Retrieval method for aggregating multi-dimensional retrieval strategies

Abstract

The invention relates to the technical field of information retrieval, and discloses a retrieval method for aggregating multi-dimensional retrieval strategies. When retrieval is carried out, a retrieval word input by a user is obtained; based on the retrieval word, performing retrieval by using each retrieval strategy to obtain a retrieval result of each retrieval strategy; inputting the retrieval word into a semantic judgment model to obtain a retrieval strategy weight adjustment parameter; calculating according to the retrieval strategy weight adjustment parameters to obtain the final weight of each retrieval strategy; and according to the final weight of each retrieval strategy, sorting the retrieval results of each retrieval strategy to obtain a final retrieval result. According to the method, the weights of full-text retrieval, vectorization retrieval and graph database retrieval are dynamically adjusted according to semantic features of user retrieval words through retrieval strategy weight adjustment parameters accurately obtained by using a semantic judgment model, and it is ensured that optimal retrieval results can be obtained in different query scenes.

Description

Technical Field

[0001] The present invention relates to the field of information retrieval technology, and in particular to a retrieval method that aggregates multi-dimensional retrieval strategies. Background Art

[0002] In the era of information explosion, document retrieval systems play a vital role. They help users quickly locate the required content from massive amounts of information. Document retrieval strategies include full-text retrieval technology, vectorized retrieval, graph database retrieval, etc. Among them, traditional document retrieval methods mainly rely on full-text retrieval technology, that is, by scanning all the text in the document to find text fragments that match the user's query keywords. This method is simple and direct and is suitable for most text-based retrieval needs. Vectorized retrieval has received widespread attention as an emerging retrieval method. This method first converts the text into a vector representation in a high-dimensional space, and then uses a machine learning algorithm to calculate the similarity between texts, thereby achieving more intelligent and accurate retrieval. Compared with traditional full-text retrieval, vectorized retrieval can capture information at the semantic level and improve the ability to understand complex situations such as synonyms and contextual relationships. Graph database retrieval has also become an important part of the modern information retrieval field. It forms a network structure (or graph) by establishing associations between entities, and then can more deeply explore the potential connections between data. For example, in academic literature retrieval, graph database retrieval can help discover the working connections between different researchers or track the development of a certain scientific research result.

[0003] However, with the surge in data volume and users' increasing requirements for the accuracy of retrieval results, the limitations of the current single retrieval strategy have gradually been exposed, and it is still difficult to provide the best solution when faced with complex retrieval tasks.

[0004] Therefore, there is an urgent need for a retrieval method that aggregates multi-dimensional retrieval strategies, which can effectively integrate multiple retrieval strategies and improve retrieval efficiency and accuracy. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a retrieval method for aggregating multi-dimensional retrieval strategies, including: Get the search term entered by the user; Based on the search terms, use each search strategy to perform a search and obtain search results of each search strategy; Input the search term into the semantic judgment model to obtain the search strategy weight adjustment parameter; Calculating according to the retrieval strategy weight adjustment parameters to obtain the final weight of each retrieval strategy; According to the final weights of the search strategies, the search results of the search strategies are sorted to obtain the final search results.

[0006] Furthermore, before obtaining the user's search terms, data processing and storage are also included, specifically: For each search strategy, the text is parsed and processed respectively to obtain the text format required for the search process of each search strategy and store it.

[0007] Furthermore, each search strategy includes text search, vectorized search and graph database search.

[0008] Furthermore, before performing a graph database search, it also includes performing semantic calculations on the search terms to obtain the entity relationship strength and entity quantity of the search terms.

[0009] Furthermore, the retrieval strategy weights obtained by the semantic judgment model adjust the clarity, semantic complexity and semantic richness of the keywords in the parameter retrieval terms.

