Large Model-Based Search Optimization Method, Device, and Medium

Abstract

The present application discloses a search optimization method, device, and medium based on a large model, which relates to the field of data processing. The method includes: receiving query information input by a user and preprocessing the query information; performing semantic analysis on the query information based on a pre-trained deep learning model to extract key information contained therein; using the query information and the key information as inputs to generate relevant first search results through a pre-trained large model; obtaining a business entity corresponding to the query information, determining compressed structure information corresponding to the business entity, and optimizing the first search results based on the compressed structure information to obtain second search results; and sorting and displaying according to the second search results. It can accurately analyze the deep intention and context of the user's query, reduce irrelevant results caused by keyword ambiguity, and can quickly generate candidate results based on the reasoning ability of the large model, optimizing the search efficiency.

Description

Technical Field

[0001] This application relates to the field of data processing, and particularly to a search optimization method, device, and medium based on large models. Background Art

[0002] With the development of Internet technology, Internet information has shown an explosive growth trend.

[0003] As an important tool for Internet information retrieval, the accuracy and efficiency of search engines directly affect user experience and the effective dissemination of Internet information. However, with the continuous increase in the amount of information and the increasing complexity of query requirements, traditional search engine technologies have been difficult to meet users' needs for high-quality search results.

[0004] Traditional search engines mainly rely on keyword matching and page weight ranking algorithms to provide search results. This approach is overwhelmed when dealing with massive amounts of information and complex queries. On the one hand, simple keyword matching is difficult to accurately capture the deep semantics of user queries, resulting in a large number of irrelevant or low-quality results; on the other hand, traditional page weight ranking algorithms often only consider surface features such as external links and internal structures of web pages, while ignoring important factors such as content quality and user experience. Summary of the Invention

[0005] To solve the above problems, this application proposes a search optimization method based on large models, including:

[0006] Receiving query information input by a user and preprocessing the query information;

[0007] Based on a pre-trained deep learning model, performing semantic analysis on the query information to extract key information contained therein;

[0008] Using the query information and the key information as inputs, generating relevant first search results through a pre-trained large model;

[0009] Obtaining a business entity corresponding to the query information, determining compressed structure information corresponding to the business entity, and optimizing the first search results based on the compressed structure information to obtain second search results;

[0010] Sorting and displaying according to the second search results.

[0011] In one example, preprocessing the query information specifically includes:

[0012] Obtaining a corresponding string according to the query information;

[0013] Performing text preprocessing on the string;

[0014] Perform word segmentation on the said string, and perform part-of-speech tagging on the word segmentation results.

[0015] In one example, based on a pre-trained deep learning model, perform semantic analysis on the said query information, and extract the key information contained therein, specifically including:

[0016] Based on a pre-trained deep learning model, determine the minimum text units contained in the word segmentation results of the said query information;

[0017] Perform position encoding on the said minimum text units;

[0018] Output corresponding context vectors for the said minimum text units; the context vectors include a first vector for aggregating semantic representations for intent classification and a second vector for fine-grained semantic features for entity recognition;

[0019] Perform intent classification according to the first vector, and perform key information extraction according to the intent classification and the second vector.

[0020] In one example, obtain the business entity corresponding to the said query information, and determine the compressed structure information corresponding to the business entity, specifically including:

[0021] Generate a corresponding first business entity according to the said key information;

[0022] Based on the corresponding relationship in the local database, expand the said key information, so as to generate a corresponding second business entity for each first business entity;

[0023] According to the corresponding relationship, compress each first business entity and its corresponding second business entity to generate corresponding compressed structure information.

[0024] In one example, optimize the said first search results based on the compressed structure information to obtain second search results, specifically including:

[0025] For the compressed structure information corresponding to each first business entity, select at least one second business entity as the selected entity corresponding to the compressed structure information;

[0026] Combine the selected entities corresponding to all compressed structure information to obtain an entity combination;

[0027] Perform relevance analysis on the first search results according to the entity combination, and use the first search results with the corresponding first relevance score higher than the preset score as the sub-search results corresponding to the entity combination;

[0028] For each sub-search result, several search results with the highest first relevance score are respectively selected therefrom, and after deduplication, they are combined to obtain a second search result.

