Information retrieval method, device, prediction model training method and equipment
Patent Information
- Application Number
- CN202311247101.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-25
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2043-09-25
AI Technical Summary
[0003]本申请提供一种信息检索方法、装置、信息检索设备及存储介质,旨在解决基于搜索词对数据进行数据检导致搜索推荐信息准确性低的问题,以提升信息检索的准确性
[0054]This application provides information retrieval methods, devices, information retrieval equipment and storage media, by obtaining search words and text titles associated with the search words; and obtaining reference search word clusters associated with each of the text titles, the reference search words The cluster is constructed based on the historical browsing information and the historical search words corresponding to each of the historical browsing information; and then based on the search words, each of the text titles, and the reference search word clusters corresponding to each of the text titles, Determine the target text title corresponding to the search term, and output the target recommendation information corresponding to the target text title. In this solution, based on the user's historical browsing information and the historical search words corresponding to each historical browsing information, a reference search word cluster corresponding to the text title is constructed, and the text title associated with the search word is initially obtained, and the text title is used to replace the information content. Simplify the amount of data analysis. Further, analyze the reference search word clusters corresponding to the text title and the search words, that is, use the reference search word clusters to replace the text corresponding to the text title to perform search information analysis and reduce search words. There is a difference in the information distribution space between the recommended information and the reference search word cluster to replace the text corresponding to the text title, ensuring the accuracy and comprehensiveness of the reference search word cluster, and then determining the target text title corresponding to the search word in the text title. , and further determine the corresponding target recommendation information based on the target text title to ensure the accuracy of the target recommendation information.
Smart Images

Figure CN117725304B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of computer technology, and in particular to an information retrieval method, device, prediction model training method and equipment. Background Art
[0002] With the development of the text era, information retrieval has been widely used. For example, text retrieval is used in question and answer retrieval recommendation, product retrieval promotion, text retrieval recommendation, etc. In related technologies, text retrieval generally searches the database for the search terms entered by the user in the search system, and outputs the retrieval recommendation information related to the search terms in the database. However, due to the large amount of information in the database and the complexity of the information space, directly performing data analysis and retrieval on the database based on the search terms is prone to low accuracy of search recommendation information due to information distribution bias. Summary of the invention
[0003] The present application provides an information retrieval method, apparatus, information retrieval device and storage medium, aiming to solve the problem of low accuracy of search recommendation information caused by data retrieval based on search terms, so as to improve the accuracy of information retrieval.
[0004] In a first aspect, the present application provides an information retrieval method, comprising:
[0005] Obtain a search term and a text title associated with the search term;
[0006] Acquire a reference search word cluster associated with each of the text titles, wherein the reference search word cluster is constructed based on historical browsing information and a historical search word corresponding to each of the historical browsing information;
[0007] According to the search word, each of the text titles, and the reference search word cluster corresponding to each of the text titles, a target text title corresponding to the search word is determined, and target recommendation information corresponding to the target text title is output.
[0008] In a possible implementation of the present application, the step of constructing the reference search term cluster includes:
[0009] According to the user's historical behavior data, obtain the user's browsing history information;
[0010] For each piece of historical browsing information, obtain the historical search words corresponding to the historical browsing information, count the repetition frequency of each historical search word, determine the reference historical search words corresponding to the historical browsing information, and construct a reference search word cluster corresponding to the historical browsing information based on the reference historical search words.
[0011] In a possible implementation of the present application, determining the target text title corresponding to the search term according to the search term, each of the text titles, and the reference search term cluster corresponding to each of the text titles includes:
[0012] Input the search term, each of the text titles, and the reference search term cluster associated with each of the text titles into a prediction model, wherein the prediction model is a BERT model;
[0013] Extract features from the search term and each reference historical search term in the reference search term cluster respectively by using a feature extraction module in the prediction model to obtain search term features and reference historical search term features;
[0014] Processing the search term features and the reference historical search term features through a calculation module in the prediction model to obtain a relevance between the search term and each of the reference search term clusters;
[0015] According to the relevance corresponding to the reference search word cluster, a target reference search word cluster is extracted from the reference search word cluster, and a text title corresponding to the target reference search word cluster is set as a target text title corresponding to the search word.
[0016] In a possible implementation of the present application, the prediction model is trained by the following method:
[0017] Constructing a training sample set based on the user's historical search terms and the historical recommendation information corresponding to the historical search terms;
[0018] According to each training sample in the training sample set, the preset BERT model is trained to obtain a prediction model;
[0019] The training sample construction steps are as follows:
[0020] For each historical search term, execute:
[0021] Counting the browsing frequency of each of the historical recommendation information corresponding to the historical search terms to obtain an average browsing frequency;
[0022] Determining the relevance between each of the historical recommendation information and the historical search term according to the magnitude relationship between the browsing frequency corresponding to each of the historical recommendation information and the average browsing frequency;
[0023] The historical search word and each of the historical browsing information with set relevance associated with the historical search word are set as a training sample.
[0024] In a possible implementation of the present application, before inputting the search term, each of the text titles, and the reference search term cluster associated with each of the text titles into the prediction model, the method further includes:
[0025] For each text title, each of the reference historical search terms in the reference search term cluster associated with the text title is input into a feature extraction module in the prediction model, and a reference historical search term feature corresponding to each of the reference historical search terms is output;
[0026] The feature mean of the reference historical search term features corresponding to the reference search term cluster is calculated, and the parameters of the feature extraction module are adjusted according to the feature mean.
