Data processing method, device and equipment and readable storage medium

By generating context content vectors and updating query statements, using semantic understanding to rewrite users' query expressions, the problem of insufficient keyword matching in existing search technology is solved, the accuracy and efficiency of search is improved, and the query that is more in line with user intentions is provided. result.

CN119003576APending Publication Date: 2024-11-22TENCENT TECHNOLOGY (SHENZHEN) CO LTD
View PDF 0 Cites 5 Cited by

Patent Information

Application Number
CN202411067875.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-05
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Existing search technology is difficult to handle complex queries when users enter keywords inaccurate or incompletely, resulting in low correlation and accuracy of search results. Users need to spend time screening and evaluating results, and low retrieval efficiency.

Method used

By obtaining historical interaction information, generating context content vectors and updating query statements, using semantic understanding to rewrite and optimize user's query expression, improving semantic relevance, and providing prompts on the search page to guide users to re-retrieve.

Benefits of technology

It improves the accuracy and efficiency of content retrieval, reduces the time and energy of users in filtering search results, and provides query results that are more in line with user intentions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119003576A_ABST
    Figure CN119003576A_ABST
Patent Text Reader

Abstract

The invention discloses a data processing method and device, equipment and a readable storage medium, and the method comprises the steps: carrying out the content retrieval based on an initial query statement, and obtaining candidate result contents used for carrying out sorting display in a retrieval page; if the interaction operation of the object meets the re-retrieval condition, historical interaction information is obtained; the historical interaction information comprises N historical query statements and a historical interaction result set; the N historical query statements comprise initial query statements, and the historical interaction result set comprises candidate result contents interacted by objects in the candidate result contents; generating a context content vector corresponding to historical interaction result content in the historical interaction information, and generating an update query statement through the context content vector and the N historical query statements; the update query statement is used for being displayed in a retrieval prompt box; the retrieval prompt box is used for guiding the object to re-retrieve by updating the query statement. By adopting the method and the device, the accuracy and the efficiency of content retrieval can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a data processing method, device, equipment and readable storage medium. Background Art

[0002] When users search for a question, they often do not know the keywords of the question well. Existing search technologies are mainly based on keyword matching, and their ability to understand the semantics of queries is weak. If the keywords entered by the user are inaccurate or incomplete, for example, when the user enters part of the keywords of the question, or the synonyms of the keywords in the question, the keyword matching cannot understand the structure of the phrase or sentence, and it is difficult to handle some complex query expressions, such as natural language queries, long-tail queries, etc., which may lead to the inability to find relevant information, and the relevance and accuracy of the search results are low, and the user's true intention cannot be accurately captured. At the same time, the search results of existing technologies often contain a large amount of information, and users need to spend time screening and evaluating the content one by one, and adjust the keywords for multiple searches, which has low search efficiency. Summary of the invention

[0003] The embodiments of the present application provide a data processing method, apparatus, device and readable storage medium, which can improve the accuracy and efficiency of content retrieval.

[0004] On the one hand, an embodiment of the present application provides a data processing method, including:

[0005] Perform content retrieval based on the initial query statement to obtain M candidate result contents for sorting and displaying in the search page; M is a positive integer;

[0006] If the object's interaction operation on the search page meets the re-search condition, historical interaction information is obtained; the historical interaction information includes N historical query statements and a historical interaction result set corresponding to the N historical query statements; the N historical query statements include the initial query statement, and the historical interaction result content in the historical interaction result set includes the candidate result content that the object has interacted with among the M candidate result contents; N is a positive integer;

[0007] Generate a context content vector corresponding to the content of the historical interaction results in the historical interaction information, generate a content text prompt corresponding to each historical query statement through the context content vector and N historical query statements, and generate an update query statement based on the N content text prompts; the update query statement is used to be displayed in a search prompt box on the search page; the search prompt box is used to guide the object to be re-searched through the update query statement.

[0008] Among them, it also includes:

[0009] Generate a business time range based on the query time of the initial query statement, and obtain S business query statements within the business time range except the initial query statement; S is a positive integer;

[0010] Perform feature extraction on the initial query statement to obtain an initial query vector, perform feature extraction on S business query statements respectively to obtain S business query vectors, perform cross-attention processing on the initial query vector and the S business query vectors to obtain attention scores corresponding to the S business query vectors respectively, determine the business query vector whose attention score is greater than or equal to the score threshold as the target business query vector, and determine the business query statement corresponding to the target business query vector as the historical query statement.

[0011] Among them, the initial query vector and the S business query vectors are cross-attention processed to obtain the attention scores corresponding to the S business query vectors, including:

[0012] Perform a dot multiplication operation on the S business query vectors and the query parameter matrix to obtain a query vector, perform a dot multiplication operation on the S business query vectors and the key parameter matrix to obtain a key vector, and perform a dot multiplication operation on the initial query vector and the value parameter matrix to obtain a value vector;

[0013] An attention score vector is generated based on the query vector and the key vector. The attention score vector is reduced in dimension based on the dimension of the key vector. The reduced-dimensional attention score vector is normalized to obtain a normalized vector. The normalized vector is point-multiplied with the value vector to obtain the attention scores corresponding to the S business query vectors.

[0014] The process of generating a context content vector corresponding to the content of the historical interaction result in the historical interaction information includes:

[0015] Based on the interactive operation information corresponding to the historical interactive result content in the historical interactive information, the interactive weight corresponding to the historical interactive result content is generated, and the feature extraction of the historical interactive result content is performed to obtain a content feature vector; the content feature vector is used to represent the context information of the historical interactive result content;

[0016] A context content vector corresponding to the historical interaction result content is generated based on the interaction weight and the content feature vector.

[0017] The interactive operation information includes H types of unit interactive information with different interactive operation types, the number of historical interactive result contents is T, and the T historical interactive result contents include historical interactive result contents A i , the interactive operation information includes the historical interactive result content A i Corresponding interactive operation information B i, i is a positive integer less than or equal to T; based on the interactive operation information corresponding to the historical interactive result content in the historical interactive information, the interactive weight corresponding to the historical interactive result content is generated, including:

[0018] Based on interactive operation information B i The interaction operation type and duration of each unit interaction information in the interaction operation are used to determine the content of the historical interaction results A. i The corresponding interaction score;

[0019] When the interaction scores corresponding to the T historical interaction result contents are obtained, the sum of the interaction scores of the T historical interaction result contents is determined as the total interaction score. i The corresponding interaction score generates the historical interaction result content A i The corresponding interaction weights.

[0020] Among them, feature extraction is performed on the content of historical interaction results to obtain content feature vectors, including:

[0021] Performing text segmentation on the content of the historical interaction results to obtain a word unit sequence corresponding to the content of the historical interaction results; a word unit in the word unit sequence refers to the smallest basic unit obtained after the text information is subjected to text segmentation processing;

[0022] The keyword frequency in the historical interaction result content is feature extracted to obtain the word frequency embedding vector, the semantic information of each word in the historical interaction result content is encoded to obtain the semantic embedding vector, the position information of each word in the historical interaction result content is encoded to obtain the position embedding vector, and the paragraph information of each word in the historical interaction result content is encoded to obtain the paragraph embedding vector; the keyword frequency is the word frequency of the query keyword in the historical interaction result content;

[0023] A content feature vector corresponding to the content of the historical interaction results is generated based on the word frequency embedding vector, semantic embedding vector, position embedding vector and paragraph embedding vector.

[0024] The number of context content vectors is T, where T is a positive integer; the N historical query statements include the historical query statements C i , T context content vectors include historical query sentences C i The corresponding Q context content vectors, Q is a positive integer less than or equal to T; the content text prompt corresponding to each historical query statement is generated through the context content vector and the N historical query statements, and the update query statement is generated based on the N content text prompts, including:

[0025] For historical query statements C iThe corresponding Q context content vectors are concatenated to obtain the historical query statement C i The corresponding global content vector is based on the historical query statement C i and historical query statement C i The corresponding global content vector generates the historical query statement C i Corresponding content text prompt;

[0026] When the content text prompts respectively corresponding to N historical query statements are obtained, an update query statement is generated based on the N content text prompts.

[0027] The update query statement is generated based on the N content text prompts, including:

[0028] Based on the query time sequence between the N historical query statements, generate text prompt weights corresponding to the N content text prompts respectively;

[0029] An update query statement is generated based on the N text prompt weights and the N content text prompts.

[0030] Among them, based on the historical query statement C i and historical query statement C i The corresponding global content vector generates the historical query statement C i The corresponding content text prompts include:

[0031] For historical query statements C i Perform feature extraction to obtain historical query statements C i The corresponding historical query feature vector D i , for the historical query feature vector D i Perform self-attention processing to obtain the historical query feature vector D i The corresponding word-unit score vector; the word-unit score vector contains the historical query statement C i The attention score corresponding to each word in ;

[0032] For historical query statements C i The corresponding global content vector and historical query statement C i The corresponding word-unit score vector is processed with cross attention to obtain the query hint vector, and the query hint vector is decoded to obtain the historical query statement C i The corresponding content text prompt.

[0033] The search prompt box is also used to display the search explanation text; the method also includes:

[0034] Based on historical query statement C i The corresponding Q context content vectors and historical query statements C iCorresponding content text prompt, generate historical query statement C i The corresponding unit explanation text;

[0035] When the unit interpretation texts corresponding to the N historical query statements are obtained, a retrieval interpretation text for the update query statement is generated based on the N unit interpretation texts.

[0036] Among them, it also includes:

[0037] Obtaining interaction scores corresponding to the candidate result contents interacted with by the object in the M candidate result contents; the interaction scores are determined based on the interaction operation type and the interaction operation duration;

[0038] If the number of candidate result contents interacted by the object is greater than or equal to the content number threshold, and the interaction scores corresponding to the candidate result contents interacted by the object are all lower than the interaction score threshold, it is determined that the object's interaction operation on the search page meets the re-search condition.

[0039] Among them, it also includes:

[0040] Obtain the unit word frequency corresponding to each word in the M candidate result contents, and generate the document weights corresponding to the M candidate result contents based on the unit word frequency;

[0041] A search score corresponding to each candidate result content is generated based on the M document weights, and the M candidate result contents are sorted and displayed on the search page based on the search score.

[0042] Among them, it also includes:

[0043] Obtain the query keyword in the initial query statement, perform word frequency statistics on the M candidate result contents based on the query keyword, obtain the keyword frequencies corresponding to the M candidate result contents respectively, and generate the relevance scores corresponding to the M candidate result contents respectively based on the M keyword frequencies; the keyword frequency is the word frequency of the query keyword in the candidate result contents;

[0044] The inverse document frequency corresponding to the query keyword is obtained, and a search score corresponding to each candidate result content is generated based on the inverse document frequency and M relevance scores. The M candidate result contents are sorted and displayed based on the search scores on the search page.

[0045] The M candidate result contents include the target candidate result content; and the relevance scores corresponding to the M candidate result contents are generated based on the M keyword frequencies, including:

[0046] Based on the target document length of the target candidate result content and the average document length of the M candidate result contents, an adjustment factor is generated, and based on the adjustment factor and the keyword frequency corresponding to the target candidate result content, a relevance score corresponding to the target candidate result content is generated.

[0047] Among them, it also includes:

[0048] Obtain the object question text, perform feature extraction on the object question text, obtain the object question vector, and obtain the global content vectors corresponding to the N historical query statements corresponding to the update query statement; the object question text is input by the object in the search prompt box on the search page for the update query statement, and the global content vector is generated based on the historical interaction result content corresponding to the historical query statement;

[0049] The object question vector and the global content vector are cross-attention processed to obtain a question result vector, and a question result text is generated based on the question result vector; the question result text is used to be displayed in the search prompt box on the search page.

[0050] An embodiment of the present application provides a data processing device, including:

[0051] The initial search module is used to search for content based on the initial query statement, and obtain M candidate result contents for sorting and displaying in the search page; M is a positive integer;

[0052] The information acquisition module is used to acquire historical interaction information if the interaction operation of the object on the search page meets the re-search condition; the historical interaction information includes N historical query statements and a historical interaction result set corresponding to the N historical query statements; the N historical query statements include an initial query statement, and the historical interaction result content in the historical interaction result set includes the candidate result content interacted by the object among the M candidate result contents; N is a positive integer;

[0053] A statement update module is used to generate a context content vector corresponding to the content of the historical interaction results in the historical interaction information, generate a content text prompt corresponding to each historical query statement through the context content vector and N historical query statements, and generate an updated query statement based on the N content text prompts; the updated query statement is used to be displayed in a search prompt box on the search page; the search prompt box is used to guide the object to be re-searched through the updated query statement.