[0029] In one example, the method further includes:

[0030] Determine the number of results of the second search result obtained by combining after deduplication;

[0031] If the number of results is higher than the preset number range, based on the number of corresponding second business entities in the entity combination, the corresponding sub-search results are deleted in ascending order until the number of results meets the preset number range;

[0032] If the number of results is lower than the preset number range, based on the number of corresponding second business entities in the entity combination, the corresponding sub-search results are added in descending order until the number of results meets the preset number range, or all sub-search results have been added.

[0033] In one example, sorting and displaying according to the second search result specifically includes:

[0034] Obtain the access permission of the user;

[0035] Filter the second search result according to the access permission to obtain a third search result;

[0036] Obtain the second relevance score of the large model for the first search result, and according to the second relevance score, obtain the first sorting corresponding to the third search result;

[0037] Obtain the access data of the third search result, and obtain the corresponding quality score according to the access data, and adjust the first sorting according to the quality score to obtain a second sorting;

[0038] Select the corresponding display method, and based on the display method, sort and display the third search result according to the second sorting.

[0039] In one example, the method further includes:

[0040] Obtain the operation information of the user for the third search result;

[0041] According to the operation information, obtain the personalized behavior information of the user;

[0042] According to the personalized behavior information, perform personalized optimization on the sorting method and display method corresponding to the user.

[0043] On the other hand, the present application also proposes a search optimization device based on a large model, including:

[0044] At least one processor; and,

[0045] A memory communicatively connected to the at least one processor; wherein,

[0046] The memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the search optimization method based on a large model as described in any of the above examples.

[0047] On the other hand, the present application also proposes a non-volatile computer storage medium storing computer-executable instructions, and the computer-executable instructions are configured as: the search optimization method based on a large model as described in any of the above examples.

[0048] The search optimization method based on a large model proposed by the present application can bring the following beneficial effects:

[0049] 1. Through deep model + large model for in-depth semantic understanding, accurately parsing the deep intention and context of the user's query, reducing irrelevant results caused by keyword ambiguity, and being able to quickly generate candidate results based on the reasoning ability of the large model to optimize the search efficiency.

[0050] 2. Combining dynamic optimization of the business entity compression structure to ensure that the search results highly match the user's needs and business rules, improving the relevance of the search results, and being able to support the automated parsing and response of multi-conditions and implicit requirements (such as price range, scenario preference) to enhance the adaptability to complex queries. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation to the present application. In the drawings:

[0052] Figure 1 It is a schematic flow chart of the search optimization method based on a large model in an embodiment of the present application;

[0053] Figure 2 It is a schematic flow chart of the search optimization method based on a large model in a certain situation in an embodiment of the present application;

[0054] Figure 3 It is a schematic diagram of the search optimization device based on a large model in an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] To make the objectives, technical solutions, and advantages of this application more clear, the following will clearly and completely describe the technical solutions of this application in combination with specific embodiments of this application and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

[0056] The following will detail the technical solutions provided by each embodiment of this application in combination with the drawings.

[0057] Search optimization can be carried out based on deep learning and natural language processing technologies to improve the performance of the search engine. By constructing a deep learning model, in-depth understanding and analysis of user queries and web page content are carried out to provide more accurate and valuable search results.

[0058] However, there are still the following deficiencies:

[0059] 1. The contradiction between model complexity and performance

[0060] To capture the deep semantics of user queries and web page content, a complex deep learning model needs to be constructed. However, the increase in model complexity will lead to an increase in computational cost and training difficulty, thus limiting the performance of the model in practical applications.