[0027] In a possible implementation of the present application, the outputting the target recommendation information corresponding to the target text title includes:
[0028] Extracting key information from the target recommendation information corresponding to each of the target text titles, wherein the key information includes at least one of a key text segment, a key image, and a key audio segment;
[0029] Adding each of the target text titles and the corresponding key information to each text column in a preset recommendation page template;
[0030] A hyperlink is created between each of the text columns and the target recommendation information corresponding to the target text title in the text column, and a recommendation page is obtained and displayed.
[0031] In a possible implementation of the present application, after creating a hyperlink between each of the text columns and the target recommendation information corresponding to the target text title in the text column, obtaining a recommendation page and displaying it, the method further includes:
[0032] Receiving a user operation on the recommendation page, the operation including clicking or sliding;
[0033] If the user operation triggers a hyperlink corresponding to any text field in the recommendation page, the target recommendation information corresponding to the hyperlink is displayed;
[0034] When the recommendation page closing instruction triggered by the user operation is received, the recommendation page is controlled to be closed, and the displayed target recommendation information corresponding to the search term is obtained, the search term is set as a historical search term, and the displayed target recommendation information is set as historical browsing information.
[0035] In a second aspect, the present application further provides an information retrieval device, the device comprising:
[0036] A first acquisition module: used to acquire a search term and a text title associated with the search term;
[0037] A second acquisition module: used for acquiring reference search word clusters associated with each of the text titles, wherein the reference search word clusters are constructed based on historical browsing information and historical search words corresponding to each of the historical browsing information;
[0038] Determination module: used to determine the target text title corresponding to the search term based on the search term, each text title, and the reference search term cluster corresponding to each text title, and output target recommendation information corresponding to the target text title.
[0039] In a third aspect, the present application further provides an information retrieval device, the information retrieval device comprising:
[0040] one or more processors;
[0041] Memory; and
[0042] One or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the processor to implement the steps in any one of the information retrieval methods.
[0043] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is loaded by a processor to execute the steps in any one of the information retrieval methods described.
[0044] In a fifth aspect, the present application also provides a prediction model training method, comprising:
[0045] Get historical search terms and historical browsing information that the user has browsed;
[0046] Setting each of the historical search terms and each of the historical recommendation information with set relevance associated with each of the historical search terms as a training sample to obtain a training sample set;
[0047] For each piece of historical browsing information, obtaining a historical search term corresponding to the historical browsing information, counting the repetition frequency of each of the historical search terms, determining a reference historical search term corresponding to the historical browsing information, and constructing a search term cluster corresponding to the historical browsing information according to the reference historical search term;
[0048] According to each of the training samples in the training sample set and the search word family, a preset training model is trained to obtain a prediction model.
[0049] In one possible implementation of the present application, the historical recommendation information with set relevance is obtained by the following steps:
[0050] For each of the historical search terms, execute:
[0051] Counting the browsing frequency of each of the historical recommendation information corresponding to the historical search terms to obtain an average browsing frequency;
[0052] Determining the relevance between each of the historical recommendation information and the historical search term according to the magnitude relationship between the browsing frequency corresponding to each of the historical recommendation information and the average browsing frequency;
[0053] The relevance of the historical recommendation information to the historical text title is set.
[0054] The present application provides an information retrieval method, an apparatus, an information retrieval device and a storage medium, which obtain a search term and a text title associated with the search term; and obtain a reference search term cluster associated with each of the text titles, wherein the reference search term cluster is constructed based on historical browsing information and a historical search term corresponding to each of the historical browsing information; then, based on the search term, each of the text titles, and the reference search term cluster corresponding to each of the text titles, determine a target text title corresponding to the search term, and output target recommendation information corresponding to the target text title. In this scheme, based on the user's historical browsing information and each historical search term corresponding to the historical browsing information, a reference search term cluster corresponding to the text title is constructed, and the text title associated with the search term is preliminarily obtained, and the text title is used to replace the information content to simplify the amount of data analysis. Furthermore, the reference search term cluster corresponding to the text title and the search term are analyzed, that is, the reference search term cluster is used to replace the text corresponding to the text title for search information analysis, thereby reducing the information distribution space difference between the search term and the recommended information, and constructing a reference search term cluster to replace the text corresponding to the text title to ensure the accuracy and comprehensiveness of the reference search term cluster, and then determine the target text title corresponding to the search term in the text title, and further determine the corresponding target recommendation information based on the target text title to ensure the accuracy of the target recommendation information. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.
[0056] Figure 1 is a schematic diagram of a scenario of an information retrieval method provided in an embodiment of the present application;
[0057] Figure 2 It is a schematic diagram of an example flow chart of the information retrieval method provided in the embodiments of the present application;
[0058] Figure 3 It is a schematic diagram of an embodiment of the process of constructing a reference search word cluster in the information retrieval method provided in the embodiments of the present application;
[0059] Figure 4 It is a schematic diagram of an embodiment of the process of determining the target text title in the information retrieval method provided in the embodiment of the present application;
[0060] Figure 5 It is a schematic diagram of an embodiment of the process of outputting target recommendation information in the information retrieval method provided in the embodiment of the present application;
[0061] Figure 6 is a flowchart of another embodiment of the information retrieval method provided in the embodiments of the present application;
[0062] Figure 7 This is a flowchart of an implementation plan of the prediction model training method provided in the examples of the present application;
[0063] Figure 8 This is a schematic diagram of an implementation flow of a prediction model training method provided in an embodiment of the present application;
[0064] Fig. 9 is a schematic diagram of the structure of an embodiment of the information retrieval device provided in the embodiments of the present application;
[0065] Fig.10 It is a schematic diagram of the structure of an embodiment of the information retrieval device provided in the embodiments of the present application. DETAILED DESCRIPTION
[0066] 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 invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.
[0067] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, which are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.