[0054] In a possible implementation, the initial query module is further configured to perform the following operations:

[0055] Generate a business time range based on the query time of the initial query statement, and obtain S business query statements within the business time range except the initial query statement; S is a positive integer;

[0056] Perform feature extraction on the initial query statement to obtain an initial query vector, perform feature extraction on S business query statements respectively to obtain S business query vectors, perform cross-attention processing on the initial query vector and the S business query vectors to obtain attention scores corresponding to the S business query vectors respectively, determine the business query vector whose attention score is greater than or equal to the score threshold as the target business query vector, and determine the business query statement corresponding to the target business query vector as the historical query statement.

[0057] In a possible implementation, the initial query module is used to perform cross-attention processing on the initial query vector and the S business query vectors, and when the attention scores corresponding to the S business query vectors are obtained, the following operations are specifically performed:

[0058] Perform a dot multiplication operation on the S business query vectors and the query parameter matrix to obtain a query vector, perform a dot multiplication operation on the S business query vectors and the key parameter matrix to obtain a key vector, and perform a dot multiplication operation on the initial query vector and the value parameter matrix to obtain a value vector;

[0059] An attention score vector is generated based on the query vector and the key vector. The attention score vector is reduced in dimension based on the dimension of the key vector. The reduced-dimensional attention score vector is normalized to obtain a normalized vector. The normalized vector is point-multiplied with the value vector to obtain the attention scores corresponding to the S business query vectors.

[0060] In a possible implementation, when the statement updating module is used to generate a context content vector corresponding to the content of the historical interaction result in the historical interaction information, it is specifically used to perform the following operations:

[0061] Based on the interactive operation information corresponding to the historical interactive result content in the historical interactive information, the interactive weight corresponding to the historical interactive result content is generated, and the feature extraction of the historical interactive result content is performed to obtain a content feature vector; the content feature vector is used to represent the context information of the historical interactive result content;

[0062] A context content vector corresponding to the historical interaction result content is generated based on the interaction weight and the content feature vector.

[0063] In a possible implementation, the interactive operation information includes H types of unit interactive information with different interactive operation types, the number of historical interactive result contents is T, and the T historical interactive result contents include historical interactive result contents A i , the interactive operation information includes the historical interactive result content A i Corresponding interactive operation information B i, i is a positive integer less than or equal to T; the statement update module is used to generate the interaction weight corresponding to the historical interaction result content based on the interaction operation information corresponding to the historical interaction result content in the historical interaction information, and specifically to perform the following operations:

[0064] Based on interactive operation information B i The interaction operation type and duration of each unit interaction information in the interaction operation are used to determine the content of the historical interaction results A. i The corresponding interaction score;

[0065] When the interaction scores corresponding to the T historical interaction result contents are obtained, the sum of the interaction scores of the T historical interaction result contents is determined as the total interaction score. i The corresponding interaction score generates the historical interaction result content A i The corresponding interaction weights.

[0066] In a possible implementation, the statement update module is used to extract features from the content of the historical interaction results, and when obtaining the content feature vector, it is specifically used to perform the following operations:

[0067] Performing text segmentation on the content of the historical interaction results to obtain a word unit sequence corresponding to the content of the historical interaction results; a word unit in the word unit sequence refers to the smallest basic unit obtained after the text information is subjected to text segmentation processing;

[0068] The keyword frequency in the historical interaction result content is feature extracted to obtain the word frequency embedding vector, the semantic information of each word in the historical interaction result content is encoded to obtain the semantic embedding vector, the position information of each word in the historical interaction result content is encoded to obtain the position embedding vector, and the paragraph information of each word in the historical interaction result content is encoded to obtain the paragraph embedding vector; the keyword frequency is the word frequency of the query keyword in the historical interaction result content;

[0069] A content feature vector corresponding to the content of the historical interaction results is generated based on the word frequency embedding vector, semantic embedding vector, position embedding vector and paragraph embedding vector.

[0070] In a possible implementation, the number of context content vectors is T, where T is a positive integer; the N historical query statements include historical query statements C i , T context content vectors include historical query sentences C i The corresponding Q context content vectors, Q is a positive integer less than or equal to T; the statement update module is used to generate the content text prompt corresponding to each historical query statement through the context content vector and the N historical query statements, and when the update query statement is generated based on the N content text prompts, it is specifically used to perform the following operations:

[0071] For historical query statements C i The corresponding Q context content vectors are concatenated to obtain the historical query statement C i The corresponding global content vector is based on the historical query statement C i and historical query statement C i The corresponding global content vector generates the historical query statement C i Corresponding content text prompt;

[0072] When the content text prompts respectively corresponding to N historical query statements are obtained, an update query statement is generated based on the N content text prompts.

[0073] In a possible implementation, when the statement update module is used to generate an update query statement based on N content text prompts, it is specifically used to perform the following operations:

[0074] Based on the query time sequence between the N historical query statements, generate text prompt weights corresponding to the N content text prompts respectively;

[0075] An update query statement is generated based on the N text prompt weights and the N content text prompts.

[0076] In a possible implementation, the statement update module is used to update the historical query statement C i and historical query statement C i The corresponding global content vector generates the historical query statement C i When the corresponding content text prompt is displayed, it is used to perform the following operations:

[0077] For historical query statements C i Perform feature extraction to obtain historical query statements C i The corresponding historical query feature vector D i , for the historical query feature vector D i Perform self-attention processing to obtain the historical query feature vector D i The corresponding word-unit score vector; the word-unit score vector contains the historical query statement C i The attention score corresponding to each word in ;

[0078] For historical query statements C i The corresponding global content vector and historical query statement C i The corresponding word-unit score vector is processed with cross attention to obtain the query hint vector, and the query hint vector is decoded to obtain the historical query statement C i The corresponding content text prompt.

[0079] In a possible implementation, the search prompt box is further used to display the search explanation text; and the statement update module is further used to perform the following operations:

[0080] Based on historical query statement C i The corresponding Q context content vectors and historical query statements C i Corresponding content text prompt, generate historical query statement C i The corresponding unit explanation text;

[0081] When the unit interpretation texts corresponding to the N historical query statements are obtained, a retrieval interpretation text for the update query statement is generated based on the N unit interpretation texts.

[0082] In a possible implementation, the content acquisition module is further configured to perform the following operations:

[0083] Obtaining interaction scores corresponding to the candidate result contents interacted with by the object in the M candidate result contents; the interaction scores are determined based on the interaction operation type and the interaction operation duration;

[0084] If the number of candidate result contents interacted by the object is greater than or equal to the content number threshold, and the interaction scores corresponding to the candidate result contents interacted by the object are all lower than the interaction score threshold, it is determined that the object's interaction operation on the search page meets the re-search condition.

[0085] In a possible implementation, the initial query module is further configured to perform the following operations:

[0086] Obtain the unit word frequency corresponding to each word in the M candidate result contents, and generate the document weights corresponding to the M candidate result contents based on the unit word frequency;

[0087] A search score corresponding to each candidate result content is generated based on the M document weights, and the M candidate result contents are sorted and displayed on the search page based on the search score.

[0088] In a possible implementation, the initial query module is further configured to perform the following operations:

[0089] Obtain the query keyword in the initial query statement, perform word frequency statistics on the M candidate result contents based on the query keyword, obtain the keyword frequencies corresponding to the M candidate result contents respectively, and generate the relevance scores corresponding to the M candidate result contents respectively based on the M keyword frequencies; the keyword frequency is the word frequency of the query keyword in the candidate result contents;

[0090] The inverse document frequency corresponding to the query keyword is obtained, and a search score corresponding to each candidate result content is generated based on the inverse document frequency and M relevance scores. The M candidate result contents are sorted and displayed based on the search scores on the search page.

[0091] In a possible implementation, the M candidate result contents include target candidate result contents; when the initial query module is used to generate the relevance scores corresponding to the M candidate result contents respectively based on the M keyword frequencies, it is specifically used to perform the following operations:

[0092] In a possible implementation, the statement update module is further configured to perform the following operations:

[0093] Obtain the object question text, perform feature extraction on the object question text, obtain the object question vector, and obtain the global content vectors corresponding to the N historical query statements corresponding to the update query statement; the object question text is input by the object in the search prompt box on the search page for the update query statement, and the global content vector is generated based on the historical interaction result content corresponding to the historical query statement;

[0094] The object question vector and the global content vector are cross-attention processed to obtain a question result vector, and a question result text is generated based on the question result vector; the question result text is used to be displayed in the search prompt box on the search page.

[0095] On the one hand, an embodiment of the present application provides a computer device, including: a processor, a memory, and a network interface;

[0096] The processor is connected to the memory and the network interface, wherein the network interface is used to provide a data communication function, and the memory is used to store a computer program. When the computer program is executed by the processor, the computer device executes the method provided in the embodiment of the present application.

[0097] On the one hand, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. The computer program is suitable for being loaded and executed by a processor so that a computer device having the processor executes the method provided by the embodiment of the present application.

[0098] In one aspect, an embodiment of the present application provides a computer program product, which includes a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the method provided in the embodiment of the present application.

[0099] The embodiment of the present application performs content retrieval based on the initial query statement to obtain M candidate result contents for sorting and displaying in the search page. If the interactive operation of the object on the search page meets the re-search condition, the historical interaction information is obtained, wherein the historical interaction information includes N historical query statements and a historical interaction result set corresponding to the N historical query statements, the N historical query statements include the initial query statement, and the historical interaction result content in the historical interaction result set includes the candidate result content that the object has interacted with in the M candidate result contents. A context content vector corresponding to the historical interaction result content in the historical interaction information is generated, an update query statement is generated through the context content vector and the N historical query statements, a context content vector representing the context information in each historical interaction result content is generated through semantic understanding, the interaction history and query expression of the object are learned and understood through the N historical query statements, and the N historical query statements are rewritten and optimized based on the context content vector, so that the user's true intention can be better understood, and a semantically related update query statement can be obtained. Therefore, searching through the update query statement can improve the accuracy of content retrieval. When the object's interactive operation on the search page meets the re-search conditions, the system can actively display the updated query statement rewritten by the above method in the search prompt box on the search page, and guide the object to re-search through the updated query statement through the search prompt box. There is no need for the object to understand all the result content and adjust keywords for active re-search, which greatly improves the efficiency of content screening and content retrieval. BRIEF DESCRIPTION OF THE DRAWINGS

[0100] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0101] Figure 1 This is a schematic diagram of a network architecture provided by an embodiment of the present application;

[0102] Figure 2 This is a data processing scenario provided by the embodiment of the present application. Figure 1 ;

[0103] Figure 3 This is a flow diagram of a data processing method provided in an embodiment of the present application. Figure 1 ;

[0104] Figure 4 This is a data processing scenario provided by the embodiment of the present application. Figure 2 ;

[0105] Figure 5 This is a flow diagram of a data processing method provided in an embodiment of the present application. Figure 2 ;

[0106] Figure 6 is a structural schematic diagram of a data processing device provided in an embodiment of the present application;

[0107] Figure 7 It is a structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0108] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0109] It can be understood that in the specific implementation of the present application, the user (object or player) data involved, when the above and following embodiments of the present application are applied to specific products or technologies, needs to obtain user permission or consent, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of the relevant regions.

[0110] Among them, if it is necessary to collect object (such as user, etc.) data in this application, a prompt interface or pop-up window will be displayed before or during the collection. The prompt interface or pop-up window is used to prompt the user that certain data is currently being collected. Only after the user confirms the prompt interface or pop-up window, the relevant steps of data acquisition will be started, otherwise the process will end. Moreover, the acquired user data will be used in reasonable and legal scenarios or purposes. Optionally, in some scenarios where user data needs to be used but has not been authorized by the user, authorization can be requested from the user, and the user data can be used when the authorization is passed.

[0111] See also Figure 1 , Figure 1 Schematic diagram of a network architecture provided by an embodiment of the present application. Figure 1As shown, the network architecture may include a business server 100 and a terminal device cluster, and the terminal device cluster may include terminal device 10a, terminal device 10b, ..., terminal device 10n, wherein any terminal device in the terminal device cluster may have a communication connection with the business server 100, for example, there is a communication connection between the terminal device 10a and the business server 100, and there is a communication connection between the terminal device 10b and the business server 100, wherein the above-mentioned communication connection does not limit the connection method, and may be directly or indirectly connected through a wired communication method, or directly or indirectly connected through a wireless communication method, or through other methods, and the present application does not make any restrictions here.

