and complex sentence decomposition to improve the accuracy and completeness of text processing;

[0030] Machine learning module: connects to the legal knowledge base and is used to learn legal rules and principles from the legal knowledge base, including feature extraction, model training and evaluation;

[0031] Legal reasoning engine: connects the legal knowledge base and machine learning module, and is used to perform legal reasoning based on the legal rules and principles in the legal knowledge base and machine learning module. It combines case reasoning, rule reasoning and deep learning reasoning technology to generate professional and accurate legal advice and answers for the search text entered by the user.

[0032] The advantages of the present invention compared with the prior art are:

[0033] Accurate retrieval: By combining natural language processing with deep learning technology, we can deeply understand the semantics of user search requests, accurately extract feature words and normalize them, effectively avoid search omissions due to insufficient semantic understanding, greatly improve the accuracy of search results, and make the retrieved legal texts highly consistent with user needs.

[0034] Efficient sorting: Based on the vector space model and cosine similarity algorithm, the text similarity is calculated and the search results are sorted in descending order according to the similarity, so that the most relevant legal texts are displayed first, saving users time in screening information and improving search efficiency.

[0035] Personalized service: Using collaborative filtering and content-based recommendation algorithm dual-engine drive, combined with dynamically updated user interest models, accurately push legal information that meets the personalized needs of users, meet the diverse search demands of different users in different scenarios, and enhance user experience.

[0036] Continuous optimization: With the help of a regularly updated legal knowledge base, a dynamically adjusted user interest model, and a recommendation algorithm optimized by feedback, the system can keep up with the development of international commercial law and changes in user behavior, always maintain good search performance, and provide guarantees for long-term and stable legal search services. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a flow chart of an international commercial law retrieval method based on artificial intelligence of the present invention.

[0038] Figure 2 It is a system block diagram of an international commercial law retrieval system based on artificial intelligence of the present invention.

[0039] Figure 3 The present invention is a similarity calculation flow chart of an international commercial law retrieval method based on artificial intelligence. DETAILED DESCRIPTION

[0040] The following is a further detailed description of an international commercial law search system and method based on artificial intelligence of the present invention in conjunction with the accompanying drawings.

[0041] Combined with Figure 1-3 , the present invention is introduced in detail.

[0042] S1. Feature word extraction: Through lexical analysis and part-of-speech tagging in natural language processing technology, feature words are accurately extracted from the search request input by the user; at the same time, combined with semantic understanding, synonyms and near-synonyms are normalized. On this basis, a deep learning model is introduced to capture potential key feature words in complex contexts; combined with deep semantic understanding, synonyms, near-synonyms, and polysemous words are intelligently normalized. Let the user's search request text be T, and after deep preprocessing, the word set W = w 1 ,w 2 ,…,w n , through the pre-trained deep semantic knowledge base and the dynamically updated semantic mapping function f s Normalize the semantics of each word and get the feature word set F = f s (w 1 ),fs (w 2 ),…f s (w n ), where f s Based on the pre-trained semantic knowledge base, synonym conversion and polysemy analysis are realized. For professional legal terms, a professional term dictionary is established and updated in real time, combined with semantic normalization processing to further improve the accuracy of feature word extraction.

[0043] S2. Similarity calculation: The vector space model is combined with the cosine similarity algorithm to calculate the similarity and obtain the similarity between the feature words of each legal text in the legal knowledge base and the search request. In the specific calculation, the feature words of each legal text in the legal knowledge base and the search request are vectorized; the TF-IDF algorithm is used to assign weights to each feature word and construct a text vector. Then the feature word f in the legal text L i The TF-IDF value is calculated as TF(f i ,L)×IDF(f i ), where TF(f i ,L) is f i The frequency of words in L, N is the total number of legal texts, DF(f i ) is the one containing f i The number of legal texts; for the retrieval request feature word vector V q =(v q1 ,v q2 ,…,v qm ) and the legal text vector V l =(v l1 ,v l2 ,…,v lm ), their cosine similarity calculation formula is:

[0044]

[0045] The larger the cosine similarity value, the higher the similarity. For the reference clauses and explanatory text information in the legal text, hierarchical weighting processing is adopted, and different weights are assigned according to the closeness of their relationship with the core legal provisions, so that the similarity calculation is more in line with the actual value of the legal text.