[0068] In the embodiment of the present application, "and / or" describes the association relationship of the associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / ", unless otherwise specified, generally indicates that the associated objects before and after are in an "or" relationship.
[0069] In this application, the word "exemplary" is used to mean "serving as an example, illustration, or illustration." Any embodiment described in this application as "exemplary" is not necessarily to be construed as being preferred or advantageous over other embodiments. The following description is given to enable any person skilled in the art to implement and use the invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the invention can be implemented without using these specific details. In other instances, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in this application.
[0070] In the prior art, information retrieval is generally performed through correlation analysis between search terms and the information to be recommended, that is, correlation is the core task in the search scenario. Specifically, the correlation task is to judge the correlation between the current note and the user's search query based on the user's search term query (referred to as the search term in the full text), that is, the degree of correlation between the current note and the user's search query. For example, when the user's search query is: How about Dior's lipstick, if the note introduces Dior lipstick, it is considered relevant, if it is a MAC lipstick, it is considered irrelevant. For the user's search query, the notes are returned to the user in descending order of relevance to the query.
[0071] However, in the search model of the related art, by simultaneously inputting the search term query and the information to be recommended, exemplarily, the information to be recommended is the body of the note, and after the search model, the output of the CLS position is taken as the similarity score of the current query-note. This method needs to process a long text, including the search term query, the note title and the note body. The note body may have thousands of characters, so the amount of calculation required is very large, resulting in poor performance of the model, occupying a lot of resources, and failing to meet the online latency requirements; and the same model calculates the search term query and the semantic feature vector of the information to be recommended, and the spatial distribution of the two is different, resulting in inaccurate similarity scores output by the model, and low accuracy of information recommendation based on the similarity.
[0072] Therefore, the embodiments of the present application provide an information retrieval method, apparatus, prediction model training method, information retrieval device and storage medium, which construct a reference search term cluster based on historical browsing information and historical search terms corresponding to each of the historical browsing information, replace the text corresponding to the text title based on the reference search term cluster, ensure the accuracy and comprehensiveness of the reference search term cluster, and perform analysis and processing based on the reference search term cluster and the search terms to determine the target recommended information, reduce the information distribution space difference between the search terms and the recommended information, and ensure the accuracy of the target recommended information.
[0073] The information retrieval method in the embodiment of the present invention is applied to an information retrieval device, which is arranged in an information retrieval device. The information retrieval device is provided with one or more processors, a memory, and one or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the information retrieval method; the information retrieval device can be a terminal, such as a mobile phone or a tablet computer, and the information retrieval device can also be a server, or a service cluster composed of multiple servers.
[0074] like Figure 1 As shown, Figure 1This is a scenario diagram of the information retrieval method in an embodiment of the present application. The information retrieval scenario in the embodiment of the present invention includes an information retrieval device 100 (the information retrieval device 100 is integrated with an information retrieval apparatus), and a computer-readable storage medium corresponding to the information retrieval is run in the information retrieval device 100 to execute the information retrieval steps.
[0075] Understandably, Figure 1 The information retrieval device in the scenario of the information retrieval method shown, or the devices included in the information retrieval device, does not constitute a limitation on the embodiments of the present invention. That is, the number of devices and types of devices included in the scenario of the information retrieval method, or the number of devices and types of devices included in each device, do not affect the overall implementation of the technical solution in the embodiments of the present invention, and can all be regarded as equivalent replacements or derivatives of the technical solution claimed to be protected in the embodiments of the present invention.
[0076] The information retrieval device 100 in the embodiment of the present invention is mainly used for: obtaining search terms and text titles associated with the search terms; obtaining reference search term clusters associated with each text title, the reference search term clusters being constructed based on historical browsing information and historical search terms corresponding to each piece of historical browsing information; determining a target text title corresponding to the search term based on the search terms, each text title, and the reference search term clusters corresponding to each text title, and outputting target recommendation information corresponding to the target text title.
[0077] In the embodiment of the present invention, the information retrieval device 100 may be an independent information retrieval device, or an information retrieval device network or information retrieval device cluster composed of information retrieval devices. For example, the information retrieval device 100 described in the embodiment of the present invention includes but is not limited to a computer, a network host, a single network information retrieval device, a set of multiple network information retrieval devices, or a cloud information retrieval device composed of multiple information retrieval devices. The cloud information retrieval device is composed of a large number of computers or network information retrieval devices based on cloud computing.
[0078] Those skilled in the art will understand that Figure 1 The application environment shown in the figure is only one application scenario of the present application solution and does not constitute a limitation on the application scenario of the present application solution. Other application environments may also include Figure 1 More or fewer information retrieval devices, or information retrieval device network connection relationships, such as Figure 1 Only one information retrieval device is shown. It can be understood that the scenario of the information retrieval method can also include one or more other information retrieval devices, which are not limited here. The information retrieval device 100 can also include a memory for storing data, such as search term clusters corresponding to each text title.
[0079] In addition, in the scenario of the information retrieval method of the present application, the information retrieval device 100 may be provided with a touch-sensitive display device 200, or the information retrieval device 100 may not be provided with a display device and may be communicated with an external display device, and the display device 200 is used to output the result of the execution of the information retrieval method in the information retrieval device. The information retrieval device 100 may access a backend database 300 (the backend database may be in the local memory of the information retrieval device, or may be provided in the cloud), and the backend database 300 may store text related to information retrieval, for example, an initial image in the backend database 300, or pre-set filtering parameters.
[0080] It should be noted that Figure 1 The scenario diagram of the information retrieval method shown is only an example. The scenario of the information retrieval method described in the embodiment of the present invention is to more clearly illustrate the technical solution of the embodiment of the present invention, and does not constitute a limitation on the technical solution provided by the embodiment of the present invention.
