Large model-based enterprise-level search result optimization method and device, and medium

Abstract

The present application relates to the technical field of electric digital data processing. Disclosed are a large model-based enterprise-level search result optimization method and device, and a medium. The method comprises: by means of an enterprise search engine, returning a search result list consisting of a plurality of search results corresponding to a search query requested by a user; performing semantic analysis on the search query by means of a large language model, so as to obtain an answer corresponding to the search query; on the basis of the query answer, adjusting the search result list to obtain an adjusted fusion search result list; on the basis of a preset weighting calculation model, allocating corresponding search matching weightings to the fusion search result list; for each search result in the fusion search result list, acquiring user operation data fed back by the user for the search result, and, on the basis of the user operation data, calculating a user operation weighting corresponding to the search result; and, on the basis of the user operation weightings and the search matching weightings, re-adjusting the fusion search result list to obtain a target search result.

Description

A method, device, and medium for enterprise-level search result optimization based on a large model.

[0001] This application claims priority to Chinese Patent Application No. 202410564668.2, filed on May 9, 2024, entitled "A Method, Device and Medium for Optimizing Enterprise-Level Search Results Based on a Large Model", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of electronic digital data processing technology, specifically to an enterprise-level search result optimization method, device, and medium based on a large model. Background Technology

[0003] With the rapid development of information technology, search engine technology is widely used in Internet services. However, since similar user needs also exist within enterprises, enterprise-level search engines are playing an increasingly important role in information retrieval, data analysis, and knowledge management within organizations.

[0004] Search engines contain a vast amount of information. To ensure users obtain the information they need, search results need continuous optimization to improve the search experience and efficiency. Currently, common optimization methods rely primarily on user clicks and selections, optimizing search result rankings based on user actions. However, this approach depends on massive amounts of user behavior data. Compared to internet services, internal enterprise search differs by orders of magnitude in terms of user traffic, making it difficult to provide accurate enterprise-level search result rankings.

[0005] Summary of the Invention

[0006] To address the aforementioned issues, this application proposes an enterprise-level search result optimization method based on a large model, comprising:

[0007] The enterprise search engine returns a list of search results consisting of multiple search results corresponding to the user's requested search question;

[0008] The search question is input into a preset language model, and the language model performs semantic analysis on the search question to obtain the answer to the search question.

[0009] Based on the answers to the questions, the search results list is adjusted to obtain an adjusted merged search results list;

[0010] According to a preset weight calculation model, corresponding search matching weights are assigned to the fused search results list; wherein, the search matching weights are positively correlated with the order of the search results in the fused search results list;

[0011] The fused search results list is pushed to the user. For each search result in the fused search results list, user operation data is collected from the user in response to the search result. Based on the user operation data, the user operation weight corresponding to the search result is calculated.

[0012] Based on the user operation weight and the search matching weight, the fused search results list is readjusted to obtain the target search results.

[0013] In one implementation of this application, the search results list is adjusted based on the answer to the question to obtain an adjusted fused search results list, specifically including:

[0014] The question answer and multiple search results in the search results list are vectorized to obtain the corresponding question answer vector and search result vector;

[0015] For each search result vector, calculate the cosine similarity between the search result vector and the question answer vector;

[0016] The order of multiple search results in the search results list is adjusted according to the descending order of the cosine similarity to obtain the adjusted fused search results list.

[0017] In one implementation of this application, before readjusting the fused search results list based on the user operation weight and the search matching weight, the method further includes:

[0018] Identify the related business activities corresponding to different positions within the company;

[0019] Obtain the set of search questions generated by different positions in multiple search scenarios when performing search operations, as well as the set of business questions generated by the related business in each search scenario;

[0020] Based on preset association rules, the association strength between the user's job position and the associated business is determined according to the set of search questions and the set of business questions.

[0021] Obtain the preset mapping relationship between association strength and user correlation coefficient, and determine the user correlation coefficient that matches the association strength based on the mapping relationship.