[0046] S3. Sorting search results: Sort the search results according to similarity, and give priority to displaying the most relevant legal texts. That is, according to the similarity values ​​calculated in step S2, sort the search results in descending order, and put the legal texts with high similarity in the front and display them to users first.

[0047] S4. Personalized recommendation: Use collaborative filtering algorithm combined with content-based recommendation algorithm to collect users' historical search records and current search requests, build user interest model, and recommend legal information that users may be interested in. Collaborative filtering finds user groups with similar search behaviors and mines relevant legal information that they have searched but the current user has not searched; content-based recommendation analyzes the correlation between the feature words of the current search request and the content of the unsearched text in the legal knowledge base. The specific method of building the user interest model is as follows: The historical search feature word set of user u is H u =h u1 ,h u2 ,…,h uk , the current search request feature word set is F, for the legal text L j , collaborative filtering similarity calculation is sim cf (u,L j ), calculated based on content similarity as sim cb (F,L j ), comprehensive recommendation score R = λsim cf (u,L j )+(1-λ)sim cb (F,L j ), where λ balances the weights of the two algorithms. During the recommendation process, the user interest model is regularly updated and verified, and the model parameters are dynamically adjusted according to the recent changes in user search behavior and dynamic updates of industry regulations. At the same time, a recommendation effect feedback mechanism is established to collect user selection behavior data on recommended content and reversely optimize the recommendation algorithm.

[0048] The present invention also provides an artificial intelligence-based international commercial law search system for implementing the above method, and the system is composed as follows:

[0049] User interface: used for interaction between users and the system, inputting legal search information, displaying search results and case analysis content. The design is simple and intuitive, making it convenient for different user groups to operate.

[0050] Legal Knowledge Base: A database containing international commercial law regulations, legal interpretations, and legal cases. It is regularly updated and maintained to ensure the timeliness and accuracy of the data, providing a solid data foundation for retrieval.

[0051] Natural language processing module: connects the user interface with the legal knowledge base, and is used to convert the text of the legal knowledge base and the search text entered by the user into a computer-understandable format, including text preprocessing, vocabulary embedding, dependency parsing and named entity recognition, to achieve in-depth processing of natural language and provide accurate input for subsequent steps.

[0052] Machine learning module: connects to the legal knowledge base and is used to learn legal rules and principles from the legal knowledge base, including feature extraction, model training and evaluation, continuously optimizes model performance and improves retrieval accuracy. In the feature extraction stage, the key features are screened by combining the text preprocessing results with the deep learning model output.

[0053] Legal reasoning engine: connects the legal knowledge base and machine learning module, and is used to perform legal reasoning based on the legal rules and principles in the legal knowledge base and machine learning module, generate legal advice and answers for the search text entered by the user, and assist users in understanding and applying legal knowledge.

[0054] The specific implementation process of the international commercial law retrieval system and method based on artificial intelligence of the present invention is as follows:

[0055] Building an international commercial law search system:

[0056] The AI-based international commercial law search system consists of a user interface, a legal knowledge base, a natural language processing module, a machine learning module, and a legal reasoning engine. The modules work together to provide a full-process intelligent service from user input of search requests to accurate output of legal search results and recommendations.

[0057] The user interface is located at the front end of the system as a window for users to interact with the system. Its simple and intuitive design facilitates operation by users of different professional backgrounds. Users enter legal search information here, and the system responds and displays search results and case analysis content in real time. In terms of specific construction, it is built using HTML, CSS and JavaScript technologies. The search box is set up to facilitate users to enter text. The result display area presents legal provisions, cases and other information in a clear layout. The operation button design conforms to human-computer interaction habits. With the help of AJAX technology, it ensures smooth asynchronous interaction with the back-end server, avoids page freezes, and improves user experience.