[0081] Based on the scenario of the above information retrieval method, an embodiment of the information retrieval method is proposed.
[0082] like Figure 2 FIG. 2 is a flow chart of an embodiment of an information retrieval method in an embodiment of the present application. The information retrieval method includes steps S201-S203:
[0083] S201. Obtain a search term and a text title associated with the search term.
[0084] The search term, i.e., the search term query input by the user, may be, for example, "XX foundation review", "best movie", etc.
[0085] Among them, the text title, that is, the text title corresponding to the information to be recommended, is exemplified as follows: if the information to be recommended is a note, then the text title can be the note text title of the note to be recommended; if the information to be recommended is a video, then the text title can be the video text title of the video to be recommended. That is, the text title can be understood as the identification text information of each information to be recommended. It can be understood that the text title can be the existing text title of the information to be recommended, or the title text of each information to be recommended in the database. Exemplarily, when the information enters the database, a unique title text is assigned to the information to be recommended that enters the database.
[0086] Specifically, the text title associated with the search term is obtained, that is, the text title associated with the search term is obtained through data search. It can be understood that the specific implementation method of obtaining the text title is not specifically limited in this application, and is exemplary:
[0087] In one implementation scheme of the present application, a recommended information database is searched based on a search term to obtain text titles of all information to be recommended that is related to the search term. It can be understood that due to the gap in the spatial distribution of information, the text titles found are text titles of information to be recommended that is related to the search term, and their number (i.e., the number of information to be recommended) will be relatively large.
[0088] In another embodiment of the present application, a similarity search is performed between the search term and a preset search term library to obtain associative search terms similar to the search term, and the text title of the information to be recommended associated with the search term and the associative search term in the historical search record is obtained.
[0089] Specifically, in one implementation scheme of the present application, the information retrieval method is applied to an intelligent terminal, such as a mobile phone, a tablet, etc. The intelligent terminal has a software platform corresponding to the information retrieval method, and an output port (such as an input box) of the search term query is provided in the software platform. The user inputs the search term query to be retrieved based on the input port. After the search term is obtained based on the input port, the software platform searches the recommended information database based on the search term query, and obtains the text titles of all the information to be recommended that are related to the search term, that is, performs a preliminary search to obtain a large number of text titles of the information to be recommended that correspond to the search term query.
[0090] S202: Obtain reference search term clusters associated with each text title. The reference search term clusters are constructed based on historical browsing information and historical search terms corresponding to each piece of historical browsing information.
[0091] Among them, historical browsing information, that is, historical search terms that the user has searched historically, the information that the user has browsed historically in the recommended historical target recommendation information, it can be understood that the historical browsing information can be a note, a video, a piece of music, etc. Historical search terms, that is, historical search terms query that the user has input historically.
[0092] Specifically, the construction of the reference search word cluster is not specifically limited in this application, and is exemplified as follows:
[0093] In one of the implementation schemes of the present application, by counting the historical search terms corresponding to each piece of historical browsing information, all the historical search terms are aggregated to construct a reference search term cluster of the text title corresponding to the historical browsing information.
[0094] In another embodiment of the present application, by counting the historical search terms corresponding to each historical browsing information, according to the search frequency of the historical browsing information corresponding to the historical search terms, a part of the historical search terms are extracted and aggregated to construct a reference search term cluster for the text title corresponding to the historical browsing information.
[0095] Among them, the reference search term cluster, that is, all searched search terms (search term query) are aggregated for a piece of information to be recommended. It can be understood that the reference search term cluster includes multiple reference historical search terms. For example, for a doc (text) that has been browsed historically, all searched search term queries are aggregated to generate a reference search term cluster corresponding to the text. For example, for a review text of XX foundation liquid, the searched historical search term queries include "XX foundation liquid evaluation", "XX foundation liquid is easy to use", "XX foundation liquid is good", etc., that is, all or part of the search term queries in "XX foundation liquid evaluation", "XX foundation liquid is easy to use", "XX foundation liquid is good", etc. can be aggregated into the reference search term cluster of the review text of XX foundation liquid.
[0096] S203: Determine the target text title corresponding to the search term based on the search term, each text title, and the reference search term cluster corresponding to each text title, and output target recommendation information corresponding to the target text title.
[0097] The target text title is a part of the text title in the target text title, and the target recommendation information is the recommendation information corresponding to the target text title.
[0098] Specifically, after determining the search term, text title and reference search term clusters corresponding to each text title, the terminal uses the reference search term clusters corresponding to each text title to replace the information to be recommended corresponding to the text title to perform similarity calculations with the search term respectively. The calculation is based on the reference search term cluster to ensure the comprehensiveness and accuracy of the reference search term cluster in summarizing the recommended information. At the same time, short text calculations reduce the amount of calculations, avoid the spatial distribution gap between the search term and the information to be recommended, ensure the accuracy of data calculations, simplify the amount of calculation data, occupy less resources, and improve the search speed. Further, based on the calculated similarity between the search term and each reference search term cluster, at least one target reference search term cluster with a larger similarity is determined according to the similarity, and based on the association between the reference search term cluster and the text title, the target text title corresponding to the target reference search term cluster is determined, and then the target recommended information corresponding to the target text title is determined, so as to realize fast and accurate information retrieval.
[0099] Further, based on the above implementation scheme, see Figure 3 , Figure 3 This is a schematic flow chart of an embodiment of the information retrieval method provided in the embodiment of the present application for constructing a reference search word cluster, including steps S301-S302:
[0100] S301. Obtain historical browsing information browsed by the user based on the user's historical behavior data.