[0022] In one implementation of this application, the fused search results list is readjusted based on the user operation weight and the search matching weight to obtain the target search results, specifically including:

[0023] Determine the specific related business to which the search question belongs and the user relevance coefficient corresponding to the specific related business;

[0024] For each search result, determine the product between the user operation weight corresponding to the search result and the user relevance coefficient, and sum the product with the search matching weight to obtain the target weight corresponding to the search result;

[0025] The order of multiple search results in the fused search results list is readjusted according to the target weight from high to low to obtain the target search results.

[0026] In one implementation of this application, determining the specified associated service to which the search question belongs and the user relevance coefficient corresponding to the specified associated service specifically includes:

[0027] Determine the first related business to which the search question belongs;

[0028] When there are multiple first associated services, determine all the search questions requested by the user within a preset time, as well as the auxiliary search questions that are dependent on the search questions among all the search questions;

[0029] Determine the second related business to which the subordinate search question belongs, and filter out the related business with the highest occurrence frequency from the first related business and the second related business as the designated related business to which the search question belongs.

[0030] In one implementation of this application, determining the subordinate search problems that are dependent on the search problem among all the search problems specifically includes:

[0031] Collect search log data and determine whether there is a search order dependency between any two historical search questions based on the search order among the historical search questions in the search log data.

[0032] If the number of times that any two historical search problems have a search order dependency is greater than a preset value, then the two historical search problems are used as nodes, and the search order is used as the direction of the edge to construct a search dependency graph corresponding to the historical search problems.

[0033] Based on the search dependency graph, identify the subordinate search questions that are dependent on the search question among all the search questions.

[0034] In one implementation of this application, the user operation data includes the total number of clicks made by the user on the merged search results list during a single search, and the browsing time of the search results. Based on the user operation data, the user operation weight corresponding to the search results is calculated, specifically including:

[0035] Determine the default search term, return the standard search results corresponding to the search term through the enterprise search engine, and determine the number of first search results corresponding to the standard search results;

[0036] The number of second search results corresponding to multiple search results in the merged search results list is determined, and the user operation weight corresponding to the search results is calculated using the following formula, based on the number of first search results, the number of second search results, the total number of clicks, and the browsing time:

[0037] Among them, w op The values ​​represent the weight of user actions, where M represents the number of first searches, m represents the number of second searches, R represents the total number of clicks, and t represents the browsing time.

[0038] In one implementation of this application, semantic analysis of the search question is performed using the language big model to obtain the answer to the search question, specifically including:

[0039] The search problem is vectorized using the large language model to obtain the corresponding search problem vector.

[0040] The search question vector is compared with the vector data in the enterprise knowledge vector library to determine the similarity between the search question vector and the vector data;

[0041] Select specified vector data with a similarity greater than a preset similarity from the enterprise knowledge vector library, and perform semantic analysis on the specified vector data and the search question vector to determine the corresponding search intent;

[0042] Based on the search intent, generate the answer to the search question.

[0043] This application embodiment provides an enterprise-level search result optimization device based on a large model, the device comprising:

[0044] At least one processor;

[0045] And, a memory communicatively connected to the at least one processor;

[0046] The memory stores instructions that can be executed by the at least one processor, which enable the at least one processor to perform an enterprise-level search result optimization method based on a large model as described above.

[0047] This application provides a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as follows:

[0048] An enterprise-level search result optimization method based on a large model, as described in any of the preceding items.

[0049] The enterprise-level search result optimization method based on a large model proposed in this application can bring the following benefits: Beneficial effects:

[0050] This approach generates question-and-answer results based on a large language model. By evaluating the similarity between the question-and-answer results and the search results returned by the search engine, it can better understand user intent and filter out search results with higher similarity and better match the user's actual needs. It optimizes the order of search results, improves search matching accuracy, and avoids over-reliance on the influence of user action data on search results. Collecting user action data in response to search results and calculating user action weights allows for a better understanding of user preferences and behavioral patterns. This enables further optimization of search result ranking based on user action weights, making the search engine results more aligned with user expectations. Attached Figure Description

[0051] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0052] Figure 1 is a flowchart illustrating an enterprise-level search result optimization method based on a large model provided in an embodiment of this application;

[0053] Figure 2 is a schematic diagram of the structure of an enterprise-level search result optimization device based on a large model provided in an embodiment of this application. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0055] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0056] As shown in Figure 1, an enterprise-level search result optimization method based on a large model provided in this application includes:

[0057] S101: Through the enterprise search engine, return a list of search results consisting of multiple search results corresponding to the search question requested by the user.