[0010] Furthermore, the final weight of each search strategy is obtained by calculating according to the search strategy weight adjustment parameter, which is specifically: The adjustment weight of each search strategy is calculated according to the search strategy weight adjustment parameter, and the calculation formula is: W` text =α× S text +α× (1-S complexity ) ; W` vector =γ× S complexity +δ× S semantic ; W` graph =ε× S relation +ζ×S entity ; Among them, W` text is the adjustment weight for text retrieval, W` vector is the adjustment weight for vectorized retrieval, W` graph is the adjustment weight for graph database retrieval, S text Indicates the clarity of the search term, S complexity represents the semantic complexity of the search term, S semantic Indicates the semantic richness of the search term, S relation Indicates the entity relationship strength of the search term, S entity It represents the number of entities of the search term, and α, β, γ, δ, ε, and ζ are adjustment coefficients.

[0011] The adjusted weights of each search strategy are normalized to obtain the final weight of each search strategy.

[0012] Furthermore, the normalized calculation formula is as follows: W text= ; W vector= ; W graph= ; Among them, W text is the final weight of text retrieval, W vector is the final weight of vectorized retrieval, W graph The final weight retrieved for the graph database.

[0013] On the other hand, the present application provides a retrieval system for aggregating multi-dimensional retrieval strategies, which executes any of the above-mentioned retrieval methods for aggregating multi-dimensional retrieval strategies, wherein the retrieval system comprises: a retrieval term acquisition module, a retrieval module, a retrieval strategy weight acquisition module, and a result output module; The search term acquisition module is used to acquire the search term input by the user; The search module uses each search strategy to perform search based on the search term to obtain search results of each search strategy; The retrieval strategy weight acquisition module is used to input the search term into the semantic judgment model to obtain the retrieval strategy weight adjustment parameter; calculate according to the retrieval strategy weight adjustment parameter to obtain the final weight of each retrieval strategy; The result output module is used to sort the search results of each search strategy according to the final weight of each search strategy to obtain the final search result.

[0014] The embodiments of the present invention have the following technical effects: When the present invention performs a search, the search terms are input into the semantic judgment model to obtain the search strategy weight adjustment parameters; then the search strategy weight adjustment parameters are calculated to obtain the final weights of each search strategy; thereby the search results of each search strategy are sorted to obtain the final search results. By using the search strategy weight adjustment parameters accurately obtained by the semantic judgment model, the weights of full-text search, vectorized search and graph database search are dynamically adjusted according to the semantic features of the user's search terms, ensuring that the best search results can be obtained in different query scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 It is a flow chart of a retrieval method for an aggregated multi-dimensional retrieval strategy provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0018] At present, the single full-text search technology has gradually exposed its limitations, and there are also some significant problems in practical applications: although full-text search can cover a wide range of document content, it is easy to produce a large number of irrelevant results, especially when the query conditions are not specific enough, it is difficult to ensure that the retrieved results fully meet the user's expectations; vectorized search, although there have been breakthroughs in semantic understanding, but because of the need to map the text to a high-dimensional space, this not only increases the computational cost, but may also cause some subtle differences to be ignored, affecting the final retrieval accuracy; graph database retrieval has weak support for unstructured data, and building and maintaining complex graph structures is itself a challenge, especially when processing large-scale data sets, performance may be affected.

[0019] However, the existing technology usually focuses on the use of a certain retrieval method alone, and lacks the effective integration of multiple retrieval technologies. This means that even if each technology has reached a high level in its respective field, it is still difficult to provide the best solution when faced with complex retrieval tasks. Therefore, the key issue is how to integrate the advantages of full-text retrieval, vectorized retrieval and graph database retrieval to improve retrieval efficiency and accuracy. The present invention provides a retrieval method that aggregates multi-dimensional retrieval strategies, aiming to significantly improve the efficiency and accuracy of massive document retrieval by integrating the three methods of full-text retrieval, vectorized retrieval and graph database retrieval. In view of the limitations of a single retrieval method in the prior art, the method first uses a semantic judgment model to accurately obtain the retrieval strategy weight adjustment parameters, and then dynamically adjusts the weights of full-text retrieval, vectorized retrieval and graph database retrieval according to the semantic features of the user's search terms, to ensure that the best retrieval results can be obtained in different query scenarios.

[0020] Semantic analysis based on large models: Traditional weight adjustment methods usually rely on static rules or simple keyword matching, which cannot accurately understand the semantics of the query. The present invention introduces a semantic judgment model to perform in-depth semantic analysis on the query, capture the complex semantic relationships in the query, and thus adjust the weight more accurately.