[0101] Among them, historical behavior data, that is, user behavior data based on historical target recommendation information, exemplarily, multiple target historical recommendation information is displayed on a non-touch display screen at the same time, and the historical behavior data can be that a mouse click triggers the behavior of viewing a certain historical target recommendation information; or multiple target historical recommendation information is displayed on a touch-sensitive display screen at the same time, and the user clicks, slides or touches the touch-sensitive display screen to trigger the behavior of viewing a certain historical target recommendation information.
[0102] It is understandable that the historical behavior data corresponding to the historical browsing information browsed by the user can be stored in the user browsing log or the user behavior log.
[0103] S302, for each historical browsing information, obtain the historical search words corresponding to the historical browsing information, count the repetition frequency of each historical search word, determine the reference historical search words corresponding to the historical browsing information, and construct a reference search word cluster corresponding to the historical browsing information based on the reference historical search words.
[0104] Among them, the repetition frequency, that is, for each historical browsing information, the same number of the same search terms in all historical search terms that have been searched. It can be understood that the user may search for the same historical search term at different historical search times, and click to browse the same historical target recommendation information (that is, historical browsing information).
[0105] Specifically, in one implementation of the present application, the terminal obtains the user's search click log, and extracts the corresponding historical browsing information and the historical search terms corresponding to the historical browsing information in the search click log. The repetition frequency of the historical search terms is counted for each historical browsing information, and then a query-doc graph is constructed. For example, for a historical browsing doc, all searched historical search term queries are aggregated, and the top 20 historical search term queries with the highest repetition frequency (the top 20 historical search term queries with the highest repetition frequency) are taken out from the current historical search term query cluster, and are considered to be synonymous queries of the current historical browsing doc, that is, the top 20 historical search term queries with the highest repetition frequency (the top 20 historical search term queries with the highest repetition frequency) are aggregated into a reference search term cluster of the historical browsing doc.
[0106] Further, based on the above implementation scheme, see Figure 4 , Figure 4 This is a flow chart of an embodiment of determining a target text title in an information retrieval method provided in an embodiment of the present application, including steps S401-S404:
[0107] S401, input the search words, each text title, and the reference search word cluster associated with each text title into the prediction model, which is the BERT model.
[0108] The prediction model is obtained through pre-training. Specifically, the prediction model is used to predict the relevance between each text title and the search term. It can be understood that the prediction model can be stacked by multiple layers of transform structures.
[0109] S402 . Perform feature extraction on the search term and each reference historical search term in the reference search term cluster respectively through a feature extraction module in the prediction model to obtain search term features and reference historical search term features.
[0110] Specifically, the feature extraction module in the prediction model performs feature extraction on the search word and each reference historical search word in the reference search word cluster to obtain the search word feature and the reference historical search word feature. Exemplarily, the feature extraction module is an embedding module, and the embedding module performs embedding feature extraction on the search word and each reference historical search word in the reference search word cluster to obtain the search word embedding and the reference historical search word embedding.
[0111] S403: Process the search term features and the reference historical search term features through the calculation module in the prediction model to obtain the relevance of the search term to each reference search term cluster.
[0112] Furthermore, for each reference search term cluster, the calculation module in the prediction model calculates the vector distance between each reference historical search term and the reference historical search term, and determines the relevance between the search term and the reference search term cluster according to the distance.
[0113] S404. Extract a target reference search word cluster from the reference search word cluster according to the relevance corresponding to the reference search word cluster, and set the text title corresponding to the target reference search word cluster as the target text title corresponding to the search word.
[0114] Furthermore, after determining the relevance corresponding to each reference search term cluster, that is, the relevance between the reference search term cluster and the text title, the prediction model determines the relevance between the text title corresponding to the reference search term cluster and the search term.
[0115] Furthermore, based on the above implementation scheme, the present application also provides an implementation scheme for prediction model training, comprising the steps of:
[0116] (1) Construct a training sample set based on the user's historical search terms and the historical recommendation information corresponding to the historical search terms;
[0117] The historical recommendation information is information recommended based on historical search terms, and the historical browsing information is all or part of the historical recommendation information.
[0118] Among them, the historical search terms, that is, the search terms queries that the user has searched historically, and the historical recommendation information corresponding to the historical search terms, that is, multiple historical target recommendation information obtained for each search term query.
[0119] Specifically, the steps for constructing training samples are as follows:
[0120] A. For each historical search term, count the browsing frequency of each historical recommended information corresponding to the historical search term to obtain the average browsing frequency;
[0121] B. determining the relevance between each of the historical recommendation information and the historical search term according to the magnitude relationship between the browsing frequency corresponding to each of the historical recommendation information and the average browsing frequency;
[0122] C. Setting the historical search term and each of the historical browsing information with set relevance associated with the historical search term as a training sample.
[0123] Exemplarily, the terminal obtains the user's search log, which includes the user's historical search terms and historical browsing information corresponding to each historical search term (historical target recommendation information that has been clicked and browsed), analyzes the search log, and counts the browsing frequency of each historical browsing information corresponding to the historical search term for each historical search term. It can be understood that the user may click on the same historical target recommendation information multiple times for a historical search term, and determines the relevance of each historical browsing information based on the relationship between the browsing frequency corresponding to each historical browsing information and the average browsing frequency, that is, if the browsing frequency corresponding to the historical browsing information is greater than or equal to the average browsing frequency, the relevance of the historical browsing information is positively correlated, and if the browsing frequency corresponding to the historical browsing information is less than the average browsing frequency, the relevance of the historical browsing information is negatively correlated; based on the relevance of the historical browsing information, a positively correlated information set and a negatively correlated information set corresponding to each historical search term are generated, the positively correlated information set includes multiple positively correlated historical browsing information; the negatively correlated information set includes multiple negatively correlated historical browsing information.