[0058] An enterprise search engine is a search engine system customized to meet the internal information retrieval needs of an enterprise. Its purpose is to help employees quickly and accurately find the information they need, improve work efficiency and collaboration capabilities. These search engines are typically used for internal knowledge management, document retrieval, employee training, and other aspects of an organization.

[0059] After a user enters a search query into the enterprise search engine, the server can then use the search engine to return a list of search results corresponding to the user's requested search query. Search queries typically exist in the form of multiple search keywords, but in some cases, they can also be a complete search statement. The search results in the final search results list are arranged in descending order of relevance to the search query. For example, if a user's search query is "Project A budget sheet," the enterprise search engine will retrieve documents or text information containing the two keywords mentioned above and arrange them in descending order of relevance to form the search results list.

[0060] S102: Input the search question into the preset language model, perform semantic analysis on the search question through the language model, and obtain the answer to the search question.

[0061] The search results returned by the aforementioned enterprise search engine only consider the match between the search results and the search question. However, due to the diverse data types and complex organizational structures of enterprises, unrelated documents may publish information that mentions the search keywords but is unrelated to them, thus increasing the match rate. This significantly reduces search efficiency and user experience. Therefore, to adapt to application scenarios where enterprises have large amounts of internal data but low user access, this application embodiment provides more accurate and relevant search results based on a large language model. A large language model refers to a large-scale neural network model trained using deep learning technology, used to process natural language processing tasks. It can learn the syntax, semantics, and contextual information of a language, helping computers better understand and generate text. Available language models include GPT-3, BERT, and XLNet. Therefore, based on a deep understanding of semantics, the large language model can automatically generate answers to questions, thus providing data references for subsequent optimization of the search results list.

[0062] In one embodiment, a large language model first vectorizes the search question to obtain a corresponding search question vector. Then, this vector is compared with vector data in an enterprise knowledge vector library to determine the similarity between them. Based on this similarity, specified vector data with a similarity greater than a preset similarity can be selected from the enterprise knowledge vector library. Semantic analysis is then performed on the specified vector data and the search question vector to deeply understand the user's search intent. Finally, the large language model can generate the corresponding answer to the search question based on this intent.

[0063] S103: Based on the answers to the questions, adjust the search results list to obtain the adjusted merged search results list.

[0064] The language model can generate rich and diverse answers to questions, potentially including information or perspectives not covered by search engines. This can supplement search results and provide users with more comprehensive information. Therefore, after generating answers based on the language model, it's necessary to compare them with the search results provided by the search engine. Based on the comparison results, the search results list should be adjusted to obtain a refined, merged search results list. Comparing the answers generated by the language model with the search results list can identify potential omissions, errors, or duplicates in the search results list, allowing for corresponding adjustments and optimizations, thereby improving the quality and accuracy of the search results.

[0065] In one embodiment, both the question answer and the search results exist in text form. To better compare them, multiple search results in the question answer and search results lists need to be vectorized separately to obtain corresponding question answer vectors and search result vectors. After vectorization, the cosine similarity between each search result vector and the question answer vector needs to be calculated. Cosine similarity reflects the similarity between search results provided by two different retrieval methods. The higher the similarity, the higher the accuracy of the search results and the greater the likelihood that they meet the user's needs. Therefore, the order of multiple search results in the search results list needs to be adjusted according to the descending order of cosine similarity to obtain the adjusted fused search results list.

[0066] S104: Based on the preset weight calculation model, assign corresponding search matching weights to the fused search results list; wherein, the search matching weights are positively correlated with the order of the search results in the fused search results list.