[0112] Each terminal device in the terminal device cluster may include: smart phones, tablet computers, laptop computers, desktop computers, intelligent voice interaction devices, smart home appliances (e.g., smart TVs), wearable devices, vehicle terminals, aircraft, and other smart terminals with data processing functions. The vehicle terminal may be a terminal device in smart transportation scenarios and assisted driving scenarios. It should be understood that Figure 1 Each terminal device in the terminal device cluster shown in the figure can be installed with an application client having a data processing function. When the application client runs in each terminal device, it can be respectively connected to the above Figure 1 The business servers 100 shown perform data interaction.

[0113] The application client may specifically include: a vehicle client, a smart home client, an entertainment client (e.g., a game client), a multimedia client (e.g., a video client), a social client, and an information client (e.g., a news client), etc. The application client in the embodiment of the present application may be integrated in a client (e.g., a social client) or may be an independent client (e.g., a news client). The embodiment of the present application does not limit the type of the application client.

[0114] Among them, the business server 100 can be a server corresponding to the application client. The business server 100 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), as well as big data and artificial intelligence platforms.

[0115] like Figure 1As shown, any terminal device in the terminal device cluster can send a content query request to the business server 100. Taking the terminal device 10a as an example, the terminal device 10a can send a content query request to the business server 100. The business server 100 can perform content retrieval based on the initial query statement in the content query request to obtain M candidate result contents, and sort the M candidate result contents based on the retrieval score to obtain a candidate result content sequence. Among them, the initial query statement can be a query word or query statement entered by the object (user) in the application client to express the information needs of the object. The retrieval score is determined based on the degree of association between the candidate result content and the initial query statement.

[0116] The business server 100 can send the candidate result content sequence to the terminal device 10a, and the terminal device 10a can sort and display the M candidate result contents in the search page of the application client. The object can view the candidate result contents in the search page. If the object's interactive operation on the search page meets the re-search condition, the terminal device 10a can send a re-search request to the business server 100, and the business server 100 can obtain historical interaction information based on the re-search request. Among them, the historical interaction information includes N historical query statements and a historical interaction result set corresponding to the N historical query statements, the N historical query statements include an initial query statement, and the historical interaction result content in the historical interaction result set includes the candidate result content that has been interacted with by the object among the M candidate result contents. For example, if the number of candidate result contents that the object has interacted with reaches a certain number, it is determined that the object's interactive operation on the search page meets the re-search condition.

[0117] The business server 100 can generate a context content vector corresponding to each historical interaction result content based on the historical interaction result content, and generate a content text prompt corresponding to each historical query statement through the context content vector and N historical query statements. The content text prompt can be a prompt, and the business server 100 can generate an update query statement based on the N content text prompts. For example, the historical interaction result set can only include candidate result content that the object has interacted with, and the N historical query statements can only include initial query statements. The business server 100 can generate several context content vectors based on the candidate result content that the object has interacted with, generate a content text prompt through the context content vector, and then generate an update query statement corresponding to the initial query statement based on the content text prompt.

[0118] The business server 100 may send the update query statement to the terminal device 10a, and the terminal device 10a may display a search prompt box in the search page of the application client, display the update query statement in the search prompt box in the search page, and guide the object to re-search through the update query statement through the search prompt box.

[0119] The embodiment of the present application generates an update query statement through a context content vector and N historical query statements, generates a context content vector representing context information in each historical interaction result content through semantic understanding, learns to understand the interaction history and query expression of the object through N historical query statements, and rewrites and optimizes the N historical query statements based on the context content vector, so as to better understand the user's true intention and obtain semantically related update query statements. Therefore, searching through the update query statement can improve the accuracy of content retrieval.

[0120] See also Figure 2 , Figure 2 This is a data processing scenario provided by an embodiment of the present application. Figure 1 .like Figure 2 As shown, the business server can be Figure 1 The service server 100 shown in FIG. 1 and the terminal device 100 may be as follows: Figure 1 Any terminal device in the terminal device cluster shown may be, for example, terminal device 10a. In terminal device 10a, the object may input an initial query statement to search for content, and terminal device 10a may send a content query request to business server 100. Business server 100 may perform content retrieval based on the initial query statement in the content query request to obtain M candidate result contents, and sort the M candidate result contents based on the retrieval score to obtain a candidate result content sequence. The initial query statement may be a query word or query statement input by the object (user) in the application client to express the object's information needs. The retrieval score is determined based on the degree of association between the candidate result contents and the initial query statement.

[0121] The business server 100 can send the candidate result content sequence to the terminal device 10a, and the terminal device 10a can sort and display the M candidate result contents in the search page of the application client. As shown in page 101, the application client can display candidate result content 1, candidate result content 2, candidate result content 3 and candidate result content 4 among the M candidate result contents in page 1 of the search page. The object can view the candidate result content in the search page. As shown in page 102, the object browses the candidate result content 2. The terminal device 10a can record the object's interactive operation information for the candidate result content 2. The interactive operation information may include the interactive operation type and the interactive operation duration. The interactive operation type may be, for example, a browsing operation, a like operation, a comment operation, etc., which is not limited in the embodiments of the present application.

[0122] As shown in page 103, the object continues to view candidate result content 5, candidate result content 6, candidate result content 7 and candidate result content 8 among the M candidate result contents on page 2 of the search page, and the terminal device 10a obtains the interaction scores corresponding to the candidate result contents interacted by the object in the search page, and the interaction scores are determined based on the interaction operation type and interaction operation duration corresponding to the object. If the number of candidate result contents interacted by the object is greater than or equal to the content quantity threshold, and the interaction scores corresponding to the candidate result contents interacted by the object are all lower than the interaction score threshold (pre-set value), it is determined that the interaction operation of the object for the search page meets the re-search condition, and the terminal device 10a can send a re-search request to the business server 100. Among them, the re-search request can include the interaction operation information of the candidate result contents interacted by the object. If the number of candidate result contents interacted by the object reaches a certain number, it is determined that the interaction operation of the object for the search page meets the re-search condition.

[0123] The business server 100 may obtain historical interaction information based on the re-retrieval request, wherein the historical interaction information includes N historical query statements and a historical interaction result set corresponding to the N historical query statements, the N historical query statements include the initial query statement, and the historical interaction result content in the historical interaction result set includes the candidate result content that has been interacted by the object among the M candidate result contents.

[0124] The business server 100 can generate a context content vector corresponding to each historical interaction result content based on the historical interaction result content, and the context content vector can include context information representing the frequency, distribution, and global semantics of keywords. The business server 100 can generate a content text prompt corresponding to each historical query statement through the context content vector and N historical query statements, and update the query statement based on the N content text prompts. For example, the N historical query statements can only include the initial query statement, and the object has only interacted with candidate result content 2, candidate result content 3, and candidate result content 5 in the search page corresponding to the initial query statement. Then the business server 100 can generate context content vector 2 based on the context information corresponding to candidate result content 2, generate context content vector 3 based on the context information corresponding to candidate result content 3, and generate context content vector 5 based on the context information corresponding to candidate result content 5. The business server 100 can generate a content text prompt based on context content vector 2, context content vector 3, and context content vector 5. The content text prompt can be a prompt, and the business server 100 can generate an updated query statement corresponding to the initial query statement through the content text prompt.

[0125] The business server 100 may send the update query statement to the terminal device 10a, as shown in page 104, and the terminal device 10a may display a search prompt box W in the search page of the application client, display the update query statement in the search prompt box W in the search page, and guide the object to re-search through the update query statement through the search prompt box W. For example, the object may click the "Confirm" control in the search prompt box W, and the terminal device 10a may re-search based on the update query statement and display the candidate result content retrieved by the update query statement.

[0126] The embodiment of the present application generates a context content vector corresponding to the content of the historical interaction results in the historical interaction information, generates an update query statement through the context content vector and N historical query statements, and generates a context content vector representing the context information in each historical interaction result content through semantic understanding, so as to better understand the user's true intention and obtain semantically related update query statements. The update query statement has a wide range of applications in many fields. The update query statement can be used for re-retrieval or for content recommendation. For example, in social media applications, the update query statement can identify the user's content preferences based on their interests and behaviors, and recommend related videos or articles. In an automatic customer service system, the update query statement can generate a more accurate question expression based on the user's usage habits and consultation history information, improve the processing efficiency of manual customer service, enhance the user experience, and make the service more personalized and intelligent.

[0127] See also Figure 3 , Figure 3 This is a flow diagram of a data processing method provided in an embodiment of the present application. Figure 1 The data processing method can be executed by a computer device, which can be Figure 1 The service server 100 or any terminal device in the terminal device cluster shown may be, for example, the terminal device 10a. The following will take the data processing method executed by a computer device as an example for explanation. The data processing method may at least include the following steps S101 to S103:

[0128] Step S101, performing content retrieval based on the initial query statement, and obtaining M candidate result contents for sorting and displaying in the search page; M is a positive integer;

[0129] Specifically, the computer device can perform content retrieval based on the initial query statement to obtain M candidate result contents, sort the M candidate result contents based on the retrieval score, and obtain a candidate result content sequence. The initial query statement can be a query word or query statement entered by the object (user) in the application client to express the information needs of the object. The retrieval score is determined based on the degree of association between the candidate result contents and the initial query statement. The candidate result contents can be multimedia data, and can be a combination of multiple types of media information, such as documents, images, audio, video, etc.

[0130] Content sorting can be based on methods such as document frequency (Term Frequency-Inverse Document Frequency, TF-IDF), relevance probability (Best Matching 25, BM25), word embedding recall (Embedding), etc., and the embodiments of the present application are not limited here.

[0131] Step S102, if the object's interaction operation on the search page meets the re-search condition, then obtain historical interaction information; the historical interaction information includes N historical query statements and a historical interaction result set corresponding to the N historical query statements; the N historical query statements include the initial query statement, and the historical interaction result content in the historical interaction result set includes the candidate result content that the object has interacted with among the M candidate result contents; N is a positive integer;

[0132] Specifically, the object can view the candidate result content in the search page, the object can interact with each candidate result content in the search page, and the computer device can detect whether the object's interactive operation meets the re-search condition during the object interaction. If the object's interactive operation on the search page meets the re-search condition, the computer device can obtain historical interaction information based on the re-search request. Among them, the historical interaction information includes N historical query statements and a historical interaction result set corresponding to the N historical query statements, the N historical query statements include the initial query statement, and the historical interaction result content in the historical interaction result set includes the candidate result content that the object has interacted with in the M candidate result content. The historical query statement can be other query statements associated with the initial query statement input by the object within the business time range determined by the initial query statement. The historical interaction result set includes the result content that the object has interacted with under the N historical query statements. The re-search condition can refer to a condition that prompts the object to re-search. If the number of candidate result contents that the object has interacted with reaches a certain number, it is determined that the object's interactive operation on the search page meets the re-search condition.

[0133] Step S103, generate a context content vector corresponding to the content of the historical interaction results in the historical interaction information, generate a content text prompt corresponding to each historical query statement through the context content vector and N historical query statements, and generate an update query statement based on the N content text prompts; the update query statement is used to be displayed in the search prompt box in the search page; the search prompt box is used to guide the object to be re-searched through the update query statement.

[0134] For details, please refer to Figure 4 , Figure 4 This is a data processing scenario proposed in the embodiment of the present application. Figure 2 ,like Figure 4 As shown, the computer device can obtain N historical query statements and a set of historical interaction results in the historical interaction information. The N historical query statements may include historical query statement 1, historical query statement 2, ..., historical query statement N. The computer device may generate a context content vector corresponding to each historical interaction result content based on the historical interaction result content. The context content vector may contain context information representing keyword frequency, topic distribution, and global semantics. The computer device may generate a content text prompt corresponding to each historical query statement through the context content vector and the N historical query statements. The content text prompt may be a prompt. An update query statement is generated based on the N content text prompts. The N content text prompts may include content text prompt 1 corresponding to historical query statement 1, content text prompt 2 corresponding to historical query statement 2, ..., content text prompt N corresponding to historical query statement N. For example, the computer device may generate content text prompt 1 corresponding to historical query statement 1 through historical query statement 1 and the context content vector corresponding to historical query statement 1.