[0124] Exemplarily, the click information of all notes under the same historical search term query is counted, the click frequency of all notes under the current query, and the average click rate of the current query are counted, and the average click rate is used as the threshold, and the query-note click rate is divided into positive examples (positive correlation) and negative examples (negative correlation).
[0125] (2) According to each training sample in the training sample set, the preset BERT model is trained to obtain a prediction model;
[0126] Each historical search word in the training sample, as well as the text title corresponding to the historical browsing information corresponding to each historical search word and the corresponding reference search word cluster are replaced as input, and the relevance of the historical browsing information to the historical search word is used as the output label. The preset BERT model is trained to obtain a prediction model.
[0127] Furthermore, based on the above implementation scheme, the present application also provides an implementation scheme for training a feature extraction module in a prediction model, including:
[0128] (1) For each text title, each reference historical search term in the reference search term cluster associated with the text title is input into a feature extraction module in the prediction model, and the reference historical search term features corresponding to each reference historical search term are output;
[0129] (2) Calculate the feature mean of the reference historical search term features corresponding to the reference search term cluster, and adjust the parameters of the feature extraction module according to the feature mean.
[0130] Specifically, in the implementation scheme of the present application, each reference historical search term in the reference search term cluster corresponding to each text title is input into the embedding module in the prediction model to obtain the embedding representation corresponding to each reference historical search term query in the reference search term cluster, that is, the reference historical search term embedding is obtained, and the reference historical search term embedding representation corresponding to each reference search term cluster is averaged to obtain the feature mean, and the parameters of the embedding module are adjusted according to the feature mean to increase the robustness of the embedding module to data noise.
[0131] Further, based on the above implementation scheme, see Figure 5 , Figure 5 This is a flow chart of an embodiment of the output of target recommendation information in the information retrieval method provided in the embodiment of the present application, including steps S501-S503:
[0132] S501: Extract key information from target recommendation information corresponding to each target text title.
[0133] Among them, the key information includes at least one of a key text segment, a key image and a key audio segment. It can be understood that the target recommendation information may be a video, a text note, a picture and text note, or an audio, etc. Correspondingly, the key text segment can be the total stage or the starting segment in the text note, or the key field extracted based on the text note, etc. The key image can be any image in the picture and text note or any image frame in the video (such as a cover frame), and the key audio can be any music clip in the video or audio.
[0134] Exemplarily, where any of the target recommended information is a picture and text note (i.e., a note including pictures and text), after determining the target text title, the terminal obtains the picture and text note corresponding to the target text title, extracts the first image of the picture and text title as the key image, extracts high-frequency words and sentences in the picture and text note as the key text segments, and then obtains the key image and key text segments corresponding to the picture and text note corresponding to the target text title.
[0135] S502: Add each target text title and corresponding key information to each text column in the preset recommendation page template.
[0136] Among them, the recommendation page template, that is, the recommendation page display template corresponding to each recommendation platform, can be understood that the corresponding display controls can be determined in each text column in the preset recommendation page template corresponding to the graphic information and audio information, and the controls can be adjusted according to the key information. For example, if the key information includes key images and key text ends, the text column includes image display controls and text display controls; when the terminal obtains the target text title and the corresponding key information, the key information is added to the corresponding position in the text column.
[0137] S503: Create hyperlinks between each text column and the target recommendation information corresponding to the target text title in the text column to obtain a recommendation page.
[0138] Specifically, after adding each key information to each text column in the preset recommendation page template, the terminal creates a hyperlink between each text column and the target recommendation information corresponding to the target text title in the text column, so that the text column responds to the user's behavior data when the user browses, triggers the link jump of any text column, and displays the details page of the target recommendation information corresponding to the text column. It can be understood that the behavior data is clicked or slid, and the click includes single-click, double-click, long press, short press, etc., and the sliding includes up and down sliding, left and right sliding, and point sliding, etc.
[0139] Further, based on the above implementation scheme, see Figure 6 , Figure 6 : is another embodiment flow chart of the information retrieval method provided in the embodiment of the present application, including steps S601-S609:
[0140] S601: Obtain a search term and a text title associated with the search term.
[0141] S602: Obtain reference search term clusters associated with each text title. The reference search term clusters are constructed based on historical browsing information and historical search terms corresponding to each piece of historical browsing information.
[0142] S603: Determine a target text title corresponding to the search term based on the search term, each text title, and a reference search term cluster corresponding to each text title.
[0143] S604: extract key information from the target recommendation information corresponding to each target text title, wherein the key information includes at least one of a key text segment, a key image, and a key audio segment.
[0144] S605: Add each target text title and corresponding key information to each text column in the preset recommendation page template.
[0145] S606: Create hyperlinks between each text column and the target recommendation information corresponding to the target text title in the text column to obtain a recommendation page.
[0146] For the specific implementation of steps S601-S606, please refer to any of the above implementation schemes.
[0147] S607: Display the recommendation page and receive user operations on the recommendation page.
[0148] Specifically, in one implementation scheme of the present application, the terminal includes a touch-sensitive display interface, and the terminal displays the recommendation page based on the touch-sensitive display interface, and accepts user operations based on the recommendation page, wherein the operations include clicking or sliding, and the clicks include single-clicking, double-clicking, long pressing, short pressing, etc., and the sliding includes sliding up and down, sliding left and right, and point sliding, etc.
[0149] S608: If the user operation triggers a hyperlink corresponding to any text field in the recommendation page, the target recommendation information corresponding to the hyperlink is displayed.