[0067] The merged search results list combines the analysis results of the search engine and the language big data model, resulting in the final order of search results. Search results appearing earlier in the list are more likely to meet the user's actual search needs. Therefore, a pre-defined weighting calculation model is used to assign corresponding search matching weights to the merged search results list. These search matching weights reflect the degree of match between the search results order and the user's actual needs, and are positively correlated with the order of the search results in the merged search results list. In other words, the search matching weights for search results decrease sequentially according to their order.

[0068] It should be noted that, due to the large number of search results in the merged search results list, and the fact that users only view a few of the earlier results when browsing the search results, the weight calculation model only considers the earlier, more similar search results when assigning search matching weights. Therefore, it is necessary to filter out a preset number of search results with a similarity greater than a preset threshold or those that are among the earlier results in the merged search results list. Assuming that the final number of highly similar search results is five, the weight calculation model can assign search matching weights to each search result according to the idea of ​​average distribution, assigning weight values ​​of 1, 0.8, 0.6, 0.4, and 0.2 respectively. It should be noted that the preset threshold and preset number can be set according to actual search needs, and this application does not limit them.

[0069] S105: Push the integrated search results list to the user. For each search result in the integrated search results list, collect user operation data in response to the search results, and calculate the user operation weight corresponding to the search results based on the user operation data.

[0070] After assigning search matching weights to the merged search results list, the list needs to be pushed to users. Upon receiving the search results, users will perform actions such as clicking and browsing. The server needs to collect user action data for each search result in the merged search results list and calculate the corresponding user action weight based on this data. This user action data reflects user satisfaction and relevance with the search results. By calculating the user action weights, the ranking of search results can be optimized and adjusted to provide a search experience that better meets user needs.

[0071] In one embodiment, the user operation data takes into account the overall number of clicks and browsing duration of the user on the search results. Before calculating the user operation weight, it is necessary to evaluate the standard search ability of the enterprise search engine, and this process can be achieved through default search standard words. Generally, the search standard words use words with a relatively high word frequency, such as "of", "is", "has", etc. Record the standard search results returned by the enterprise search engine and the corresponding number of the first search results, and at the same time, it is also necessary to determine the number of the second search results corresponding to multiple search results in the integrated search result list. According to the number of the first search results, the number of the second search results, the total number of clicks, and the browsing duration, the user operation weight corresponding to the search results can be calculated, and this process can be achieved through the following formula:

[0072] where, w op represents the user operation weight, M represents the number of the first search results, m represents the number of the second search results, R represents the total number of clicks, and t represents the browsing duration.

[0073] After calculating the user operation weight, if it is necessary to adjust the integrated search result list based on this data, it is also necessary to additionally consider the relevance between the user's own position and the search problem. By analyzing the position association information, the search engine can more quickly locate the position or business that best matches the search problem, which not only improves the search efficiency but also increases the user's satisfaction with the search results. Therefore, the greater the relevance between the position and the search problem, the relatively higher the matching degree of the finally returned search results to the user's search needs. Each position corresponds to different associated businesses. For example, the associated businesses corresponding to the sales position can be market research, sales data analysis, customer relationship management, etc., and the associated businesses corresponding to R & D personnel can be programming skills training, software development process, technical document writing, etc. Different search problems belong to different associated businesses. For example, searching for the sales data of a certain product belongs to the associated business of sales data analysis, and searching for the system usage guide belongs to the associated business of skills training. Since the number of search problems is large and the scope involved is wide, when evaluating the relevance between the position and the search problem, it can be converted into evaluating the relevance between the position and the associated business. Then, the relevance between a certain search problem and the position can be determined according to the relevance between the associated business to which the search problem belongs and the position.

[0074] In one embodiment, the relevance between job positions and related business activities can be represented by user relevance coefficients. First, it's necessary to identify the related business activities corresponding to different job positions within the company. Then, obtain the set of search questions generated by different job positions in multiple search scenarios, as well as the set of business questions generated by related business activities in each search scenario. Search scenarios include the background, purpose, and search conditions of the user's query request. For example, a search scenario could be a user searching for product documents or a user searching for employee training materials. Because different job positions involve different product types and training skills, the search questions generated in different search scenarios will also differ. Similarly, different related business activities within a search scenario will generate different search questions. For example, in the search scenario of searching for product documents, when the related business is market research, the corresponding search question could be to search for the current market share of products and user feedback information; when the related business is sales data analysis, the corresponding search question could be to search for sales documents of a certain product within a specified time period.