[0135] The computer device can generate an update query statement through content text prompt 1, content text prompt 2, ... content text prompt N, and the update query statement can be generated by semantic understanding and rewriting based on N content text prompts through a pre-trained large language model. The large language model can be BERT (Bidirectional Encoder Representations from Transformers), GPT (Generative Pretrained Transformer), etc., wherein the update query statement is used to be displayed in a search prompt box on the search page, and the search prompt box is used to guide the object to be re-searched through the update query statement.

[0136] The embodiment of the present application can rewrite and optimize the context content vector and N historical query statements through a pre-trained large language model. When the object clicks on certain search results (historical interaction result content), the system can organize the content of the search results into a context, such as a context content vector, and form a prompt based on the context content vector to input a large language model. The large language model can rewrite the initial query statement or N historical query statements based on the prompt to generate a more accurate query expression that meets the user's intention. The process of generating and updating the query local area can be a real-time feedback mechanism, which can enable the system to dynamically adjust the retrieval strategy and continuously optimize the query and results, thereby improving the relevance of the retrieval and user satisfaction.

[0137] It can be understood that the updated query statement can be recommended to the object in the form of a suggestion for the object to select and refer to. The object can directly use the rewritten query, or further modify and improve it. The updated query statement can be used in the next round of retrieval to obtain more accurate and relevant result content. The computer device can continuously record the object's interactive operation information and the corresponding historical interactive result content, repeatedly extract and generate the context content vector corresponding to the historical interactive result content, and realize dynamic optimization of queries and results.

[0138] The embodiment of the present application performs content retrieval based on the initial query statement to obtain M candidate result contents for sorting and displaying in the search page. If the interactive operation of the object on the search page meets the re-search condition, the historical interaction information is obtained, wherein the historical interaction information includes N historical query statements and a historical interaction result set corresponding to the N historical query statements, the N historical query statements include the initial query statement, and the historical interaction result content in the historical interaction result set includes the candidate result content that the object has interacted with in the M candidate result contents. A context content vector corresponding to the historical interaction result content in the historical interaction information is generated, an update query statement is generated through the context content vector and the N historical query statements, a context content vector representing the context information in each historical interaction result content is generated through semantic understanding, the interaction history and query expression of the object are learned and understood through the N historical query statements, and the N historical query statements are rewritten and optimized based on the context content vector, so that the user's true intention can be better understood, and a semantically related update query statement can be obtained. Therefore, searching through the update query statement can improve the accuracy of content retrieval. When the object's interactive operation on the search page meets the re-search conditions, the system can actively display the updated query statement rewritten by the above method in the search prompt box on the search page, and guide the object to re-search through the updated query statement through the search prompt box. There is no need for the object to understand all the result content and adjust keywords for active re-search, which greatly improves the efficiency of content screening and content retrieval.

[0139] See also Figure 5 , Figure 5 This is a flow diagram of a data processing method provided in an embodiment of the present application. Figure 2 The data processing method can be executed by a computer device, which can be Figure 1 The service server 100 or any terminal device in the terminal device cluster shown may be, for example, the terminal device 10a. The following will take the data processing method executed by a computer device as an example for explanation. The data processing method may at least include the following steps S201 to S205:

[0140] Step S201, performing content retrieval based on the initial query statement, and obtaining M candidate result contents for sorting and displaying in the search page; M is a positive integer;

[0141] Specifically, the computer device can perform content retrieval based on the initial query statement to obtain M candidate result contents, sort the M candidate result contents based on the retrieval score, and obtain a candidate result content sequence. The initial query statement can be a query word or query statement entered by the object (user) in the application client to express the information needs of the object. The retrieval score is determined based on the degree of association between the candidate result contents and the initial query statement. The candidate result contents can be multimedia data, and can be a combination of multiple types of media information, such as documents, images, audio, video, etc.

[0142] Optionally, the M candidate result contents may be obtained based on document frequency sorting, and the process may be: obtaining the unit word frequency corresponding to each word in the M candidate result contents, generating the document weights corresponding to the M candidate result contents based on the unit word frequency; generating the search score corresponding to each candidate result content based on the M document weights, and sorting and displaying the M candidate result contents based on the search score on the search page.

[0143] Specifically, the computer device can obtain the unit term frequency (Term Frequency) corresponding to each word (term) in the M candidate result contents, and generate the document weights corresponding to the M candidate result contents based on the unit term frequency. i,j For example, the document weight w i,j It can be determined based on unit word frequency and inverse document frequency, and its calculation process can be shown as formula (1):

[0144]

[0145] Among them, term i can be any term in the document (candidate result content), tf i,jis the unit word frequency of term i in document j, M represents the total number of documents, df i represents the number of documents in the document collection that contain term i, Represents the inverse document frequency of term i. The inverse document frequency of term i can also be expressed as IDF(i).

[0146] The computer device can be based on the M document weights w i,j Generate a search score corresponding to each candidate result content. The search score can be based on the score of different terms i and the document weight w i,j The computer device can sort and display the M candidate result contents based on the search scores in the search page.

[0147] It can be understood that for document j, if a term only appears in this document, the higher its importance, the higher its document weight w in document j. i,j At the same time, if a term appears in many documents, its document weight w i,j Will be lower.

[0148] Optionally, the M candidate result contents may be obtained based on relevance probability sorting, and the process may be: obtaining the query keyword in the initial query statement, performing word frequency statistics on the M candidate result contents based on the query keyword, obtaining the keyword frequencies corresponding to the M candidate result contents, and generating relevance scores corresponding to the M candidate result contents based on the M keyword frequencies; the keyword frequency is the word frequency of the query keyword in the candidate result contents; obtaining the inverse document frequency corresponding to the query keyword, generating a retrieval score corresponding to each candidate result content based on the inverse document frequency and the M relevance scores, and sorting and displaying the M candidate result contents based on the retrieval score on the retrieval page.

[0149] Specifically, for the initial query statement Q, the initial query statement Q may include the query keyword q i The computer device can perform word frequency statistics on the M candidate result contents (document D) based on the query keywords, and obtain the keyword frequency f(q i ,D), based on the M keyword frequencies f(q i ,D) Generate the relevance scores corresponding to the content of M candidate results, and obtain the inverse document frequency IDF (q i ), inverse document frequency IDF(q i ) can refer to the calculation method of the above formula (1), which will not be repeated in detail in the embodiments of the present application.

[0150] The computer device can be based on the inverse document frequency IDF (q i) and M relevance scores generate the search score score(D,Q) corresponding to each candidate result content. The M candidate result contents are sorted and displayed based on the search score on the search page. The calculation process can be shown as formula (2):

[0151]

[0152] Where D represents the document, i.e., the content of M candidate results, |D| represents the length of document D, avgdl represents the average length of the document set, and k1 and b are adjustment factors. k1 is used to adjust the effect of word frequency on the retrieval score, and the usual value range is 0.1 to 2.0. b is used to adjust the effect of document length on the retrieval score, and the usual value range is 0.0 to 1.0.

[0153] It is understandable that the embodiment of the present application can input the initial query statement into a large language model, such as a BERT model, a GPT model, etc., to improve the semantic understanding ability of the initial query, rather than directly using keyword matching to perform content retrieval. The large language model rewrites and expands the initial query statement to generate semantically relevant query expressions. This semantic-level matching makes the retrieval results more accurate and comprehensive, makes up for the shortcomings of keyword matching, and improves the accuracy of content retrieval.

[0154] Step S202: if the object's interaction operation on the search page meets the re-search condition, historical interaction information is obtained; the historical interaction information includes N historical query statements and a historical interaction result set corresponding to the N historical query statements; the N historical query statements include an initial query statement, and the historical interaction result content in the historical interaction result set includes the candidate result content that the object has interacted with among the M candidate result contents; N is a positive integer;

[0155] Specifically, the object can view the candidate result content in the search page, the object can interact with each candidate result content in the search page, and the computer device can detect whether the object's interactive operation meets the re-search condition during the object interaction. If the object's interactive operation on the search page meets the re-search condition, the computer device can obtain historical interaction information based on the re-search request. Among them, the historical interaction information includes N historical query statements and a historical interaction result set corresponding to the N historical query statements, the N historical query statements include the initial query statement, and the historical interaction result content in the historical interaction result set includes the candidate result content that the object has interacted with in the M candidate result content. The historical query statement can be other query statements associated with the initial query statement input by the object within the business time range determined by the initial query statement. The historical interaction result set includes the result content that the object has interacted with under the N historical query statements. The re-search condition can refer to a condition that prompts the object to re-search. If the number of candidate result contents that the object has interacted with reaches a certain number, it is determined that the object's interactive operation on the search page meets the re-search condition.

[0156] The process of a computer device determining whether a re-search condition is met may be: obtaining the interaction scores corresponding to the candidate result contents interacted with by the object in M ​​candidate result contents; the interaction scores are determined based on the interaction operation type and the interaction operation duration; if the number of candidate result contents interacted with by the object is greater than or equal to the content number threshold, and the interaction scores corresponding to the candidate result contents interacted with by the object are all lower than the interaction score threshold, then it is determined that the object's interaction operation on the search page meets the re-search condition.

[0157] Specifically, the computer device can obtain the interaction scores corresponding to the candidate result contents interacted by the object in the search page, and the interaction scores are determined based on the interaction operation type and interaction operation duration corresponding to the object. In addition to explicit behaviors such as clicking, the interaction operation information can also integrate implicit behavior data such as mouse movement, scrolling, and staying.

[0158] Optionally, when the object switches the display page of the candidate result content in the search page, for example, when clicking the next page, it can also be determined that the object's interactive operations on the search page meet the re-search conditions. When the object clicks the search box on the search page for entering the initial query statement or modifies the initial query statement, it can also be determined that the object's interactive operations on the search page meet the re-search conditions. The embodiments of the present application do not limit the specific content of the re-search conditions.

[0159] Among them, the method for determining the historical query statement is: generating a business time range based on the query time of the initial query statement, obtaining S business query statements other than the initial query statement within the business time range; S is a positive integer; performing feature extraction on the initial query statement to obtain an initial query vector, performing feature extraction on the S business query statements respectively to obtain S business query vectors, performing cross-attention processing on the initial query vector and the S business query vectors to obtain attention scores corresponding to the S business query vectors respectively, determining a business query vector with an attention score greater than or equal to a score threshold as a target business query vector, and determining the business query statement corresponding to the target business query vector as a historical query statement.

[0160] Specifically, the computer device can generate a business time range based on the query time of the initial query statement. For example, the business time range can be the time range from the query time of the initial query statement to 1 hour ago. The computer device can obtain S business query statements within the business time range except the initial query statement, perform feature extraction on the initial query statement to obtain an initial query vector, perform feature extraction on the S business query statements respectively to obtain S business query vectors, perform cross-attention processing on the initial query vector and the S business query vectors, and obtain attention scores corresponding to the S business query vectors respectively. The process of performing cross-attention processing on the initial query vector and the S business query vectors can be: performing a point multiplication operation on the S business query vectors and the query parameter matrix to obtain a query vector, performing a point multiplication operation on the S business query vectors and the key parameter matrix to obtain a key vector, performing a point multiplication operation on the initial query vector and the value parameter matrix to obtain a value vector; generating an attention score vector based on the query vector and the key vector, performing dimensionality reduction processing on the attention score vector based on the dimension of the key vector, normalizing the reduced dimensionality attention score vector to obtain a normalized vector, performing a point multiplication operation on the normalized vector and the value vector to obtain the attention scores corresponding to the S business query vectors respectively.

[0161] Specifically, the computer device can combine the S business query vectors with the query parameter matrix W Q Perform a dot multiplication operation to obtain the query vector Q, and add the S business query vectors to the key parameter matrix W K Perform a dot multiplication operation to obtain the key vector K, and add the initial query vector to the value parameter matrix W V Perform a dot multiplication operation to obtain a value vector V. The computer device can use the transpose operation result K of the query vector Q and the key vector K T Perform a dot multiplication operation to obtain the attention score vector. Based on the dimension value d of the key vector K K The square root of The attention score matrix is ​​reduced in dimension, and the reduced attention score vector is normalized by softmax to obtain a normalized vector. The normalized vector is multiplied by the value vector V to obtain the attention scores corresponding to the S business query vectors. The process can be shown as formula (3):

[0162]

[0163] Step S203, based on the interactive operation information corresponding to the historical interaction result content in the historical interaction information, generate an interactive weight corresponding to the historical interaction result content, perform feature extraction on the historical interaction result content, and obtain a content feature vector; the content feature vector is used to characterize the context information of the historical interaction result content; based on the interactive weight and the content feature vector, generate a context content vector corresponding to the historical interaction result content.