[0150] Furthermore, when the terminal detects that a user operation triggers a hyperlink corresponding to any text field in the recommendation page, the recommendation page is controlled to jump to the text page (details page) of the target recommendation information corresponding to the hyperlink for display; exemplarily, it is preset that when the user clicks on any text field in the recommendation page, the hyperlink corresponding to the text field is triggered. Then, when the terminal detects that the user clicks on any text field in the recommendation page, the hyperlink corresponding to any text field in the recommendation page is triggered, and the recommendation page is controlled to jump to the text page (details page) of the target recommendation information corresponding to the hyperlink for display, and the text page includes all information of the target recommendation information.
[0151] S609. When a recommendation page closing instruction triggered by a user operation is received, the recommendation page is controlled to be closed, and the displayed target recommendation information corresponding to the search term is obtained, the search term is set as a historical search term, and the displayed target recommendation information is set as historical browsing information.
[0152] It can be understood that the user operation that triggers the recommendation page closing instruction can be a preset user operation. For example, it can be a left swipe operation or a click swipe operation. For example, if the user operation that triggers the recommendation page closing instruction is a left swipe operation, then when the terminal detects that the user operation is a left swipe operation, the recommendation page is controlled to be closed, and the platform main page can be switched to display. It can be understood that if the user operation triggers a hyperlink, the corresponding hyperlink jump display is triggered, and the recommendation page will not be closed. The recommendation page will be displayed at the bottom layer of the display window, and the text page corresponding to the hyperlink will be displayed at the upper layer.
[0153] Furthermore, after the control recommendation page is closed, the target recommendation information displayed corresponding to the closed display page is counted, that is, the target recommendation information that has triggered the hyperlink. It can be understood that the search term is set as the historical search term, and the displayed target recommendation information is set as the historical browsing information corresponding to the search term, so as to refer to the update of the search term cluster. Exemplarily, the search term and the corresponding displayed historical browsing information are updated to the user's search log, that is, the search term and the corresponding displayed target recommendation information are updated in the user's search log, and the search term and the corresponding displayed historical browsing information are further updated to the search click log, that is, the clicked displayed target recommendation information and the search term corresponding to the target recommendation information are updated in the search click log.
[0154] It is understandable that the reference search term cluster may be updated after new historical search terms and new historical browsing information are detected, or may be updated according to a preset update frequency.
[0155] Specifically, in the implementation scheme of the present application, the terminal obtains the search term and the text title associated with the search term; obtains the reference search term cluster associated with each text title, the reference search term cluster is constructed based on the historical browsing information and the historical search term corresponding to each historical browsing information; after determining the target text title corresponding to the search term based on the search term, each text title, and the reference search term cluster corresponding to each text title, a recommendation page is generated and displayed based on the target recommendation information corresponding to the text title, and user operations on the recommendation page are received, including clicks or slides; if the user operation triggers a hyperlink corresponding to any text column in the recommendation page, the target recommendation information corresponding to the hyperlink is displayed; when a recommendation page closing instruction triggered by a user operation is received, the recommendation page is controlled to close, and the displayed target recommendation information corresponding to the search term is obtained, the search term is set as the historical search term, and the displayed target recommendation information is set as the historical browsing information. Real-time updates of the reference search term cluster are achieved to ensure the accuracy and comprehensiveness of the data.
[0156] The implementation scheme of the present application provides an information retrieval method, by obtaining search terms and text titles associated with the search terms; and obtaining reference search term clusters associated with each text title, the reference search term clusters are constructed based on historical browsing information and historical search terms corresponding to each historical browsing information; then, according to the search terms, each text title, and the reference search term clusters corresponding to each text title, the target text title corresponding to the search term is determined, and the target recommendation information corresponding to the target text title is output. In this scheme, according to the user's historical browsing information and the historical search terms corresponding to each historical browsing information, a reference search term cluster corresponding to the text title is constructed, and the text title associated with the search term is preliminarily obtained, and the information content is replaced with the text title to simplify the amount of data analysis. Further, the reference search term cluster corresponding to the text title and the search term are analyzed, that is, the reference search term cluster is used to replace the text corresponding to the text title for search information analysis, reducing the information distribution space difference between the search terms and the recommended information, and the reference search term cluster is constructed to replace the text corresponding to the text title, ensuring the accuracy and comprehensiveness of the reference search term cluster, and then determining the target text title corresponding to the search term in the text title, and further determining the corresponding target recommendation information according to the target text title, ensuring the accuracy of the target recommendation information.
[0157] In combination with any of the above embodiments, further, see Figure 7 , Figure 7 A schematic flow chart of an implementation scheme of the prediction model training method provided in the embodiment of the present application includes steps S701-S704:
[0158] S701, obtain historical search terms and historical browsing information browsed by the user.
[0159] The historical search terms, that is, the search terms that the user has searched historically, the definition of the search terms will not be elaborated in detail.
[0160] The historical browsing information refers to information browsed by the user in the historical recommendation information recommended based on historical search terms. Specifically, the historical browsing information can be obtained through user logs.
[0161] S702: Set each of the historical search terms and each of the historical recommendation information with set relevance associated with each of the historical search terms as a training sample to obtain a training sample set.
[0162] Specifically, the historical recommendation information with set relevance is obtained through the following steps:
[0163] (1) For each of the historical search terms, execute:
[0164] (2) counting the browsing frequency of each of the historical recommended information corresponding to the historical search terms to obtain an average browsing frequency;
[0165] (3) determining the relevance between each of the historical recommendation information and the historical search term according to the magnitude relationship between the browsing frequency corresponding to each of the historical recommendation information and the average browsing frequency;
[0166] (4) Setting the relevance of the historical recommendation information to the historical text title.