[0075] The search question set can be Q. j ={q1,q 2, …q j}, the set of related problems is P i ={p1,p 2, …p i After obtaining the search question set and the business question set, the strength of the association between the user's job position and related business can be determined through preset association rules.

[0076] The association rule can be represented as: and M represents the job position, and N represents related business functions. Let represent the empty set, and ... and confidence level These two attributes. Using the fitness function, the strength of the correlation between a user's job position and related business functions can be calculated, as shown in the following formula:

[0077] in, Let j represent the fitness function used to calculate the strength of the association between a user's job position and related business functions, where j represents the number of job positions. Here, min_sup represents the minimum support, and min_conf represents the minimum confidence.

[0078] After obtaining the correlation strength between job positions and related businesses through the above correlation rules, it is necessary to obtain the mapping relationship between the preset correlation strength and user correlation coefficient. In this way, based on the mapping relationship, the user correlation coefficient that matches the correlation strength can be determined when the correlation strength is clear.

[0079] Therefore, after clarifying the correlation between the user's job position and various related businesses through the above process, it is necessary to determine the specific related business to which the search question belongs and the user relevance coefficient corresponding to that specific related business.

[0080] It's important to note that a search query may relate to multiple specific related services. For example, when searching for sales data of a product, both market research and sales data analysis might involve sales data. Therefore, it's necessary to more precisely pinpoint the specific related services based on the dependencies between the user's search queries.

[0081] First, determine the primary related business to which the search question belongs. If there are multiple primary related businesses, identify all search questions requested by the user within a preset time period, as well as any dependent search questions within those search questions. Typically, users retrieve a lot of related information when searching for information. For example, when searching for product sales data, they first search for sales figures. If the sales data is poor, they need to understand the reasons for the poor sales, which can be achieved by obtaining user feedback data. Therefore, searching for user feedback data depends on the search question of searching for product sales data. It's understandable that while dependent search questions and dependent search questions may belong to different related businesses, those with overlapping related businesses are likely to have a stronger correlation with the search question. In this case, the related business can be directly designated as the specified related business for the search question. Therefore, after identifying the dependent search questions, determine the secondary related business to which they belong. Then, from the primary and secondary related businesses, select the related business with the highest frequency of occurrence as the specified related business to which the search question belongs.

[0082] In one embodiment, the determination of auxiliary search questions is based on a search dependency graph. The search dependency graph consists of search questions and directed edges connecting different search questions. The direction of the edges represents the dependency relationship between different search relationships. Essentially, a dependency relationship means that the formulation of one search question depends on the formulation of another search question, and the search order of the two is relatively fixed. Therefore, the search question at the endpoint of a directed edge depends on the search question at the starting point. The search dependency graph needs to be constructed based on pre-collected search log data. After collecting the search log data, it is necessary to determine whether there is a search order dependency between any two historical search questions based on the search order among the historical search questions in the search log data. If the number of times any two historical search questions have a search order dependency exceeds a preset value, then any two historical search questions are used as nodes, and the search order is used as the direction of the edge to construct the search dependency graph corresponding to the historical search questions. Thus, based on the search dependency graph, auxiliary search questions that have a dependency relationship with the user's proposed search question can be determined from all search questions.

[0083] S106: Based on the user operation weight and search matching weight, readjust the merged search results list to obtain the target search results.

[0084] After determining the user relevance coefficient corresponding to the search question, the server combines the user relevance coefficient, user action weight, and search matching weight to recalculate the target weight corresponding to each search result. Then, the merged search results list is readjusted according to the order of the target weights to obtain the target search results.