[0061] 2. Data sparsity and cold start problems

[0062] For newly emerging queries or web pages, it is difficult to provide accurate search results due to the lack of sufficient historical data and user feedback. Such data sparsity and cold start problems limit the generalization ability and adaptability of the search engine.

[0063] 3. Continuous optimization of user experience

[0064] Although more accurate search results can be provided, there is still room for improvement in terms of user experience. For example, how to sort and display search results according to users' personalized needs and preferences, and how to provide users with a more intuitive and convenient search experience, etc.

[0065] As Figure 1 and Figure 2 shown, the embodiments of this application provide a search optimization method based on a large model, including:

[0066] S101: Receive the query information input by the user and preprocess the query information.

[0067] When the user inputs a query through the search box or other interfaces of the search engine, receive the user's query information to obtain the corresponding string.

[0068] Preprocess the query information (query string), including the text preprocessing process, such as removing irrelevant information like spaces, punctuation marks, special characters, etc., for subsequent processing.

[0069] Use a word segmentation tool to perform word segmentation on the query string, breaking down the query into several meaningful words or phrases. Based on word segmentation, further perform part-of-speech tagging to label the part of speech of each word or phrase to better understand the intent of the query.

[0070] S102: Based on a pre-trained deep learning model, perform semantic analysis on the query information and extract the key information contained therein.

[0071] Call a pre-trained deep learning model, such as BERT, Transformer, etc., to perform in-depth semantic analysis on the preprocessed query information.

[0072] The deep learning model can analyze the keywords, phrases, and sentence structures in the query, understand the intent and context of the query. The model will also extract key information in the query, such as the topic, category, time, etc., providing important clues for subsequent searches.

[0073] Specifically, BERT-base (12-layer Transformer encoder, 12 attention heads, hidden layer dimension 768) can be selected as the model architecture, and a multi-task learning head can be set. Add two parallel task branches after the BERT output layer, namely the intent classification head (including a fully connected layer + Softmax), which outputs the query intent category (such as product search, knowledge Q&A, event query, etc.), and the sequence annotation head (including BiLSTM + CRF), which annotates the key entities in the query (such as time, location, product name, etc.).

[0074] Based on the general BERT, according to the business scenario corresponding to the current search function, use vertical domain corpus (such as e-commerce product descriptions, enterprise documents) as training parameters for secondary pre-training to enhance domain semantic understanding.

[0075] First, perform the pre-training stage through the general corpus and the vertical domain corpus for masked language modeling (MLM) + next sentence prediction (NSP), and then perform the fine-tuning stage through the labeled search query dataset to fine-tune the model parameters.

[0076] Among them, the training parameters can be as shown in Table 1 below.

[0077] Table 1 Deep Learning Model Training Parameter Table

[0078]

[0079] Based on this, when extracting key information, based on a pre-trained deep learning model, the minimum text units included in the word segmentation result of the query information are determined, and position encoding is performed on the minimum text units. The position encoding includes position encoding within a single sentence and segment encoding.

[0080] A corresponding context vector is output for the minimum text unit. The context vector includes a first vector for the aggregated semantic representation for intent classification and a second vector for the fine-grained semantic features for entity recognition.

[0081] Taking the above as an example, it outputs a context vector (corresponding to 768 dimensions) for each minimum text unit (Token). The first vector (CLS vector) is used for the aggregated semantic representation for intent classification, and the second vector (Token vector) is used for the fine-grained semantic features for entity recognition.

[0082] At this time, intent classification is performed according to the first vector. The first vector is input into a fully connected layer, and according to multiple preset intent categories, the corresponding intent category probabilities are calculated. Key information extraction is performed according to the intent classification and the second vector. Each second vector is input into a BiLSTM-CRF layer for sequence labeling, so as to extract the corresponding entities, attributes, conditions, etc., which can be directly used as key information, or can be further filtered based on blacklists, whitelists, etc. in the vertical domain to which they belong, or corresponding weights can be set, so as to obtain corresponding content according to the second vector and extract key information.