[0164] Specifically, the computer device may generate an interaction weight corresponding to the historical interaction result content based on the interaction operation information corresponding to the historical interaction result content in the historical interaction information. The interaction operation information includes H types of unit interaction information with different interaction operation types, the number of historical interaction result contents is T, and the T historical interaction result contents include historical interaction result contents A i , the interactive operation information includes the historical interactive result content A i Corresponding interactive operation information B i , i is a positive integer less than or equal to T.

[0165] The process of generating the interaction weight corresponding to the historical interaction result content based on the interaction operation information corresponding to the historical interaction result content in the historical interaction information by the computer device may be: based on the interaction operation information B i The interaction operation type and duration of each unit interaction information in the interaction operation are used to determine the content of the historical interaction results A. i Corresponding interaction scores; when the interaction scores corresponding to the T historical interaction result contents are obtained, the sum of the interaction scores of the T historical interaction result contents is determined as the total interaction score. i The corresponding interaction score generates the historical interaction result content A i The corresponding interaction weights.

[0166] Specifically, the object has a historical interaction result content A i All interactive operations may correspond to an interactive operation information, and each interactive operation may correspond to a unit of interactive information. Each interactive operation may include an interactive operation type and an interactive operation duration. The interactive operation type may be, for example, a browsing operation, a like operation, a comment operation, etc., which is not limited in the present embodiment of the application.

[0167] It can be understood that in addition to explicit behaviors such as clicking, interactive operation information can also integrate implicit behavior data such as mouse movement, scrolling, and staying to fully understand user intentions. Computer devices can also use computer vision technology to analyze multimedia information such as pictures and videos browsed by objects, extract visual features, and perform auxiliary intention recognition, thereby generating interaction weights that are more in line with the object's intentions.

[0168] Computer equipment can be based on interactive operation information B i The interaction operation type and duration of each unit interaction information in the interaction operation are used to determine the content of the historical interaction results A. i Corresponding interaction scores. For example, if the browsing operation lasts for 1 minute or less, it can be 10 points, if it lasts for more than 1 minute, it can be 20 points, if the document is browsed by liking, it can be 20 points, if the comment is liked, it can be 10 points, and the historical interaction result content A i The corresponding interaction score may be the sum of the scores of all interaction operations.

[0169] When the interaction scores corresponding to the T historical interaction result contents are obtained, the computer device can determine the sum of the interaction scores of the T historical interaction result contents as the total interaction score. i The corresponding interaction score generates the historical interaction result content A i The corresponding interaction weight, the higher the interaction weight, the more interactions the object has with the content of the historical interaction result.

[0170] The computer device can perform feature extraction on the content of the historical interaction results to obtain a content feature vector, wherein the content feature vector is used to represent the context information of the content of the historical interaction results. The process of extracting features from the content of the historical interaction results can be: performing text segmentation on the content of the historical interaction results to obtain a word-gram sequence corresponding to the content of the historical interaction results; a word-gram in the word-gram sequence refers to the smallest basic unit obtained after the text information is processed by text segmentation; performing feature extraction on the keyword frequency in the content of the historical interaction results to obtain a word frequency embedding vector, encoding the semantic information of each word-gram in the content of the historical interaction results to obtain a semantic embedding vector, encoding the position information of each word-gram in the content of the historical interaction results to obtain a position embedding vector, encoding the paragraph information of each word-gram in the content of the historical interaction results to obtain a paragraph embedding vector; the keyword frequency is the word frequency of the query keyword in the content of the historical interaction results; generating a content feature vector corresponding to the content of the historical interaction results based on the word frequency embedding vector, the semantic embedding vector, the position embedding vector and the paragraph embedding vector.

[0171] Specifically, the computer device can perform text segmentation on the content of the historical interaction results to obtain a word-unit sequence corresponding to the content of the historical interaction results. Text segmentation can be performed by splitting the text information into words (tokens) through a Tokenizer. For example, a certain text information A in the content of the historical interaction results is "Let's do tokenization!", and the word-unit sequence corresponding to the text information A can be expressed as [Let,'s,do,token,ization]. Among them, [Let,'s], [do], [token] and [ization] are all words. A word-unit in a word-unit sequence refers to the smallest basic unit obtained after the text information is split and processed. The method of text segmentation can be based on word segmentation (word-based), character segmentation (character-based), and subword segmentation (subword-based), which is not limited in the embodiments of the present application.

[0172] The computer device can extract the features of the keyword frequency in the historical interaction result content to obtain the word frequency embedding vector. Taking the text information A containing word 1, word 2, word 3 and word 4 as an example, the word frequency embedding can embed the text information A into E TF1 、E TF2 、E TF3 and E TF4 , the word frequency embedding can represent the word frequency of each word in the text information A, where the keyword frequency is the word frequency of the query keyword in the historical interaction result content. The computer device can encode the semantic information of each word in the historical interaction result content to obtain a semantic embedding vector. The semantic embedding can embed the text information A into E 词元1 、E 词元2 、E 词元3 and E 词元4 Semantic embedding can represent the semantic encoding of each word in the text information A respectively.

[0173] The computer device can encode the position information of each word in the historical interaction result content to obtain a position embedding vector. Position embedding can embed text information A into E1, E2, E3 and E4 to capture the position order of each word in text information A and understand the semantic structure. The computer device can encode the paragraph information of each word in the historical interaction result content to obtain a paragraph embedding vector. Paragraph embedding can embed text information A into E1, E2, E3 and E4. A 、E A 、E A and E A , used to distinguish different paragraphs and sentences in text information A. If the embedding of each token is the same, it is the same sentence. If the paragraph embedding of text information A is EA 、E A 、E B and E B , then E A and E A Part represents a sentence, E B and E B Part indicates another sentence.

[0174] The computer device performs vector concatenation based on the word frequency embedding vector, the semantic embedding vector, the position embedding vector and the paragraph embedding vector to obtain a content feature vector corresponding to the content of the historical interaction result. Text feature encoding can be implemented through a BERT model, which is not limited in the embodiments of the present application.

[0175] Step S204, generating an update query statement through the context content vector and N historical query statements;

[0176] Specifically, the number of context content vectors may be T, where T is a positive integer. The N historical query statements may include historical query statements C i , T context content vectors include historical query sentences C i Corresponding to the Q context content vectors, the computer device can analyze the historical query statement C i The corresponding Q context content vectors are concatenated to obtain the historical query statement C i The corresponding global content vector is based on the historical query statement C i and historical query statement C i The corresponding global content vector process can be: i Perform feature extraction to obtain historical query statements C i The corresponding historical query feature vector D i , for the historical query feature vector D i Perform self-attention processing to obtain the historical query feature vector D i The corresponding word-unit score vector; the word-unit score vector contains the historical query statement C i The attention score corresponding to each word in the historical query sentence C i The corresponding global content vector and historical query statement C i The corresponding word-unit score vector is processed with cross attention to obtain the query hint vector, and the query hint vector is decoded to obtain the historical query statement C i The corresponding content text prompt.

[0177] Specifically, the computer device can generate a context content vector corresponding to each historical interaction result content based on the historical interaction result content. The context content vector can contain context information representing keyword frequency, topic distribution, and global semantics. The computer device can generate a content text prompt corresponding to each historical query statement through the context content vector and N historical query statements. The content text prompt can be a prompt. The update query statement is generated based on the N content text prompts. The N content text prompts can include content text prompt 1 corresponding to historical query statement 1, content text prompt 2 corresponding to historical query statement 2, ..., content text prompt N corresponding to historical query statement N. For example, the computer device can generate content text prompt 1 corresponding to historical query statement 1 through historical query statement 1 and the context content vector corresponding to historical query statement 1.

[0178] Take historical query statement C among N historical query statements i For example, the computer device can query the historical statement C i Perform feature extraction to obtain historical query statements C i The corresponding historical query feature vector D i , for the historical query feature vector D i Perform self-attention processing to obtain the historical query feature vector D i The corresponding word-unit score vector. The word-unit score vector contains the historical query statement C i The attention score corresponding to each word in .

[0179] The computer device can query the historical statement C i The corresponding global content vector and historical query statement C i The corresponding word-unit score vector is processed with cross attention to obtain the query hint vector, and the query hint vector is decoded to obtain the historical query statement C i Corresponding content text prompt, generate historical query statement C i The corresponding content text prompt i , the process can be shown as formula (4):

[0180]

[0181] in, Indicates that in the historical query statement C i The historical interaction results retrieved from the search are as follows: Q represents the initial query statement, and C represents the historical query statement. i The historical interaction result contents retrieved under the search include historical interaction result contents A1, historical interaction result contents A2, ..., historical interaction result contents A n. f represents the prompt constructor, and [;] represents the vector concatenation operation.

[0182] The computer device can query the historical query statement C i The historical interaction result content A1, historical interaction result content A2, ..., historical interaction result content A1 retrieved from the n Perform vector concatenation processing to obtain historical query statement C i The corresponding global content vector for the historical query statement C i Perform feature extraction to obtain historical query statements C i The corresponding historical query feature vector D i , for the historical query feature vector D i Perform self-attention processing to obtain the historical query feature vector D i The corresponding word-unit score vector. Perform cross-attention processing on the global content vector and the word-unit score vector to obtain the content of each historical interaction result and the historical query statement C i The attention score corresponding to each word in the , and the word with an attention score greater than or equal to a certain threshold is determined as the prompt keyword.

[0183] The prompt constructor can be a natural language template, such as "Please explain [Q] in detail, [Q]'s [x1], [Q]'s [x2], ... ", where Q is the initial query statement, and x1 and x2 are prompt keywords. For example, Q can be "What is ESG data?" The computer device can generate a query based on the historical query statement C. i The generated hint keyword x1 can be "E stands for Environmental, S stands for Social, and G stands for Governance", and the hint keyword x2 can be "ESG data is used to measure the impact of an organization on the environment and society". i The constructed content text prompt may be "Please explain in detail what ESG data is. The E in ESG data stands for Environmental, S stands for Social, and G stands for Governance. ESG data is used to measure the impact of an organization on the environment and society." The embodiment of the present application does not limit the specific content of the prompt constructor.

[0184] When the content text prompts corresponding to the N historical query statements are obtained, the computer device can generate an updated query statement Q based on the N content text prompts. ′ , multiple prompts can be input into the LM (Language Model) model to obtain the output result, and the output result is determined as the update query statement Q ′ , the process can be shown as formula (5):

[0185] Q ′ =LM(prompt1,prompt2,…,prompt n ) Formula (5)

[0186] Among them, LM represents the language generation function of the large language model.

[0187] The large language model can receive multiple prompts, which can be used to guide the large language model to generate text that rewrites the initial query statement Q. The large language model can parse the keywords and phrases in the prompt, understand its semantics and context, and convert the prompts into semantic representations, such as word embeddings. The large language model can perform cross-attention processing on the word embeddings of each prompt to obtain the fused features of multiple prompts, and decode the fused features to generate an updated query statement Q. ′ The updated query statement can be generated by semantic understanding and rewriting based on the context content vector and N historical query statements through a pre-trained large language model. The large language model can be a BERT model, a GPT model, etc., wherein the updated query statement is used to be displayed in the search prompt box on the search page, and the search prompt box is used to guide the object to re-search through the updated query statement.

[0188] The embodiment of the present application can rewrite and optimize the context content vector and N historical query statements through a pre-trained large language model. When the object clicks on certain search results (historical interaction result content), the system can organize the content of the search results into a context, such as a context content vector, and form a prompt based on the context content vector to input a large language model. The large language model can rewrite the initial query statement or N historical query statements based on the prompt to generate a more accurate query expression that meets the user's intention. The process of generating and updating the query local area can be a real-time feedback mechanism, which can enable the system to dynamically adjust the retrieval strategy and continuously optimize the query and results, thereby improving the relevance of the retrieval and user satisfaction.

[0189] Large language models can personalize query rewriting based on the user's historical behavior (interaction operation information) and contextual information (context content vector). Different users may have different intentions and preferences for the same query. Large language models can personalize and optimize queries by learning object features and interaction history, and perform more accurate understanding and rewriting. This personalized and context-aware retrieval method can provide users with a more intimate and intelligent search experience.