[0085] Specifically, for each search result, the product of the user action weight and the user relevance coefficient is determined, and this product is summed with the search matching weight to obtain the target weight for the search result. This target weight is a comprehensive weight that takes into account factors such as the question answer provided by the language model, user action interest, and question relevance, thus more comprehensively reflecting the matching degree between the search results and the user's actual needs. Therefore, the order of multiple search results in the merged search results list is readjusted according to the target weight from high to low to obtain the target search results. After obtaining the target search results, the server controls the front-end interface to adjust the order of the search results and displays the adjusted target search results to the user in real time, enabling the user to obtain search results that better meet their search needs in a timely manner, improving the user's search efficiency and accuracy.

[0086] The above are embodiments of the methods proposed in this application. Based on the same idea, some embodiments of this application also provide devices and non-volatile computer storage media corresponding to the above methods.

[0087] Figure 2 is a schematic diagram of the structure of an enterprise-level search result optimization device based on a large model provided in an embodiment of this application. As shown in Figure 2, it includes:

[0088] At least one processor; and,

[0089] At least one processor-communication-connected memory; wherein,

[0090] The memory stores instructions that can be executed by at least one processor, and the instructions, when executed by at least one processor, enable at least one processor to:

[0091] The enterprise search engine returns a list of search results consisting of multiple search results corresponding to the user's requested search question;

[0092] The search question is input into a pre-defined language model, which performs semantic analysis on the search question to obtain the answer to the search question.

[0093] Based on the answers to the questions, the search results list is adjusted to obtain the adjusted merged search results list;

[0094] Based on a pre-defined weight calculation model, corresponding search matching weights are assigned to the fused search results list; wherein, the search matching weights are positively correlated with the order of the search results in the fused search results list.

[0095] The integrated search results list is pushed to the user. For each search result in the integrated search results list, user action data is collected based on the user's feedback on the search results. Based on the user action data, the user action weight corresponding to the search result is calculated.

[0096] Based on user action weights and search matching weights, the merged search results list is readjusted to obtain the target search results.

[0097] This application provides a non-volatile computer storage medium storing computer-executable instructions, which are configured as follows:

[0098] The enterprise search engine returns a list of search results consisting of multiple search results corresponding to the user's requested search question;

[0099] The search question is input into a pre-defined language model, which performs semantic analysis on the search question to obtain the answer to the search question.

[0100] Based on the answers to the questions, the search results list is adjusted to obtain the adjusted merged search results list;

[0101] Based on a pre-defined weight calculation model, corresponding search matching weights are assigned to the fused search results list; wherein, the search matching weights are positively correlated with the order of the search results in the fused search results list.

[0102] The integrated search results list is pushed to the user. For each search result in the integrated search results list, user action data is collected based on the user's feedback on the search results. Based on the user action data, the user action weight corresponding to the search result is calculated.

[0103] Based on user action weights and search matching weights, the merged search results list is readjusted to obtain the target search results.

[0104] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0105] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0106] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0107] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more flowchart illustrations and / or one or more block diagrams.

[0108] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0109] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0110] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0111] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0112] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0113] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0114] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. An enterprise-level search result optimization method based on a large model, wherein, The method includes: The enterprise search engine returns a list of search results consisting of multiple search results corresponding to the user's requested search question; The search question is input into a preset language model, and the language model performs semantic analysis on the search question to obtain the answer to the search question. Based on the answers to the questions, the search results list is adjusted to obtain an adjusted merged search results list; According to a preset weight calculation model, corresponding search matching weights are assigned to the fused search results list; wherein, the search matching weights are positively correlated with the order of the search results in the fused search results list; The fused search results list is pushed to the user. For each search result in the fused search results list, user operation data is collected from the user in response to the search result. Based on the user operation data, the user operation weight corresponding to the search result is calculated. Based on the user operation weight and the search matching weight, the fused search results list is readjusted to obtain the target search results; Before readjusting the fused search results list based on the user operation weight and the search matching weight, the method further includes: Identify the related business activities corresponding to different positions within the company; Obtain the set of search questions generated by different positions in multiple search scenarios when performing search operations, as well as the set of business questions generated by the related business in each search scenario; Based on preset association rules, the association strength between the user's job position and the associated business is determined according to the set of search questions and the set of business questions. Obtain the preset mapping relationship between association strength and user correlation coefficient, and determine the user correlation coefficient that matches the association strength based on the mapping relationship; Based on the user operation weight and the search matching weight, the fused search results list is readjusted to obtain the target search results, specifically including: Determine the specific related business to which the search question belongs and the user relevance coefficient corresponding to the specific related business; For each search result, determine the user action weight and user relevance corresponding to that search result. The product of the coefficients is then summed with the search matching weight to obtain the target weight corresponding to the search result; The order of multiple search results in the fused search results list is readjusted according to the target weight from high to low to obtain the target search results.