[0083] S103: Use the query information and the key information as inputs, and generate a relevant first search result through a pre-trained large model.

[0084] Use the preprocessed query information and key information as inputs and pass them to a pre-trained large model. The large model utilizes its powerful capabilities to perform deep learning and reasoning on the inputs, predicts the web pages or documents most relevant to the query, and forms the first search result. The large model can output a sorted result list, which contains the identifiers and relevant information of the web pages or documents most relevant to the query, and performs sorting and relevance analysis on the first search result.

[0085] Among them, the large model can use a large language model (LLM), which can directly use an open-source large language model, or can be further trained in the corresponding vertical domain based on requirements.

[0086] S104: Obtain the business entity corresponding to the query information, determine the compressed structure information corresponding to the business entity, and optimize the first search result based on the compressed structure information to obtain a second search result.

[0087] A business entity refers to an entity that has a certain relevance to the query information. For example, when the query information is to search for a product, the business entities can include the product, its complementary products, the merchants involved in the production and sales of the product, etc. When the query information is to search for knowledge, the business entities can be the knowledge itself, related knowledge, the discoverer of the knowledge, the current research status of the knowledge, etc.

[0088] The compressed structure information refers to the compressed information obtained by integrating the attributes and relationships corresponding to the business entities (this information may be stored in a database or network, etc.) and then compressing them.

[0089] Apply the compressed structure information to the first search results to further screen and optimize the results. For example, search results that do not meet the conditions can be filtered out according to the attributes of the entities, or the search results can be clustered or sorted according to the relationships between the entities.

[0090] Specifically, generate corresponding first business entities according to the key information. Based on the corresponding relationships in the local database, expand the key information, so as to generate corresponding second business entities for each first business entity.

[0091] When there are multiple key information, there can be multiple corresponding first business entities. For example, when the query information input by the user is: "Recommend mobile phones that can be used for online shopping within 5000 yuan", the identified key information can include: 5000 yuan, online shopping, mobile phone. At this time, for each key information, generate corresponding first business entities, which are respectively: price (less than 5000 yuan), e-commerce platform, mobile phone.

[0092] In the local database, corresponding relationships can be pre-stored in advance, that is, for each type of business entity, in which directions it can be expanded. Generally speaking, the structure information can be divided into two types: attributes and relationships. Attributes are used to describe the nature of the first business entity itself, while relationships are used to describe the content related to the first business entity. Of course, for some first business entities, according to the needs, only one type of expansion can be carried out among the two types of attributes and relationships.

[0093] In addition, for the personal habits of users that can be collected, the corresponding second business entities can be selected according to the personal habits.

[0094] For example, for "Price (less than 5,000 yuan)", its corresponding entity type is the price type. In its corresponding structure information, the attributes can include: pricing type (including original price, promotional price, membership price, etc.), promotional time period, etc. The relationships can include: belonging price range, purchase method (including retail, bulk purchase, etc.), etc. Thus, the second business entities that can be expanded include: promotional price within 5,000 yuan, membership price within 5,000 yuan, currently in the promotional time period and price within 5,000 yuan, bulk purchase price within 5,000 yuan, etc.

[0095] For "e-commerce platform", its corresponding entity type is the network platform type. In its corresponding structure information, the attributes can include: platform name, delivery range, payment method, platform reputation, etc. Thus, the second business entities that can be expanded include: e-commerce platform with platform name A, e-commerce platform whose platform delivery range meets the user's location, e-commerce platform with a relatively high platform reputation, etc. Among them, the user's location and the user's common payment method can be obtained in advance. Among them, which e-commerce platform name to choose can be determined through the user's historical software usage records. If it cannot be determined, one or more commonly used e-commerce platforms by the public can be selected. Similar methods can also be used for other subsequent second business entities.