[0190] Optionally, the computer device may assign different text prompt weights to historical query statements at different query times, and generate an updated query statement Q based on the N content text prompts and their corresponding weights. ′ The process may be: based on the query time sequence between the N historical query statements, generating text prompt weights corresponding to the N content text prompts respectively; and generating an update query statement based on the N text prompt weights and the N content text prompts.

[0191] Specifically, the computer device may generate text prompt weights coefj, which correspond to N content text prompts, based on the query time sequence between the N historical query statements. i , through coef i Prompt and guide the LM model's attention to the N content text cues, thereby generating an update query sentence based on the N text cue weights and the N content text cues.

[0192] The computer device may display the updated query statement in the search prompt box on the search page, and guide the subject to re-search through the updated query statement through the search prompt box. For example, the search prompt box may display "Do you want to search for the updated query statement: Confirm / Cancel", and the subject may click the "Confirm" control in the search prompt box, and the computer device may re-search based on the updated query statement and display the candidate result content retrieved by the updated query statement.

[0193] Optionally, the search prompt box is also used to display the search explanation text, and the process may be: based on the historical query statement C i The corresponding Q context content vectors and historical query statements C i Corresponding content text prompt, generate historical query statement C i Corresponding unit explanation texts; when the unit explanation texts corresponding to N historical query statements are obtained, a retrieval explanation text for the update query statement is generated based on the N unit explanation texts.

[0194] Specifically, the computer device can be based on the historical query statement C i The corresponding Q context content vectors and historical query statements C i Corresponding content text prompt, generate historical query statement C i The corresponding unit explanation text, which can explain the historical query statement C i The corresponding content text prompt is generated based on what information in the historical interaction result content.

[0195] When the unit explanation texts corresponding to N historical query statements are obtained, a retrieval explanation text for the update query statement is generated based on the N unit explanation texts. The retrieval explanation text can explain what information in the historical interaction result content is used to generate the update query statement. For example, the retrieval explanation text can be "You have just collected historical interaction result content A. The content related to {initial query statement} in historical interaction result content A is keyword B. Do you want to search for keyword B through {update query statement}?".

[0196] Optionally, if the computer device detects a frequently co-occurring query combination in N historical query statements, for example, N historical query statements all include keyword A, and other objects searching for keyword A generally also search for keyword B, the computer device can actively recommend searching for keyword B to the objects.

[0197] It can be understood that the embodiments of the present application can display the key factors and weights in the query rewriting, result sorting and other processes through the retrieval explanation text, thereby improving the transparency of the system. It can improve the interpretability of the retrieval results, show the subject why a certain result is considered relevant, and enhance the subject's trust and understanding of the system.

[0198] Step S205, obtain the object question text, perform feature extraction on the object question text to obtain the object question vector, obtain the global content vectors corresponding to the N historical query statements corresponding to the update query statement; the object question text is input by the object in the search prompt box in the search page for the update query statement, and the global content vector is generated based on the historical interaction result content corresponding to the historical query statement; cross-attention processing is performed on the object question vector and the global content vector to obtain a question result vector, and a question result text is generated based on the question result vector; the question result text is used to be displayed in the search prompt box in the search page.

[0199] Specifically, the search prompt box can also be used for question-answering dialogue, and the computer device can obtain the object question text, perform feature extraction on the object question text, and obtain the object question vector. The object question text is what the object enters in the search prompt box on the search page for the update query statement.

[0200] The computer device can obtain the global content vectors corresponding to the N historical query statements corresponding to the updated query statement. The global content vector is generated based on the historical interaction result content corresponding to the historical query statement. The computer device can perform cross-attention processing on the object question vector and the global content vector to obtain a question result vector, and generate a question result text based on the question result vector. The question result text is used to display in the search prompt box on the search page. For example, the object can enter a question through the search prompt box, such as "What are the keywords related to {initial query statement}?" The computer device can answer the question entered by the object through the question result text, such as "The keywords related to {initial query statement} include keyword A and keyword B. Keyword A refers to..., keyword B refers to...".

[0201] It can be understood that the embodiment of the present application provides an intuitive and user-friendly user interface through the search prompt box, and can provide a variety of interactive methods, such as natural language query, multi-round dialogue, visual feedback, etc., so that users can express their search needs more conveniently and naturally. By introducing intelligent question and answer and dialogue systems, users can interact with the system in the form of dialogue, obtain smooth and coherent replies, and provide a more humane and intelligent search experience. Users can not only search through simple keyword input, but also use natural language to express their needs, and have multiple rounds of dialogue and interaction with the system. Large language models can understand users' questions and feedback, provide intelligent query rewriting and result optimization suggestions, and guide users to find the required information. This intelligent and interactive user experience makes the retrieval process more natural and efficient. The embodiment of the present application can also realize retrieval control through object question text, allowing users to customize and adjust retrieval strategies, such as setting filtering conditions, adjusting relevance weights, etc., giving users more control and improving the feasibility of the embodiment of the present application.

[0202] In order to evaluate the effectiveness of the embodiments of the present application, we conducted experiments on data sets in multiple fields and compared them with traditional retrieval methods. The experimental results show that the method of the present invention has achieved significant improvements in indicators such as retrieval accuracy and user satisfaction. In terms of retrieval accuracy: use indicators such as MAP (Mean Average Precision) and NDCG (Normalized Discounted Cumulative Gain) to evaluate the relevance of retrieval results. The method of the present invention has improved the MAP and NDCG indicators by an average of 10%-20% on different data sets. In terms of user satisfaction: through user surveys and feedback collection, statistics are collected on user satisfaction with the retrieval results. The user satisfaction of the method of the present invention has increased by an average of 15%-25%, and user feedback has become more active and positive. In terms of query rewriting quality: manually evaluate the relevance and meaning of the rewritten query to the original query. Figure 1 The rewritten query generated by the method of the present invention is consistent in terms of relevance and meaning. Figure 1 In terms of consistency, the average level reached 80%-90%.

[0203] The embodiment of the present application performs content retrieval based on an initial query statement to obtain M candidate result contents for sorting and displaying in a search page. The initial query statement can be input into a large language model, such as a BERT model, a GPT model, etc., to improve the semantic understanding ability of the initial query, rather than directly using keyword matching to perform content retrieval. The large language model rewrites and expands the initial query statement to generate semantically related query expressions. This semantic-level matching makes the retrieval results more accurate and comprehensive, makes up for the shortcomings of keyword matching, and improves the accuracy of content retrieval. If the object's interactive operation on the search page meets the re-search condition, the historical interaction information is obtained, and the historical interaction information includes N historical query statements and a historical interaction result set corresponding to the N historical query statements. The N historical query statements include the initial query statement, and the historical interaction result content in the historical interaction result set includes the candidate result content interacted by the object in the M candidate result content. Generate a context content vector corresponding to the content of the historical interaction result in the historical interaction information, generate an update query statement through the context content vector and N historical query statements, generate a context content vector representing the context information in each historical interaction result content through semantic understanding, learn and understand the interaction history and query expression of the object through N historical query statements, rewrite and optimize the N historical query statements based on the context content vector, better understand the user's true intention, and obtain semantically related update query statements. Therefore, searching through the update query statement can improve the accuracy of content retrieval, and rewrite and optimize the context content vector and N historical query statements through the pre-trained large language model. When the object clicks on certain search results (historical interaction result content), the system can organize the content of the search result into a context, such as a context content vector, and form a prompt based on the context content vector to input the large language model. The large language model can rewrite the initial query statement or N historical query statements according to the prompt to generate a more accurate query expression that meets the user's intention. The process of generating an update query local area can be a real-time feedback mechanism, which can enable the system to dynamically adjust the retrieval strategy and continuously optimize the query and results, thereby improving the relevance of the retrieval and user satisfaction.

[0204] The embodiment of the present application provides an intuitive and user-friendly user interface through a search prompt box, and can provide a variety of interactive methods, such as natural language query, multi-round dialogue, visual feedback, etc., so that users can express their search needs more conveniently and naturally. When the interactive operation of the object on the search page meets the re-search condition, the system can actively display the updated query statement obtained by rewriting the above method in the search prompt box in the search page, and use the search prompt box to guide the object to re-search by updating the query statement, without the need for the object to understand all the result content and adjust the keywords for active re-search, which greatly improves the efficiency of content screening and content retrieval. By introducing an intelligent question-answering and dialogue system, users are allowed to interact with the system in the form of dialogue, obtain smooth and coherent replies, and provide a more humane and intelligent search experience. Users can not only search through simple keyword input, but also use natural language to express their needs, and have multiple rounds of dialogue and interaction with the system. Large language models can understand users' questions and feedback, provide intelligent query rewriting and result optimization suggestions, and guide users to find the required information. This intelligent and interactive user experience makes the search process more natural and efficient. The embodiment of the present application can also implement retrieval control through object question text, allowing users to customize and adjust retrieval strategies, such as setting filtering conditions, adjusting relevance weights, etc., giving users more control and improving the feasibility of the embodiment of the present application.

[0205] See also Figure 6 , Figure 6 This is a schematic diagram of the structure of a data processing device provided in an embodiment of the present application. Figure 1 .like Figure 6 As shown, the data processing device 1 includes an initial retrieval module 510 , an information acquisition module 520 and a statement updating module 530 .

[0206] The initial search module 510 is used to perform content search based on the initial query statement to obtain M candidate result contents for sorting and displaying in the search page; M is a positive integer;

[0207] The information acquisition module 520 is used to acquire historical interaction information if the interaction operation of the object on the search page meets the re-search condition; the historical interaction information includes N historical query statements and a historical interaction result set corresponding to the N historical query statements; the N historical query statements include the initial query statement, and the historical interaction result content in the historical interaction result set includes the candidate result content interacted by the object among the M candidate result contents; N is a positive integer;

[0208] The statement update module 530 is used to generate a context content vector corresponding to the content of the historical interaction results in the historical interaction information, generate a content text prompt corresponding to each historical query statement through the context content vector and N historical query statements, and generate an updated query statement based on the N content text prompts; the updated query statement is used to be displayed in the search prompt box in the search page; the search prompt box is used to guide the object to be re-searched through the updated query statement.

[0209] In a possible implementation, the initial query module is further configured to perform the following operations:

[0210] Generate a business time range based on the query time of the initial query statement, and obtain S business query statements within the business time range except the initial query statement; S is a positive integer;

[0211] Perform feature extraction on the initial query statement to obtain an initial query vector, perform feature extraction on S business query statements respectively to obtain S business query vectors, perform cross-attention processing on the initial query vector and the S business query vectors to obtain attention scores corresponding to the S business query vectors respectively, determine the business query vector whose attention score is greater than or equal to the score threshold as the target business query vector, and determine the business query statement corresponding to the target business query vector as the historical query statement.

[0212] In a possible implementation, the initial query module is used to perform cross-attention processing on the initial query vector and the S business query vectors, and when the attention scores corresponding to the S business query vectors are obtained, the following operations are specifically performed:

[0213] Perform a dot multiplication operation on the S business query vectors and the query parameter matrix to obtain a query vector, perform a dot multiplication operation on the S business query vectors and the key parameter matrix to obtain a key vector, and perform a dot multiplication operation on the initial query vector and the value parameter matrix to obtain a value vector;

[0214] An attention score vector is generated based on the query vector and the key vector. The attention score vector is reduced in dimension based on the dimension of the key vector. The reduced-dimensional attention score vector is normalized to obtain a normalized vector. The normalized vector is point-multiplied with the value vector to obtain the attention scores corresponding to the S business query vectors.

[0215] In a possible implementation, when the statement updating module 530 is used to generate a context content vector corresponding to the historical interaction result content in the historical interaction information, it is specifically used to perform the following operations:

[0216] Based on the interactive operation information corresponding to the historical interactive result content in the historical interactive information, the interactive weight corresponding to the historical interactive result content is generated, and the feature extraction of the historical interactive result content is performed to obtain a content feature vector; the content feature vector is used to represent the context information of the historical interactive result content;

[0217] A context content vector corresponding to the historical interaction result content is generated based on the interaction weight and the content feature vector.