[0021] Dynamic weight adjustment: The weight adjustment algorithm proposed in the present invention can dynamically adjust the weights of full-text search, vectorized search and graph database search according to the semantic features of the query, ensuring that the best search results can be obtained in different query scenarios. This dynamic adjustment mechanism avoids the limitations of fixed weights in traditional methods.

[0022] Flexibility and scalability: The weight adjustment algorithm of the present invention has high flexibility and scalability. By adjusting the coefficients in the formula, the system can adapt to different application scenarios and user needs. In addition, with the continuous evolution of the semantic judgment model, the system can introduce more semantic features to further improve the accuracy of weight adjustment.

[0023] Figure 1 is a flowchart of a retrieval method for an aggregated multi-dimensional retrieval strategy provided by an embodiment of the present invention, such as Figure 1 As shown, the retrieval method specifically includes the following steps: S1: Get the search term entered by the user; In some embodiments, before obtaining the user's search terms, data processing and storage are also included, specifically: For each search strategy, the text is parsed and processed respectively to obtain the text format required for the search process of each search strategy and store it.

[0024] Exemplarily, all texts are parsed using text parsing technology to extract key information such as title, text, author, etc., and preprocessing is performed, such as removing stop words, segmenting words, etc.

[0025] Then store the preprocessed text data: Json format storage: The parsed text data is converted into Json format and stored in the corresponding database for subsequent full-text retrieval. This format can quickly locate the specific location in the document.

[0026] Vectorized format storage: Use document processing embedded models to convert text into high-dimensional vectors and store these vectors in the corresponding database to support vectorized retrieval.

[0027] Knowledge graph storage: Extract knowledge graph according to knowledge graph semantic standards, construct the association relationship between documents into a graph structure, and store it in the corresponding database of Ant for graph database retrieval.

[0028] S2: Based on the search terms, use each search strategy to perform a search and obtain search results of each search strategy; In some embodiments, each search strategy includes text search, vectorized search, and graph database search.

[0029] Select the required search method according to the specific field of the search needs. For example, this application is mainly aimed at the search needs in the field of water conservancy projects. There are many special terms in the information in this field, so three search methods are selected: text search, vectorized search, and graph database search. If there are other needs, you can select the search method that needs to be integrated according to the specific search needs.

[0030] Full-text search: After the user enters the query search term, the search engine conducts a comprehensive scan of the text content stored in the Json format, extracts and matches the keywords, and returns a list of documents related to the query keywords.

[0031] Vectorized retrieval: After the user enters the query text, the document processing embedded model is used to convert the query text into a vector, and the similarity is calculated in the corresponding database to return a list of documents that are most similar to the query vector.

[0032] Graph database retrieval: After the user enters the query conditions, the corresponding graph database performs path search based on the stored knowledge graph and returns documents related to the query conditions and their associated information.

[0033] In this way, the search results of each search strategy are obtained.

[0034] In some embodiments, before searching the graph database, semantic calculation is performed on the search terms to obtain the entity relationship strength and entity quantity of the search terms, which are used to participate in the subsequent search strategy weight adjustment, improve the accuracy of the weight adjustment, and assist in the correction of the final search results. relation The higher the value, the stronger the relationship involved in the search term. The number of entities S of the search term entity The higher the value, the more entities the search term involves. The length of the search term is not limited, so the entity and relationship composition can be determined based on semantic calculation to obtain the entity relationship strength and entity quantity of the search term. S3: Input the search term into the semantic judgment model to obtain the search strategy weight adjustment parameter.

[0035] In traditional retrieval systems, the weight distribution is usually static, that is, the weights of full-text retrieval, vectorized retrieval, and graph database retrieval are fixed and cannot be dynamically adjusted according to the specific semantics of the query. The present invention performs semantic analysis on user queries through a semantic judgment model, which can more accurately understand the query intent and dynamically adjust the weights of different retrieval technologies.

[0036] For example, after the user inputs a search term, the system first uses a semantic judgment model to perform semantic analysis on the query. The semantic judgment model can capture the keywords and phrases in the query and the semantic relationships between them.