[0096] For "mobile phone", its corresponding entity type is the commodity type. In its corresponding structure information, the attributes can include: mobile phone brand, mobile phone performance, mobile phone price, etc. The relationships can include: price of the matching mobile phone case, mobile phone software ecological network, etc. Thus, the second business entities that can be expanded include: mobile phone of brand B, mobile phone with a relatively high mobile phone performance score, mobile phone with a mobile phone software ecological network of type C, etc. Among them, since the mobile phone price has been used as the first business entity, corresponding avoidance is carried out in the second business entity.

[0097] Based on this, for each first business entity, a number of related and more restricted second business entities are generated. When the number of second business entities is large, only a fixed number (such as 3) of commonly used second business entities can be selected (or personalized selection can be made according to the user's previous search records), so as to ensure the quickness of the subsequent calculation process.

[0098] According to the corresponding relationship, compress each first business entity and its corresponding second business entity to generate the corresponding compressed structure information.

[0099] At this time, when optimizing the first search results, for the compressed structure information corresponding to each first business entity, at least one second business entity can be selected as the selected entity corresponding to the compressed structure information. That is to say, at least one is selected for each compressed structure information, which is used to more specifically describe the second business entity of the first business entity. Of course, multiple can also be selected for combination and used together as the selected entity. For example, when the first business entity is a mobile phone, select mobile phones of brand B and mobile phones with a high performance score together as the second business entity, and combine them to get mobile phones of brand B with a high performance score, and use it as the corresponding selected entity.

[0100] According to the selected entities corresponding to all the compressed structure information, entity combinations are obtained. Here, all the entity combinations can be obtained by traversing. For example, there are 2 first business entities in total, namely entity A and entity B, which respectively correspond to second business entities A1 and A2, and B1 and B2. At this time, the entity combinations can include: (A1,B1) (A1,B2) (A2,B1) (A2,B2) (A1,A2,B1) (A1,A2,B2) (A1,B1,B2) (A2,B1,B2) (A1,A2,B1,B2). Among them, if there are conflicts between some entity combinations, for example, mobile phones of brand A and mobile phones of brand B have conflicts with each other, they can be selected for deletion based on requirements. Of course, since this is for search and relevance analysis this time, selecting entity combinations with conflicts does not prevent the corresponding work from being carried out, so they can also be retained based on requirements.

[0101] Perform relevance analysis on the first search results according to the entity combinations, and use the first search results with the corresponding first relevance score higher than the preset score as the sub-search results corresponding to the entity combinations. The relevance analysis can be determined by a large language model or keyword matching. The first search results include multiple websites, documents, etc. Relevance analysis can be performed through each entity combination, and for each entity combination, sort according to its corresponding first relevance score to obtain its corresponding sub-search results. At this time, since the entity combinations are different in each sub-search result, the selected sub-search results usually vary.

[0102] For each sub-search result, select several search results with the highest first relevance score respectively. At this time, usually there is a search result whose first relevance score is very high with multiple entity combinations and there are duplicates, so the second search results are obtained by combining after deduplication.

[0103] In the second search results at this time, for different entity combinations, the key information is extended in different directions, so as to perform corresponding relevance analysis, and select the content that is highly relevant in each direction to provide users with comprehensive search results.

[0104] Of course, it is also possible to determine the number of results of the second search results obtained by combining after deduplication to prevent the number of results from being too large or too small, which may affect the user experience.

[0105] A corresponding preset quantity range is set in advance. If the number of results is higher than the preset quantity range, that is, the search results are too many, then based on the number of the corresponding second business entities in the entity combination, the corresponding sub-search results are deleted in ascending order, that is, the search results corresponding to the second business entities with fewer numbers are preferentially deleted (when deleting each search result, the search results with lower first relevance scores can also be preferentially deleted) until the number of results meets the preset quantity range.