[0218] In a possible implementation, the interactive operation information includes H types of unit interactive information with different interactive operation types, the number of historical interactive result contents is T, and the T historical interactive result contents include historical interactive result contents A i , the interactive operation information includes the historical interactive result content A i Corresponding interactive operation information B i , i is a positive integer less than or equal to T; the statement updating module 530 is used to generate the interaction weight corresponding to the historical interaction result content based on the interaction operation information corresponding to the historical interaction result content in the historical interaction information, and is specifically used to perform the following operations:

[0219] Based on interactive operation information B i The interaction operation type and duration of each unit interaction information in the interaction operation are used to determine the content of the historical interaction results A. i The corresponding interaction score;

[0220] When the interaction scores corresponding to the T historical interaction result contents are obtained, the sum of the interaction scores of the T historical interaction result contents is determined as the total interaction score. i The corresponding interaction score generates the historical interaction result content A i The corresponding interaction weights.

[0221] In a possible implementation, the statement updating module 530 is used to extract features from the content of the historical interaction results, and when obtaining the content feature vector, specifically performs the following operations:

[0222] Performing text segmentation on the content of the historical interaction results to obtain a word unit sequence corresponding to the content of the historical interaction results; a word unit in the word unit sequence refers to the smallest basic unit obtained after the text information is subjected to text segmentation processing;

[0223] The keyword frequency in the historical interaction result content is feature extracted to obtain the word frequency embedding vector, the semantic information of each word in the historical interaction result content is encoded to obtain the semantic embedding vector, the position information of each word in the historical interaction result content is encoded to obtain the position embedding vector, and the paragraph information of each word in the historical interaction result content is encoded to obtain the paragraph embedding vector; the keyword frequency is the word frequency of the query keyword in the historical interaction result content;

[0224] A content feature vector corresponding to the content of the historical interaction results is generated based on the word frequency embedding vector, semantic embedding vector, position embedding vector and paragraph embedding vector.

[0225] In a possible implementation, the number of context content vectors is T, where T is a positive integer; the N historical query statements include historical query statements C i , T context content vectors include historical query sentences C i The corresponding Q context content vectors, Q is a positive integer less than or equal to T; the statement updating module 530 is used to generate the content text prompt corresponding to each historical query statement through the context content vector and the N historical query statements, and when the update query statement is generated based on the N content text prompts, it is specifically used to perform the following operations:

[0226] For historical query statements C i The corresponding Q context content vectors are concatenated to obtain the historical query statement C i The corresponding global content vector is based on the historical query statement C i and historical query statement C i The corresponding global content vector generates the historical query statement C i Corresponding content text prompt;

[0227] When the content text prompts respectively corresponding to N historical query statements are obtained, an update query statement is generated based on the N content text prompts.

[0228] In a possible implementation, when the statement updating module 530 is used to generate an update query statement based on N content text prompts, it is specifically used to perform the following operations:

[0229] Based on the query time sequence between the N historical query statements, generate text prompt weights corresponding to the N content text prompts respectively;

[0230] An update query statement is generated based on the N text prompt weights and the N content text prompts.

[0231] In a possible implementation, the statement updating module 530 is used to update the historical query statement C i and historical query statement Ci The corresponding global content vector generates the historical query statement C i When the corresponding content text prompt is displayed, it is used to perform the following operations:

[0232] For historical query statements C i Perform feature extraction to obtain historical query statements C i The corresponding historical query feature vector D i , for the historical query feature vector D i Perform self-attention processing to obtain the historical query feature vector D i The corresponding word-unit score vector; the word-unit score vector contains the historical query statement C i The attention score corresponding to each word in ;

[0233] For historical query statements C i The corresponding global content vector and historical query statement C i The corresponding word-unit score vector is processed with cross attention to obtain the query hint vector, and the query hint vector is decoded to obtain the historical query statement C i The corresponding content text prompt.

[0234] In a possible implementation, the search prompt box is also used to display the search explanation text; the statement update module 530 is also used to perform the following operations:

[0235] Based on historical query statement C i The corresponding Q context content vectors and historical query statements C i Corresponding content text prompt, generate historical query statement C i The corresponding unit explanation text;

[0236] When the unit interpretation texts corresponding to the N historical query statements are obtained, a retrieval interpretation text for the update query statement is generated based on the N unit interpretation texts.

[0237] In a possible implementation, the content acquisition module is further configured to perform the following operations:

[0238] Obtaining interaction scores corresponding to the candidate result contents interacted with by the object in the M candidate result contents; the interaction scores are determined based on the interaction operation type and the interaction operation duration;

[0239] If the number of candidate result contents interacted by the object is greater than or equal to the content number threshold, and the interaction scores corresponding to the candidate result contents interacted by the object are all lower than the interaction score threshold, it is determined that the object's interaction operation on the search page meets the re-search condition.

[0240] In a possible implementation, the initial query module is further configured to perform the following operations:

[0241] Obtain the unit word frequency corresponding to each word in the M candidate result contents, and generate the document weights corresponding to the M candidate result contents based on the unit word frequency;

[0242] A search score corresponding to each candidate result content is generated based on the M document weights, and the M candidate result contents are sorted and displayed on the search page based on the search score.

[0243] In a possible implementation, the initial query module is further configured to perform the following operations:

[0244] Obtain the query keyword in the initial query statement, perform word frequency statistics on the M candidate result contents based on the query keyword, obtain the keyword frequencies corresponding to the M candidate result contents respectively, and generate the relevance scores corresponding to the M candidate result contents respectively based on the M keyword frequencies; the keyword frequency is the word frequency of the query keyword in the candidate result contents;

[0245] The inverse document frequency corresponding to the query keyword is obtained, and a search score corresponding to each candidate result content is generated based on the inverse document frequency and M relevance scores. The M candidate result contents are sorted and displayed based on the search scores on the search page.

[0246] In a possible implementation, the M candidate result contents include target candidate result contents; when the initial query module is used to generate the relevance scores corresponding to the M candidate result contents respectively based on the M keyword frequencies, it is specifically used to perform the following operations:

[0247] In a possible implementation, the statement updating module 530 is further configured to perform the following operations:

[0248] Obtain the object question text, perform feature extraction on the object question text, obtain the object question vector, and obtain the global content vectors corresponding to the N historical query statements corresponding to the update query statement; the object question text is input by the object in the search prompt box on the search page for the update query statement, and the global content vector is generated based on the historical interaction result content corresponding to the historical query statement;

[0249] The object question vector and the global content vector are cross-attention processed to obtain a question result vector, and a question result text is generated based on the question result vector; the question result text is used to be displayed in the search prompt box on the search page.

[0250] The embodiment of the present application performs content retrieval based on an initial query statement to obtain M candidate result contents for sorting and displaying in a search page. The initial query statement can be input into a large language model, such as a BERT model, a GPT model, etc., to improve the semantic understanding ability of the initial query, rather than directly using keyword matching to perform content retrieval. The large language model rewrites and expands the initial query statement to generate semantically related query expressions. This semantic-level matching makes the retrieval results more accurate and comprehensive, makes up for the shortcomings of keyword matching, and improves the accuracy of content retrieval. If the object's interactive operation on the search page meets the re-search condition, the historical interaction information is obtained, and the historical interaction information includes N historical query statements and a historical interaction result set corresponding to the N historical query statements. The N historical query statements include the initial query statement, and the historical interaction result content in the historical interaction result set includes the candidate result content interacted by the object in the M candidate result content. Generate a context content vector corresponding to the content of the historical interaction result in the historical interaction information, generate an update query statement through the context content vector and N historical query statements, generate a context content vector representing the context information in each historical interaction result content through semantic understanding, learn the interaction history and query expression of the object through N historical query statements, rewrite and optimize the N historical query statements based on the context content vector, better understand the user's true intention, and obtain semantically related update query statements. Therefore, searching through the update query statement can improve the accuracy of content retrieval, and rewrite and optimize the context content vector and N historical query statements through the pre-trained large language model. When the object clicks on certain search results (historical interaction result content), the system can organize the content of the search result into a context, such as a context content vector, and form a prompt based on the context content vector to input the large language model. The large language model can rewrite the initial query statement or N historical query statements according to the prompt to generate a more accurate query expression that meets the user's intention. The process of generating an update query local area can be a real-time feedback mechanism, which can enable the system to dynamically adjust the retrieval strategy and continuously optimize the query and results, thereby improving the relevance of the retrieval and user satisfaction.

[0251] The embodiment of the present application provides an intuitive and user-friendly user interface through a search prompt box, and can provide a variety of interactive methods, such as natural language query, multi-round dialogue, visual feedback, etc., so that users can express their search needs more conveniently and naturally. When the interactive operation of the object on the search page meets the re-search condition, the system can actively display the updated query statement obtained by rewriting the above method in the search prompt box in the search page, and use the search prompt box to guide the object to re-search by updating the query statement, without the need for the object to understand all the result content and adjust the keywords for active re-search, which greatly improves the efficiency of content screening and content retrieval. By introducing an intelligent question-answering and dialogue system, users are allowed to interact with the system in the form of dialogue, obtain smooth and coherent replies, and provide a more humane and intelligent search experience. Users can not only search through simple keyword input, but also use natural language to express their needs, and have multiple rounds of dialogue and interaction with the system. Large language models can understand users' questions and feedback, provide intelligent query rewriting and result optimization suggestions, and guide users to find the required information. This intelligent and interactive user experience makes the search process more natural and efficient. The embodiment of the present application can also implement retrieval control through object question text, allowing users to customize and adjust retrieval strategies, such as setting filtering conditions, adjusting relevance weights, etc., giving users more control and improving the feasibility of the embodiment of the present application.

[0252] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program with a predetermined function, and works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories) or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0253] See also Figure 7 , Figure 7 Schematic diagram of the structure of a computer device provided in an embodiment of the present application. Figure 7As shown, the computer device 1000 may include: a processor 1001, a network interface 1004 and a memory 1005. In addition, the above-mentioned computer device 1000 may also include: a user interface 1003, and at least one communication bus 1002. Among them, the communication bus 1002 is used to realize the connection and communication between these components. Among them, the user interface 1003 may include a display screen (Display), a keyboard (Keyboard), and the user interface 1003 may optionally include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or it may be a non-volatile memory (non-volatile memory), such as at least one disk storage. The memory 1005 may optionally also be at least one storage device located away from the aforementioned processor 1001. As Figure 7 As shown, the memory 1005 as a computer-readable storage medium may include an operating system, a network communication module, a user interface module, and a device control application program.

[0254] In such Figure 7 In the computer device 1000 shown, the network interface 1004 can provide a network communication network element; the user interface 1003 is mainly used to provide an input interface for the user; and the processor 1001 can be used to call the device control application stored in the memory 1005 to achieve:

[0255] Perform content retrieval based on the initial query statement to obtain M candidate result contents for sorting and displaying in the search page; M is a positive integer;

[0256] If the object's interaction operation on the search page meets the re-search condition, historical interaction information is obtained; the historical interaction information includes N historical query statements and a historical interaction result set corresponding to the N historical query statements; the N historical query statements include the initial query statement, and the historical interaction result content in the historical interaction result set includes the candidate result content that the object has interacted with among the M candidate result contents; N is a positive integer;

[0257] Generate a context content vector corresponding to the content of the historical interaction results in the historical interaction information, generate a content text prompt corresponding to each historical query statement through the context content vector and N historical query statements, and generate an update query statement based on the N content text prompts; the update query statement is used to be displayed in a search prompt box on the search page; the search prompt box is used to guide the object to be re-searched through the update query statement.

[0258] It should be understood that the computer device 1000 described in the embodiment of the present application can execute the above Figure 3 and Figure 5 The description of the data processing method in any corresponding embodiment will not be repeated here. In addition, the description of the beneficial effects of adopting the same method will not be repeated here.

[0259] In addition, it should be pointed out here that: the embodiment of the present application also provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program, and when the processor executes the computer program, it can execute the above Figure 3 and Figure 5 The description of the above data processing method in any corresponding embodiment will not be repeated here. In addition, the description of the beneficial effects of the same method will not be repeated. For technical details not disclosed in the computer-readable storage medium embodiment involved in this application, please refer to the description of the method embodiment of this application.

[0260] The computer-readable storage medium may be the data processing device provided in any of the aforementioned embodiments or the internal storage unit of the computer device, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (smart media card, SMC), a secure digital (secure digital, SD) card, a flash card (flash card), etc. equipped on the computer device. Further, the computer-readable storage medium may also include both the internal storage unit of the computer device and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium may also be used to temporarily store data that has been displayed or is to be displayed.