[0037] For example, for the query "application of artificial intelligence in medical care", the semantic judgment model can recognize that "artificial intelligence" and "medical care" are core keywords, and understand that "application" is the relationship between these two keywords.

[0038] The semantic judgment model will output the semantic feature vector of the query, which contains the user's query intention. These semantic feature vectors will serve as the basis for subsequent weight adjustment.

[0039] In some embodiments, the retrieval strategy weight adjustment parameters, the clarity of keywords in the search terms, the semantic complexity and the semantic richness obtained by the semantic judgment model include the clarity of the search terms S text The higher the value, the clearer the search term is. The semantic complexity of the search term S complexity The higher the value, the more complex the semantics of the search term. The semantic richness S of the search term semantic The higher the value, the richer the search term information.

[0040] Exemplary semantic judgment models are large language models (LLMs), which can understand and generate complex content in natural language by pre-training on massive amounts of text data. They have strong language representation capabilities, can capture the deep structure, semantic relationships, and contextual information of language, and show excellent performance in various natural language processing tasks. They have strong semantic understanding capabilities: large language models have accumulated rich language knowledge and semantic understanding capabilities through large-scale corpus training; they can identify semantic relationships between words, phrases, and sentences, and accurately capture the subtle differences in language; strong generalization capabilities: because large language models are trained on large amounts of text data, they have strong generalization capabilities. Even in the face of unseen queries or tasks, they can give reasonable answers or solutions; flexibility and scalability: large language models usually have good flexibility and scalability, and can be fine-tuned or adjusted according to specific task requirements to adapt to different application scenarios.

[0041] The large language model can make predictions based on the input search terms, mainly relying on the deep neural network structure trained inside it. The specific process is as follows: Input processing: First, the model will preprocess the input search terms, including word segmentation, stop word removal, part-of-speech tagging, etc., for subsequent processing.

[0042] Encoding representation: The model then converts the processed search terms into vector representations that contain the semantic information of the search terms. This is usually achieved through the model's embedding layer.

[0043] Deep computing: Next, the model will perform multiple layers of computing in its deep neural network structure, using different attention mechanisms, fully connected layers, and other structures to deeply mine and represent the semantics of the search terms.

[0044] Output prediction: Finally, the model will predict the input search terms based on its internal training knowledge and weights, and output information such as the clarity, semantic complexity, and semantic richness of the keywords in the search terms. This information is usually obtained through a specific output layer or post-processing step based on the model's deep understanding and representation of the input search terms.

[0045] Because large language models have powerful semantic understanding and generalization capabilities, they can make accurate predictions based on the input search terms, obtain key information such as the clarity, semantic complexity, and semantic richness of the keywords in the search terms, thereby helping to improve and optimize the performance and effectiveness of the retrieval system.

[0046] This application uses the search terms entered by the user as the input of the semantic judgment model, mainly based on the following considerations: Directly reflects user intent: The search term is the core expression of the user's query, which directly reflects the user's search needs and intent. Using the search term as input enables the semantic judgment model to more accurately understand the user's query purpose, thereby providing personalized weight adjustment solutions for different queries.

[0047] Simplify input complexity: Compared with complete documents or paragraphs, search terms are usually shorter and clearer. This reduces the input complexity of the semantic judgment model, enabling it to process input data more efficiently, quickly output weight adjustment parameters, meet the needs of quickly outputting search results, and improve search efficiency.

[0048] Avoid information redundancy: In retrieval scenarios, users often focus on specific information points rather than the content of the entire document. Selecting search terms as input can avoid introducing information irrelevant to the query, reduce noise interference, and improve the accuracy and efficiency of weight adjustment.

[0049] Adapt to various search strategies: Different search strategies (such as full-text search, vectorized search, and graph database search) have different processing methods and focuses on input data. As a common input starting point, search terms can provide a unified and concise semantic basis for these strategies, facilitating subsequent weight adjustment and result fusion.

[0050] Facilitates model training and optimization: Using search terms as input simplifies the training process of the semantic judgment model. Since the search terms are relatively short and clear, it is easier for the model to learn the mapping relationship between keywords and weight adjustment parameters. In addition, this also makes it easier to tune and verify the model to ensure its stability and accuracy in different query scenarios.