[0058] The legal knowledge base is at the bottom of the system and is the data foundation of the entire retrieval system. It covers a wealth of information such as international commercial law regulations, legal interpretations, and legal cases. Its construction process is rigorous, and it is strictly controlled from data collection, cleaning to input. Its construction method is as follows:

[0059] Data collection: Organize a professional legal team to collect relevant information on international commercial law from multiple channels such as international authoritative legal documents, official commercial laws and regulations of various countries, and well-known legal case libraries, covering various treaties, laws, judicial interpretations, and typical commercial dispute cases.

[0060] Data cleaning: Remove duplicate, invalid or expired data, and standardize text formats, such as unified coding, paragraph format, etc., to ensure data quality.

[0061] Data entry: Classify and enter the cleaned data into the MySQL database, establish a reasonable data table structure, and associate laws, regulations, interpretations and cases to facilitate retrieval.

[0062] The natural language processing module is located between the user interface and the legal knowledge base, and undertakes the key task of converting natural language into a computer-understandable format, providing accurate input for subsequent retrieval processes.

[0063] The machine learning module is closely connected to the legal knowledge base, learning rules and principles from massive legal texts, continuously optimizing its own model performance, improving retrieval accuracy, and providing intelligent support for core links such as similarity calculation and feature extraction.

[0064] The legal reasoning engine relies on the results of the legal knowledge base and machine learning modules to perform deep reasoning on user search texts, generate professional and easy-to-understand legal advice and answers, and assist users in understanding and applying legal knowledge. It is a key link in the system's output of professional legal guidance.

[0065] An example of using the system to conduct an international commercial law search is as follows:

[0066] Feature word extraction:

[0067] When the user enters a search request in the interface, the natural language processing module first starts the lexical analysis and part-of-speech tagging functions, relying on Python's NLTK, Spacy and other toolkits to split the text into words and mark the parts of speech. For example, for the input "definition of liability for breach of contract in international sales of goods", it can quickly identify "international" as an adjective, "goods", "sales", "contract" as nouns, and "breach of contract", "compensation", "liability definition" as verb phrases, and preliminarily extract key content words as basic feature words.

[0068] Next, combined with the semantic understanding module, the pre-trained semantic knowledge base and the dynamically updated semantic mapping function f s , and normalize synonyms and near-synonyms. Assuming that there is a synonymous mapping relationship between "breach of contract" and "violation of contract agreement" in the semantic knowledge base, they will be uniformly converted into standard terms to ensure comprehensive retrieval.

[0069] Furthermore, a deep learning model is introduced to capture potential key feature words in complex contexts. For example, when processing long text search requests involving international goods sales disputes, the deep learning model can understand the context, dig out implicit key legal concepts, and add them to the feature word set. Taking a complex case description as an example, the model may capture the potential key feature word "force majeure" in "the impact of force majeure factors on delivery time", making feature word extraction more accurate.

[0070] For professional legal terms, the established professional terminology dictionary is updated in real time and works in conjunction with semantic normalization processing. When encountering professional terms such as "CIF trade terms", the dictionary quickly matches the standard interpretation, combines semantic processing, and accurately incorporates it as a key feature word, improving the overall feature word extraction accuracy.

[0071] Similarity calculation:

[0072] First, the feature words of each legal text and search request in the legal knowledge base are vectorized. Taking the feature word set extracted from the search request "Definition of liability for breach of contract in international sales of goods" and a certain legal text as an example, a word vector model (such as GloVe) is used to map each feature word into a vector representation of fixed dimension. Assuming that the vector of "international" is [0.1, 0.2, 0.3], the vector of "sales of goods" is [0.4, 0.5, 0.6], etc., the search request feature word vector V is constructed. q Similarly, the legal text is vectorized to obtain V l .

[0073] Next, use the TF-IDF algorithm to assign weights to each feature word and construct a text vector. Assume that there are 1,000 legal texts in the legal knowledge base (i.e., N = 1,000), and the feature word "default" appears 5 times in a specific legal text L (TF = 5), and there are 200 legal texts containing "default" (DF = 200), then IDF = log (1000 / 200) = 1.609, and the TF-IDF value of this feature word in text L is 5 × 1.609 = 8.045. In this way, the TF-IDF value is calculated for each feature word, and a text vector that reflects the importance of the feature word is constructed.