2. The enterprise-level search result optimization method based on a large model according to claim 1, wherein, Based on the answers to the questions, the search results list is adjusted to obtain an adjusted merged search results list, specifically including: The question answer and multiple search results in the search results list are vectorized to obtain the corresponding question answer vector and search result vector; For each search result vector, calculate the cosine similarity between the search result vector and the question answer vector; The order of multiple search results in the search results list is adjusted according to the descending order of the cosine similarity to obtain the adjusted fused search results list.

3. The enterprise-level search result optimization method based on a large model according to claim 1, wherein, Determining the specified related business to which the search question belongs and the user relevance coefficient corresponding to the specified related business specifically includes: Determine the first related business to which the search question belongs; When there are multiple first associated services, determine all the search questions requested by the user within a preset time, as well as the auxiliary search questions that are dependent on the search questions among all the search questions; Determine the second related business to which the subordinate search question belongs, and filter out the related business with the highest occurrence frequency from the first related business and the second related business as the designated related business to which the search question belongs.

4. The enterprise-level search result optimization method based on a large model according to claim 3, wherein, Identify the dependent search questions that are dependent on the search question from among all the search questions, specifically including: Collect search log data and determine whether there is a search order dependency between any two historical search questions based on the search order among the historical search questions in the search log data. If the number of times that any two historical search problems have a search order dependency is greater than a preset value, then the two historical search problems are used as nodes, and the search order is used as the direction of the edge to construct a search dependency graph corresponding to the historical search problems. Based on the search dependency graph, identify the subordinate search questions that are dependent on the search question among all the search questions.

5. The enterprise-level search result optimization method based on a large model according to claim 1, wherein, The user operation data includes the total number of clicks the user made on the merged search results list during a single search, as well as the duration of browsing the search results. Based on the user operation data, the user operation weight corresponding to the search results is calculated, specifically including: Determine the default search term, return the standard search results corresponding to the search term through the enterprise search engine, and determine the number of first search results corresponding to the standard search results; The number of second search results corresponding to multiple search results in the merged search results list is determined, and the user operation weight corresponding to the search results is calculated using the following formula, based on the number of first search results, the number of second search results, the total number of clicks, and the browsing time: Among them, w op The values ​​represent the weight of user actions, where M represents the number of first searches, m represents the number of second searches, R represents the total number of clicks, and t represents the browsing time.

6. The enterprise-level search result optimization method based on a large model according to claim 1, wherein, Semantic analysis of the search question is performed using the large language model to obtain the answer to the search question, specifically including: The search problem is vectorized using the large language model to obtain the corresponding search problem vector. The search question vector is compared with the vector data in the enterprise knowledge vector library to determine the similarity between the search question vector and the vector data; Select specified vector data with a similarity greater than a preset similarity from the enterprise knowledge vector library, and perform semantic analysis on the specified vector data and the search question vector to determine the corresponding search intent; Based on the search intent, generate the answer to the search question.

7. An enterprise-level search result optimization device based on a large model, wherein, The device includes: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to perform actions as described in claims 1 to 6. Any one of the following is an enterprise-level search result optimization method based on a large model.

8. A non-volatile computer storage medium storing computer-executable instructions, wherein, The computer-executable instructions are set as follows: An enterprise-level search result optimization method based on a large model, as described in any one of claims 1 to 6.

Images