[0051] In summary, the selection of search terms as the input of the semantic judgment model is based on its advantages of directly reflecting user intentions, simplifying input complexity, avoiding information redundancy, adapting to a variety of retrieval strategies, and facilitating model training and optimization.

[0052] S4: Calculating according to the retrieval strategy weight adjustment parameters to obtain the final weight of each retrieval strategy; In some embodiments, the final weight of each search strategy is obtained by calculating according to the search strategy weight adjustment parameter, specifically: S41: Calculate according to the retrieval strategy weight adjustment parameter to obtain the adjustment weight of each retrieval strategy, and the calculation formula is: W` text =α× S text +α× (1-S complexity ) ; W` vector =γ× S complexity +δ× S semantic ; W` graph =ε× S relation +ζ×S entity ; Among them, W` text is the adjustment weight for text retrieval, W` vector is the adjustment weight for vectorized retrieval, W` graph is the adjustment weight for graph database retrieval, S text Indicates the clarity of the search term. The higher the value, the clearer the search term. complexity Indicates the semantic complexity of the search term. The higher the value, the more complex the semantics of the search term. semantic Indicates the semantic richness of the search term. The higher the value, the richer the search term information. relation Indicates the entity relationship strength of the search term. The higher the value, the stronger the relationship involved in the search term. entityIt indicates the number of entities of the search term. The higher the value, the more entities the search term involves. α, β, γ, δ, ε and ζ are adjustment coefficients.

[0053] Exemplarily, the adjustment coefficient α is used to control the influence of the clarity of keywords on the full-text search weight. The default value is set to 0.5, indicating that the full-text search weight is moderate when the keywords are clear. β controls the negative influence of semantic complexity on the full-text search weight. The default value is set to 0.3, indicating that the full-text search weight is appropriately reduced when the semantic complexity is high. γ controls the influence of semantic complexity on the vectorized search weight. The default value is set to 0.4, indicating that the vectorized search weight is increased when the semantic complexity is high. δ controls the influence of semantic richness on the vectorized search weight. The default value is set to 0.6, indicating that the vectorized search weight is higher when the semantics are rich. ε controls the influence of the strength of entity relationships on the graph database search weight. The default value is set to 0.5, indicating that the graph database search weight is moderate when the entity relationship is strong. ζ controls the influence of the number of entities on the graph database search weight. The default value is set to 0.4, indicating that the graph database search weight is appropriately increased when the number of entities is large.

[0054] In some embodiments, based on considerations of program code security, the initial weight of each search strategy is first preset as the initial value of the adjusted weight of each search strategy, and the sum of each initial weight is 1; after the adjusted weight of each search strategy is obtained by calculation in S41, the calculation result of S41 is used as the final result of the adjusted weight of each search strategy.

[0055] S42: Normalize the adjusted weights of each search strategy to obtain the final weight of each search strategy.

[0056] The adjusted weights of each search strategy need to be normalized to ensure that the sum of each weight is 1. Therefore, in some embodiments, the normalized calculation formula is as follows: W text= ; W vector= ; W graph= ; Among them, W text is the final weight of text retrieval, W vector is the final weight of vectorized retrieval, W graph The final weight retrieved for the graph database.

[0057] S5: Sort the search results of each search strategy according to the final weight of each search strategy to obtain the final search result.

[0058] The retrieval method provided in this application improves retrieval accuracy: By dynamically adjusting weights, the system can prioritize the most appropriate retrieval technology based on the semantics of the query, thereby improving the accuracy and relevance of the retrieval results.

[0059] Adapt to complex query scenarios: The present invention can effectively cope with complex query scenarios, such as queries with multiple keywords, complex semantics, and queries involving entity relationships, ensuring that the best search results can be obtained in different scenarios.

[0060] Reduce manual intervention: Through an automated weight adjustment mechanism, the system can reduce manual intervention and lower operation and maintenance costs while maintaining the long-term competitiveness of the system.