[0074] Finally, according to the cosine similarity calculation formula Calculate similarity. Suppose the calculated cosine similarity between a legal text and a search request is 0.8, indicating that the two are highly correlated; while the calculated result for another legal text is 0.3, the correlation is low. For the reference clauses and explanatory text information in the legal text, hierarchical weighting is used. For example, the weight of the core legal text is set to 1, the weight of the closely related explanatory text is set to 0.6, and the weight of the more distant reference clause is set to 0.3, so that the similarity calculation is more in line with the actual value of the legal text.

[0075] The calculation example is as follows:

[0076] Assume that in a search scenario, the user enters the search request "the seller's liability for delayed delivery in an international sales contract". After natural language processing and feature word extraction, the feature word set F = {"international sales", "seller's delayed delivery", "compensation liability"} is obtained.

[0077] There are three legal documents L1, L2, and L3 in the legal knowledge base. Vectorization and TF-IDF weight calculation are performed on them:

[0078] For L1, the total number of words is 100, of which "international sale of goods" appears 8 times (TF=8), there are 2 legal documents containing this word (DF=2), N=3, then IDF=log(3 / 2)=0.176, and its TF-IDF value is 8×0.176=1.408; similarly, the TF-IDF value of "delayed delivery by the seller" in L1 is 6×log(3 / 1)=6.588; "liability for compensation" is 4×log(3 / 2)=0.704, constructing the text vector V of L1 l1 .

[0079] Similarly, calculate the text vector V of L2 and L3 l2 、V l3 .

[0080] Retrieve request feature word vector V q Calculation: Assuming that after the word vector model conversion, the vector of "international goods trading" is [0.2, 0.3, 0.1], the vector of "seller's delayed delivery" is [0.4, 0.2, 0.3], and the vector of "compensation liability" is [0.1, 0.3, 0.2], we get V q .

[0081] Calculate cosine similarity:

[0082] For L1 and V q , calculated according to the formula Substituting the vector value and the calculated TF-IDF weight value, after detailed calculation (the specific matrix calculation steps are omitted here), we can get cosθ1=0.75.

[0083] The same method is used to calculate L2 and V q cosθ2=0.45,L3 and V q cosθ3=0.62.

[0084] Search result sorting: According to the similarity values ​​calculated above, L1, L2, and L3 are sorted in descending order, i.e., L1, L3, and L2. L1 is displayed to the user first because it has the highest similarity with the search request.

[0085] Personalized recommendations:

[0086] Assuming that the user's historical search records contain relevant feature words such as "international cargo transportation insurance" and "handling of goods rejected by the buyer", the system combines the current search request and uses collaborative filtering and content-based recommendation algorithms. For example, if the "Case Analysis of the Seller's Quality Guarantee Liability in International Sales Contracts" that has been searched by similar user groups is found, and the "Special Provisions on Extension of Delivery Period in International Commercial Law" that has not been searched in the legal knowledge base but is related to the current feature word is analyzed based on the content, the comprehensive recommendation score R is calculated based on the set weight λ (assuming λ=0.6), and these legal information that may be of interest to the user is recommended.

[0087] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An international commercial law search method based on artificial intelligence, characterized by: The following steps are included: S1. Through lexical analysis and part-of-speech tagging in natural language processing technology, feature words are accurately extracted from the search request input by the user; at the same time, synonyms and near-synonyms are normalized in combination with semantic understanding; S2. Use the vector space model combined with the cosine similarity algorithm to calculate the similarity and obtain the similarity between the feature words of each legal text in the legal knowledge base and the search request; S3. Sort the search results according to similarity, and give priority to displaying legal texts with high similarity; S4. Use collaborative filtering algorithms combined with content-based recommendation algorithms to collect users’ historical search records and current search requests, build user interest models, and recommend legal information that users may be interested in.