[0261] In addition, it should be noted that: the embodiment of the present application also provides a computer program product, the computer program product includes a computer program, the computer program is stored in a computer-readable storage medium. The processor of the computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the above Figure 3 and Figure 5 Any method provided by the corresponding embodiment.

[0262] The terms "first", "second", etc. in the description, claims, and drawings of the embodiments of the present application are used to distinguish different objects, rather than to describe a specific order. In addition, the term "comprising" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, device, product, or equipment that includes a series of steps or units is not limited to the listed steps or modules, but optionally includes steps or modules that are not listed, or optionally includes other step units inherent to these processes, methods, devices, products, or equipment.

[0263] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of network elements in the above description. Whether these network elements are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described network elements for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0264] The method and related apparatus provided by the embodiment of the present application are described with reference to the method flow chart and / or structural diagram provided by the embodiment of the present application. Specifically, each process and / or box in the method flow chart and / or structural diagram, as well as the combination of the processes and / or boxes in the flow chart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable device to generate a machine, so that the instructions executed by the processor of the computer or other programmable device generate instructions for implementing the process in the process. Figure 1 A process or multiple processes and / or structures Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or structures Figure 1 These computer program instructions can also be loaded onto a computer or other programmable device so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide the functions for implementing the process. Figure 1 A flow or multiple flows and / or structures illustrate the steps of the functions specified in one block or multiple blocks.

[0265] The steps in the method of the embodiment of the present application can be adjusted in order, combined and deleted according to actual needs.

[0266] The modules in the device of the embodiment of the present application can be merged, divided and deleted according to actual needs.

[0267] The above disclosure is only the preferred embodiment of the present application, which certainly cannot be used to limit the scope of rights of the present application. Therefore, equivalent changes made according to the claims of the present application are still within the scope covered by the present application.

Claims

1. A data processing method, It is characterized in that include: Performing content retrieval based on the initial query statement to obtain M candidate result contents for sorting and displaying in the search page; M is a positive integer; If the object's interaction operation on the search page meets the re-search condition, historical interaction information is obtained; the historical interaction information includes N historical query statements and a historical interaction result set corresponding to the N historical query statements; the N historical query statements include the initial query statement, and the historical interaction result content in the historical interaction result set includes the candidate result content interacted by the object among the M candidate result contents; N is a positive integer; Generate a context content vector corresponding to the content of the historical interaction result in the historical interaction information, generate a content text prompt corresponding to each historical query statement through the context content vector and the N historical query statements, and generate an update query statement based on the N content text prompts; the update query statement is used to be displayed in a search prompt box in the search page; The search prompt box is used to guide the object to be re-searched through the update query statement.

2. The method according to claim 1, It is characterized in that Also includes: Generate a business time range based on the query time of the initial query statement, and obtain S business query statements within the business time range except the initial query statement; S is a positive integer; Perform feature extraction on the initial query statement to obtain an initial query vector, perform feature extraction on the S business query statements respectively to obtain S business query vectors, perform cross-attention processing on the initial query vector and the S business query vectors to obtain attention scores corresponding to the S business query vectors respectively, determine the business query vector whose attention score is greater than or equal to the score threshold as the target business query vector, and determine the business query statement corresponding to the target business query vector as the historical query statement.

3. The method according to claim 2, It is characterized in that The performing cross-attention processing on the initial query vector and the S business query vectors to obtain attention scores corresponding to the S business query vectors respectively includes: Performing a dot multiplication operation on the S business query vectors and the query parameter matrix to obtain a query vector, performing a dot multiplication operation on the S business query vectors and the key parameter matrix to obtain a key vector, and performing a dot multiplication operation on the initial query vector and the value parameter matrix to obtain a value vector; An attention score vector is generated based on the query vector and the key vector, the attention score vector is dimensionality reduced based on the dimension of the key vector, the reduced attention score vector is normalized to obtain a normalized vector, and a dot multiplication operation is performed on the normalized vector and the value vector to obtain the attention scores corresponding to the S business query vectors respectively.

4. The method according to claim 1, It is characterized in that The generating a context content vector corresponding to the historical interaction result content in the historical interaction information includes: Based on the interactive operation information corresponding to the historical interaction result content in the historical interaction information, an interactive weight corresponding to the historical interaction result content is generated, and features of the historical interaction result content are extracted to obtain a content feature vector; the content feature vector is used to characterize context information of the historical interaction result content; A context content vector corresponding to the historical interaction result content is generated based on the interaction weight and the content feature vector.

5. The method according to claim 4, It is characterized in that The interactive operation information includes H types of unit interactive information with different interactive operation types. The number of the historical interactive result contents is T, and the T historical interactive result contents include historical interactive result contents A i The interactive operation information includes the historical interactive result content A i Corresponding interactive operation information B i , i is a positive integer less than or equal to T; The generating, based on the interaction operation information corresponding to the historical interaction result content in the historical interaction information, the interaction weight corresponding to the historical interaction result content includes: Based on interactive operation information B i The interaction operation type and interaction operation duration of each unit interaction information in the historical interaction result content A is determined i The corresponding interaction score; When the interaction scores corresponding to the T historical interaction result contents are obtained, the sum of the interaction scores of the T historical interaction result contents is determined as the total interaction score. i The corresponding interaction score generates the historical interaction result content A i The corresponding interaction weights.

6. The method according to claim 4, It is characterized in that The feature extraction of the historical interaction result content to obtain a content feature vector includes: Performing text segmentation on the historical interaction result content to obtain a word unit sequence corresponding to the historical interaction result content; a word unit in the word unit sequence refers to the smallest basic unit obtained after the text information is subjected to text segmentation processing; Perform feature extraction on the keyword frequency in the historical interaction result content to obtain a word frequency embedding vector, encode the semantic information of each word in the historical interaction result content to obtain a semantic embedding vector, encode the position information of each word in the historical interaction result content to obtain a position embedding vector, and encode the paragraph information of each word in the historical interaction result content to obtain a paragraph embedding vector; the keyword frequency is the word frequency of the query keyword in the historical interaction result content; A content feature vector corresponding to the historical interaction result content is generated based on the word frequency embedding vector, the semantic embedding vector, the position embedding vector and the paragraph embedding vector.

7. The method according to claim 1, It is characterized in that The number of context content vectors is T, where T is a positive integer; the N historical query statements include the historical query statements C i , T context content vectors include historical query sentences C i corresponding Q context content vectors, where Q is a positive integer less than or equal to T; generating content text prompts corresponding to each historical query statement respectively through the context content vector and the N historical query statements, and generating an update query statement based on the N content text prompts, including: For the historical query statement C i The corresponding Q context content vectors are concatenated to obtain the historical query statement C i The corresponding global content vector is based on the historical query statement C i and the historical query statement C i The corresponding global content vector generates the historical query statement C i Corresponding content text prompt; When the content text prompts respectively corresponding to the N historical query statements are obtained, an update query statement is generated based on the N content text prompts.

8. The method according to claim 7, It is characterized in that The step of generating an update query statement based on N content text prompts includes: Generating text prompt weights corresponding to N content text prompts respectively based on the query time sequence between the N historical query statements; An update query statement is generated based on the N text prompt weights and the N content text prompts.

9. The method according to claim 7, It is characterized in that Based on the historical query statement C i and the historical query statement C i The corresponding global content vector generates the historical query statement C i The corresponding content text prompts include: For the historical query statement C i Perform feature extraction to obtain the historical query statement C i The corresponding historical query feature vector D i , for the historical query feature vector D i Perform self-attention processing to obtain the historical query feature vector D i The corresponding word-unit score vector; the word-unit score vector includes the historical query statement C i The attention score corresponding to each word in ; For the historical query statement C i The corresponding global content vector and the historical query statement C i The corresponding word-unit score vector is cross-attention processed to obtain a query hint vector, and the query hint vector is text-decoded to obtain the historical query statement C i The corresponding content text prompt.

10. The method according to claim 9, It is characterized in that The search prompt box is also used to display the search explanation text; the method also includes: Based on the historical query statement C i The corresponding Q context content vectors and the historical query statement C i The corresponding content text prompt generates the historical query statement C i The corresponding unit explanation text; When the unit interpretation texts respectively corresponding to the N historical query statements are obtained, a retrieval interpretation text for the update query statement is generated based on the N unit interpretation texts.

11. The method according to claim 1, It is characterized in that Also includes: Obtaining interaction scores corresponding to the candidate result contents interacted with by the object in the M candidate result contents; The interaction score is determined based on the interaction operation type and the interaction operation duration; If the number of candidate result contents interacted by the object is greater than or equal to the content number threshold, and the interaction scores corresponding to the candidate result contents interacted by the object are all lower than the interaction score threshold, it is determined that the object's interaction operation on the search page meets the re-search condition.

12. The method according to claim 1, It is characterized in that Also includes: Obtaining the unit word frequency corresponding to each word in the M candidate result contents, and generating the document weights corresponding to the M candidate result contents based on the unit word frequency; A search score corresponding to each candidate result content is generated based on the M document weights, and the M candidate result contents are sorted and displayed on the search page based on the search score.

13. The method according to claim 1, It is characterized in that Also includes: Obtaining a query keyword in the initial query statement, performing word frequency statistics on the M candidate result contents based on the query keyword, obtaining keyword frequencies corresponding to the M candidate result contents respectively, and generating relevance scores corresponding to the M candidate result contents respectively based on the M keyword frequencies; The keyword frequency is the word frequency of the query keyword in the candidate result content; The inverse document frequency corresponding to the query keyword is obtained, and a search score corresponding to each candidate result content is generated based on the inverse document frequency and M relevance scores, and the M candidate result contents are sorted and displayed on the search page based on the search scores.

14. The method according to claim 13, It is characterized in that The M candidate result contents include target candidate result content; The generating of the relevance scores corresponding to the M candidate result contents respectively based on the M keyword frequencies includes: An adjustment factor is generated based on the target document length of the target candidate result content and the average document length of the M candidate result contents, and a relevance score corresponding to the target candidate result content is generated based on the adjustment factor and the keyword frequency corresponding to the target candidate result content.

15. The method according to claim 1, It is characterized in that Also includes: Obtaining an object question text, performing feature extraction on the object question text to obtain an object question vector, and obtaining global content vectors corresponding to N historical query statements corresponding to the update query statement; the object question text is input by the object in a search prompt box on the search page for the update query statement, and the global content vector is generated based on the historical interaction result content corresponding to the historical query statement; Performing cross-attention processing on the object question vector and the global content vector to obtain a question result vector, and generating a question result text based on the question result vector; The question result text is used to be displayed in the search prompt box in the search page.

16. A data processing device, It is characterized in that include: An initial search module, used to search for content based on an initial query statement, and obtain M candidate result contents for sorting and displaying in a search page; M is a positive integer; An information acquisition module, configured to acquire historical interaction information if the object's interaction operation on the search page satisfies a re-search condition; the historical interaction information includes N historical query statements and a historical interaction result set corresponding to the N historical query statements; the N historical query statements include the initial query statement, and the historical interaction result content in the historical interaction result set includes the candidate result content interacted by the object among the M candidate result contents; N is a positive integer; A statement updating module, used to generate a context content vector corresponding to the content of the historical interaction result in the historical interaction information, generate a content text prompt corresponding to each historical query statement through the context content vector and the N historical query statements, and generate an updated query statement based on the N content text prompts; the updated query statement is used to be displayed in the search prompt box in the search page; The search prompt box is used to guide the object to be re-searched through the update query statement.

17. A computer device, It is characterized in that include: Processor, memory, and network interface; The processor is connected to the memory and the network interface, wherein the network interface is used to provide a data communication function, the memory is used to store a computer program, and the processor is used to call the computer program so that the computer device executes the method described in any one of claims 1-15.

18. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded and executed by a processor, so that a computer device having the processor executes the method according to any one of claims 1 to 15.

19. A computer program product, It is characterized in that The computer program product comprises a computer program, which is stored in a computer-readable storage medium and is suitable for being read and executed by a processor, so that a computer device having the processor executes the method according to any one of claims 1 to 15.

Citation Information

Cited By

  • Statement modification method and device and computer equipment

    CN121705303A

  • Systems and methods for improved data curation and integration

    US12481686B1

  • Method and system for evaluating effectiveness of prompt

    US12681969B2

  • Method and system for evaluating effectiveness of prompt

    US20260170029A1

  • Data processing method and apparatus, and device and readable storage medium

    WO2026031951A1