[0061] On the other hand, the present application provides a retrieval system for aggregating multi-dimensional retrieval strategies, which executes any of the above-mentioned retrieval methods for aggregating multi-dimensional retrieval strategies, wherein the retrieval system comprises: a retrieval term acquisition module, a retrieval module, a retrieval strategy weight acquisition module, and a result output module; The search term acquisition module is used to acquire the search term input by the user; The search module uses each search strategy to perform search based on the search term to obtain search results of each search strategy; The retrieval strategy weight acquisition module is used to input the search term into the semantic judgment model to obtain the retrieval strategy weight adjustment parameter; calculate according to the retrieval strategy weight adjustment parameter to obtain the final weight of each retrieval strategy; The result output module is used to sort the search results of each search strategy according to the final weight of each search strategy to obtain the final search result.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A retrieval method for aggregating multi-dimensional retrieval strategies, characterized in that: include: Get the search term entered by the user; Based on the search terms, use each search strategy to perform a search and obtain search results of each search strategy; Input the search term into the semantic judgment model to obtain the search strategy weight adjustment parameter; Calculating according to the retrieval strategy weight adjustment parameters to obtain the final weight of each retrieval strategy; According to the final weights of the search strategies, the search results of the search strategies are sorted to obtain the final search results.

2. The retrieval method of the aggregated multi-dimensional retrieval strategy according to claim 1, characterized in that: Before obtaining the user's search terms, data processing and storage are also included, specifically: For each search strategy, the text is parsed and processed respectively to obtain the text format required for the search process of each search strategy and store it.

3. The retrieval method of claim 1, characterized in that: The various search strategies include text search, vectorized search, and graph database search.

4. The retrieval method of the aggregated multi-dimensional retrieval strategy according to claim 3 is characterized in that: Before searching the graph database, semantic calculation is also performed on the search terms to obtain the entity relationship strength and entity quantity of the search terms.

5. The retrieval method of claim 4, characterized in that: The retrieval strategy weight adjustment parameter obtained by the semantic judgment model is based on the clarity, semantic complexity and semantic richness of the keywords in the retrieval term.

6. The retrieval method of the aggregated multi-dimensional retrieval strategy according to claim 3, characterized in that: The final weight of each search strategy is obtained by calculating according to the search strategy weight adjustment parameter, which is specifically: The adjustment weight of each search strategy is calculated according to the search strategy weight adjustment parameter, and the calculation formula is: W` text =α× S text +α× (1-S complexity ) ; W` vector =γ× S complexity +δ× S semantic ; W` graph =ε× S relation +ζ×S entity ; Among them, W` text is the adjustment weight for text retrieval, W` vector is the adjustment weight for vectorized retrieval, W` graph is the adjustment weight for graph database retrieval, S text Indicates the clarity of the search term, S complexity represents the semantic complexity of the search term, S semantic Indicates the semantic richness of the search term, S relation Indicates the entity relationship strength of the search term, S entity It represents the number of entities of the search term, and α, β, γ, δ, ε, and ζ are adjustment coefficients. The adjusted weights of each search strategy are normalized to obtain the final weight of each search strategy.

7. The retrieval method of claim 6, characterized in that: The normalized calculation formula is as follows: W text= ； W vector= ； W graph= ； Among them, W text is the final weight of text retrieval, W vector is the final weight of vectorized retrieval, W graph The final weight retrieved for the graph database.

8. A retrieval system for an aggregated multi-dimensional retrieval strategy, executing a retrieval method for an aggregated multi-dimensional retrieval strategy as claimed in any one of claims 1 to 7, characterized in that: The retrieval system includes: a retrieval term acquisition module, a retrieval module, a retrieval strategy weight acquisition module, and a result output module; The search term acquisition module is used to acquire the search term input by the user; The search module uses each search strategy to perform search based on the search term to obtain search results of each search strategy; The retrieval strategy weight acquisition module is used to input the search term into the semantic judgment model to obtain the retrieval strategy weight adjustment parameter; calculate according to the retrieval strategy weight adjustment parameter to obtain the final weight of each retrieval strategy; The result output module is used to sort the search results of each search strategy according to the final weight of each search strategy to obtain the final search result.

Images