2. The international commercial law search method based on artificial intelligence according to claim 1, characterized in that: In step S1, a deep learning model is also introduced to capture potential key feature words in complex contexts; at the same time, combined with deep semantic understanding, intelligent normalization processing is performed on synonyms, near-synonyms, and polysemous words. The normalization processing method is as follows: Assume that the user's search request text is T, and after deep preprocessing, we get the word set W = w1, w2, ..., w n , through the pre-trained deep semantic knowledge base and the dynamically updated semantic mapping function f s Normalize the semantics of each word and get the feature word set F = f s (w1), f s (w2),…f s (w n ), where f s Based on the pre-trained semantic knowledge base, synonym conversion and polysemy analysis operations of words are realized.

3. The international commercial law search method based on artificial intelligence according to claim 2, characterized in that: In step S2, the specific method for calculating the similarity between each legal text in the legal knowledge base and the feature word of the search request is as follows: Vectorize the feature words of each legal text and search request in the legal knowledge base; use the TF-IDF algorithm to assign weights to each feature word and construct a text vector; then the feature word f in the legal text L i The TF-IDF value is calculated as TF(f i , L)×IDF(f i ), where TF(f i , L) is f i The frequency of words in L, N is the total number of legal texts, DF(f i ) is the one containing f i Number of legal texts; For the retrieval request feature word vector V q =(v q1 ,v q2 ,…,v qm ) and the legal text vector V l =(v l1 ,v l2 ,…,v lm ), their cosine similarity calculation formula is: A larger value of the cosine similarity indicates a higher similarity.

4. The international commercial law search method based on artificial intelligence according to claim 3 is characterized by: In step S3, the search results are sorted in descending order according to the similarity values ​​calculated in step S2, and the legal texts with high similarity are prioritized and displayed to the user.

5. The international commercial law search method based on artificial intelligence according to claim 4, characterized in that: In step S4, the collaborative filtering algorithm is combined with the content-based recommendation algorithm to collect the user's historical search records and build a user interest model; collaborative filtering finds user groups with similar search behaviors and mines relevant legal information that they have searched but the current user has not searched; content-based recommendation analyzes the correlation between the feature words of the current search request and the content of the unsearched text in the legal knowledge base; the specific method of building the user interest model is as follows: The historical search feature word set of user u is H u =h u1 ,h u2 ,…,h uk , the current search request feature word set is F, for the legal text L j , collaborative filtering similarity calculation is sim cf (u,L j ), calculated based on content similarity as sim cb (F,L j ), comprehensive recommendation score R = λsim cf (u,L j )+(1-λ)sim cb (F,L j ), where λ balances the weights of the two algorithms.

6. The international commercial law search method based on artificial intelligence according to claim 2, characterized in that: In the process of capturing feature words, a professional terminology dictionary is established for professional legal terms and updated in real time, combined with semantic normalization processing.

7. The international commercial law search method based on artificial intelligence according to claim 3 is characterized by: In the process of calculating the similarity of feature words, hierarchical weighting processing is adopted for the quoted clauses and explanatory text information in the legal text, and different weights are assigned according to the degree of their connection with the core legal provisions.

8. The international commercial law search method based on artificial intelligence according to claim 5, characterized in that: In the process of recommending legal information that users may be interested in, the user interest model is regularly updated and verified, and the model parameters are dynamically adjusted based on recent changes in user search behavior and dynamic updates of industry regulations. At the same time, a recommendation effect feedback mechanism is established to collect user selection behavior data on recommended content and reversely optimize the recommendation algorithm.

9. An international commercial law search system based on artificial intelligence, characterized by: include User interface: used for interaction between users and the system, inputting legal search information, and displaying search results and case analysis content; Legal Knowledge Base: A database containing international commercial law regulations, legal interpretations, and legal cases; Natural language processing module: connects the user interface with the legal knowledge base and is used to convert the text of the legal knowledge base and the search text entered by the user into a format that can be understood by a computer, including text preprocessing, vocabulary embedding, dependency parsing and named entity recognition; Machine learning module: connects to the legal knowledge base and is used to learn legal rules and principles from the legal knowledge base, including feature extraction, model training and evaluation; Legal reasoning engine: connects the legal knowledge base and machine learning module, and is used to perform legal reasoning based on the legal rules and principles in the legal knowledge base and machine learning module, and generate legal advice and answers for the search text entered by the user.

Images