[0106] If the number of results is lower than the preset quantity range, then based on the number of the corresponding second business entities in the entity combination, the corresponding sub-search results are added in descending order, that is, the search results that are relevant to more second business entities and contain more content are preferentially added until the number of results meets the preset quantity range, or all sub-search results have been added.

[0107] S105: Sort and display according to the second search results.

[0108] For the second search results in the above text, they can be sorted and displayed according to the high and low of the first relevance score, the number of second business entities in the entity combination, etc.

[0109] Specifically, the access permission of the user can be obtained, and according to the access permission, the second search results are filtered to obtain the third search results.

[0110] Verify the user's access permission to ensure that the user can only see the search results to which they have access. According to information such as the user's role and permissions, the search results are filtered to exclude web pages or documents that do not meet the permission requirements. The filtered results will be re-sorted to ensure that the user can only see the results that are relevant to the query and to which they have access.

[0111] Obtain the second relevance score of the large model for the first search results, and according to the second relevance score, obtain the first sorting corresponding to the third search results. When the large model performs a search, in addition to giving the first search results, it can also give the corresponding relevance scores. At this time, the third search results still belong to a part of the first search results, so the corresponding second relevance scores can also be obtained and sorted according to the high and low.

[0112] Obtain the access data of the third search result, obtain the corresponding quality score based on the access data, and adjust the first sorting according to the quality score to obtain the second sorting.

[0113] Based on the preliminary first sorting, comprehensively consider factors such as web page quality and user feedback, and fine-tune the sorting result to obtain the second sorting. For example, the sorting result can be adjusted according to information such as the click-through rate, browsing duration, and evaluation of the web page. Search results with a high click-through rate, a long browsing duration, and good evaluations can have their positions in the first sorting improved to generate the corresponding second sorting.

[0114] Select the corresponding display method, and based on the display method, sort and display the third search result according to the second sorting. The display method can include various forms such as lists, abstracts, and pictures to facilitate users to quickly browse and understand the search results.

[0115] 1. Through deep models + large models for in-depth semantic understanding, accurately analyze the deep intentions and contexts of user queries, reduce irrelevant results caused by keyword ambiguity, and be able to quickly generate candidate results based on the reasoning ability of the large model to optimize search efficiency.

[0116] 2. Combine dynamic optimization of business entity compression structures to ensure that search results highly match user needs and business rules, improve the relevance of search results, and be able to support the automated parsing and response of multi-conditions and implicit requirements (such as price ranges, scenario preferences) to enhance the adaptability to complex queries.

[0117] In one embodiment, it is also possible to obtain the operation information of the user on the third search result, including information such as clicks, browsing duration, and evaluations.

[0118] According to the operation information, obtain the personalized behavior information of the user. The personalized behavior information reflects the user's search habits and needs and understands the user's satisfaction with the search results. For example, if the user has more clicks and a longer browsing duration on video websites, it is considered that the user has a greater preference for video websites.

[0119] According to the personalized behavior information, perform personalized optimization on the user's corresponding sorting method and display method. For example, when the user conducts a search later, video websites can be preferentially recommended. Of course, through this personalized behavior information, the large model and deep learning model can also be iteratively optimized to improve the accuracy of search results and the user experience. The optimization methods include adjusting model parameters, increasing training data, and improving model structures.

[0120] As Figure 3 shown, the embodiment of the present application also proposes a search optimization device based on a large model, including:

[0121] At least one processor; and,

[0122] A memory communicatively connected to the at least one processor; wherein,

[0123] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute, for example, the search optimization method based on a large model as described in any of the above embodiments.

[0124] This application also provides in an embodiment a non-volatile computer storage medium storing computer-executable instructions, and the computer-executable instructions are configured to be the search optimization method based on a large model as described in any of the above embodiments.

[0125] The various embodiments in this application are described in a progressive manner. For the same or similar parts among the various embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the device and medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and for the relevant parts, reference can be made to the partial description of the method embodiments.

[0126] The device and medium provided in the embodiments of this application correspond one-to-one to the method. Therefore, the device and medium also have beneficial technical effects similar to those of the corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the device and medium will not be elaborated here.

[0127] The above are only the embodiments of this application and are not intended to limit this application. For those skilled in the art, various changes and modifications can be made to this application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included within the scope of the claims of this application.

Claims

1. A search optimization method based on a large model, characterized in that: include: Receiving query information input by a user and preprocessing the query information; Based on a pre-trained deep learning model, semantic analysis is performed on the query information to extract key information contained therein; Taking the query information and the key information as input, generating a relevant first search result through a pre-trained large model; Acquiring the business entity corresponding to the query information and determining the compressed structure information corresponding to the business entity, specifically comprising: generating a corresponding first business entity according to the key information; expanding the key information based on the corresponding relationship in the local database, thereby generating a corresponding second business entity for each first business entity; compressing each first business entity and its corresponding second business entity according to the corresponding relationship to generate corresponding compressed structure information; The first search result is optimized based on the compressed structure information to obtain a second search result, specifically including: for each compressed structure information corresponding to the first business entity, selecting at least one second business entity as a selected entity corresponding to the compressed structure information; combining selected entities corresponding to all compressed structure information to obtain an entity combination; performing a correlation analysis on the first search result according to the entity combination, and taking a first search result with a first correlation score higher than a preset score as a sub-search result corresponding to the entity combination; for each sub-search result, selecting a number of search results with the highest first correlation score, and combining them after deduplication to obtain a second search result; The second search results are sorted and displayed.

2. The search optimization method based on a large model according to claim 1, characterized in that: Preprocessing the query information specifically includes: According to the query information, obtain a corresponding character string; Performing text preprocessing on the character string; The character string is segmented and part-of-speech tagging is performed on the segmentation result.

3. The search optimization method based on a large model according to claim 1, characterized in that: Based on the pre-trained deep learning model, the query information is semantically analyzed to extract the key information contained therein, including: Determine, based on a pre-trained deep learning model, a minimum text unit contained in a word segmentation result of the query information; Performing position encoding on the minimum text unit; Outputting a corresponding context vector for the minimum text unit; the context vector includes a first vector of aggregated semantic representation for intent classification and a second vector of fine-grained semantic features for entity recognition; Intent classification is performed according to the first vector, and key information is extracted according to the intent classification and the second vector.

4. The large model-based search optimization method according to claim 1, characterized in that: The method further comprises: Determine the number of results of the second search result obtained by combining after deduplication; If the number of results is higher than a preset number range, then based on the number of corresponding second business entities in the entity combination, the corresponding sub-search results are deleted in order from low to high until the number of results meets the preset number range; If the number of results is lower than the preset number range, then based on the number of corresponding second business entities in the entity combination, corresponding sub-search results are added in descending order until the number of results meets the preset number range, or all sub-search results are added.

5. The large model-based search optimization method according to claim 1, characterized in that: Sorting and displaying the second search result specifically includes: Obtain access rights to the user; filtering the second search result according to the access permission to obtain a third search result; Obtaining a second relevance score of the large model for the first search result, and obtaining a first ranking corresponding to the third search result according to the second relevance score; Obtaining access data of the third search result, and obtaining a corresponding quality score according to the access data, and adjusting the first ranking according to the quality score to obtain a second ranking; A corresponding display method is selected, and based on the display method, the third search results are sorted and displayed according to the second sorting.

6. The large model-based search optimization method according to claim 5, characterized in that: The method further comprises: Acquiring user operation information on the third search result; According to the operation information, obtaining the user's personalized behavior information; According to the personalized behavior information, the sorting method and display method corresponding to the user are personalized and optimized.

7. A search optimization device based on a large model, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the large model-based search optimization method as described in any one of claims 1 to 6.

8. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The computer executable instructions are configured as: a large model-based search optimization method as described in any one of claims 1 to 6.

Images