[0167] S703. For each piece of historical browsing information, obtain the historical search term corresponding to the historical browsing information, count the repetition frequency of each of the historical search terms, determine the historical search term corresponding to the historical browsing information, and construct a reference search term cluster corresponding to the historical browsing information based on the reference historical search term.
[0168] S704: Train a preset training model according to each of the training samples in the training sample set and the search word family to obtain a prediction model.
[0169] Specifically, the implementation scheme of this application realizes the acquisition of the prediction model based on transfer learning and offline training, exemplarily:
[0170] During the offline phase:
[0171] Acquiring training samples specifically includes: according to the user's search log, counting the click information of all notes (historical recommendation information) under the same query (historical search term), and counting the click frequency of all notes under the current query, that is, counting the browsing frequency of each of the historical recommendation information corresponding to the historical search term to obtain the average browsing frequency; further, calculating the average click rate of a query, that is, calculating the average browsing frequency; taking the average click rate as a threshold, dividing the query-note into positive examples and negative examples according to the click frequency, that is, determining the relevance between each of the historical recommendation information and the historical search term according to the size relationship between the browsing frequency corresponding to each of the historical recommendation information and the average browsing frequency. It can be understood that positive examples and negative examples are equivalent to setting the historical recommendation information of relevance (represented by the historical text title of the historical recommendation information, that is, setting the historical recommendation information of relevance is equivalent to setting the historical text title of relevance). Exemplarily, notes with a click frequency greater than the average click rate are set as positive examples, and notes with a click frequency less than or equal to the average click rate are set as negative examples, positive examples are marked with 1, and negative examples are marked with 0, that is, 1 and 0 are relevance.
[0172] The transfer learning data specifically includes: constructing a query-doc graph based on the user's search click log, that is, obtaining the corresponding historical browsing information for each historical search term; further, aggregating all searched queries for a doc (a historical browsing information), that is, obtaining the historical search term corresponding to each historical browsing information, taking out the top 20 queries in the current query cluster, and considering them as the synonymous queries of the current note, which can represent this note, that is, counting the repetition frequency of each historical search term, determining the reference historical search term corresponding to the historical browsing information, and constructing the reference search term cluster corresponding to the historical browsing information based on the reference historical search term. Further, obtaining the embedding representation of the synonymous query of the note, that is, obtaining the embedding representation of the query for the 20 synonymous queries (reference search term clusters) of the note, and averaging the embedding representation of the query to increase the robustness to data noise and realize the vectorization processing of the reference search term cluster.
[0173] Online stage:
[0174] First, data is preprocessed, specifically including preprocessing the input query + note title, including removing special characters, symbols, and case conversion, etc., to obtain historical search terms (it can be understood that the process of obtaining historical search terms is online processing when historical input is input, and subsequent training can directly obtain historical search terms).
[0175] Further, for each historical search word in the training sample, a corresponding sample reference search word cluster is found in the search word cluster to obtain the sample reference search word cluster corresponding to the training sample;
[0176] For details, see Figure 8 , input query (historical search term), note title (historical text title of historical recommendation information) and note synonymous query embedding representation (i.e. synonymous query in the figure, also sample reference search term cluster), output the correlation score of query and note, until the correlation score meets the convergence condition, and obtain the prediction model. That is, the training sample and the sample reference search term cluster corresponding to the training sample are input into the preset BERT model for training to obtain the prediction model.
[0177] In order to better implement the information retrieval method in the embodiment of the present application, based on the information retrieval method, an information retrieval device is also provided in the embodiment of the present application, such as Fig. 9 As shown, the information retrieval device includes modules 901-903:
[0178] The first acquisition module 901 is used to acquire a search term and a text title associated with the search term;
[0179] The second acquisition module 902 is used to acquire reference search word clusters associated with each text title, where the reference search word clusters are constructed based on historical browsing information and historical search words corresponding to each historical browsing information;
[0180] Determination module 903: is used to determine the target text title corresponding to the search term according to the search term, each text title, and the reference search term cluster corresponding to each text title, and output the target recommendation information corresponding to the target text title.
[0181] In one embodiment of the present application, the information retrieval device further includes a building module: the building module is used to:
[0182] According to the user's historical behavior data, obtain the user's browsing history information;
[0183] For each historical browsing information, the historical search words corresponding to the historical browsing information are obtained, the repetition frequency of each historical search word is counted, the reference historical search words corresponding to the historical browsing information are determined, and the reference search word cluster corresponding to the historical browsing information is constructed according to the reference historical search words.
[0184] In one embodiment of the present application, the determination module 903 is used to determine the target text title corresponding to the search term according to the search term, each text title, and the reference search term cluster corresponding to each text title, and specifically includes:
[0185] Input the search terms, each text title, and the reference search term cluster associated with each text title into the prediction model, which is the BERT model;
[0186] The feature extraction module in the prediction model extracts features of the search term and each reference historical search term in the reference search term cluster respectively to obtain search term features and reference historical search term features;
[0187] The search term features and the reference historical search term features are processed by the calculation module in the prediction model to obtain the relevance of the search term to each reference search term cluster;
[0188] According to the relevance corresponding to the reference search word cluster, a target reference search word cluster is extracted from the reference search word cluster, and the text title corresponding to the target reference search word cluster is set as the target text title corresponding to the search word.
[0189] In one embodiment of the present application, the information retrieval device further includes a model training module: the model training module is used to:
[0190] Construct a training sample set based on the user's historical search terms and the historical browsing information corresponding to the historical search terms;
[0191] According to each training sample in the training sample set, the preset BERT model is trained to obtain a prediction model;
[0192] The steps for constructing training samples are as follows:
[0193] For each historical search term, count the browsing frequency of each historical browsing information corresponding to the historical search term to obtain the average browsing frequency;
[0194] Determine the relevance of each historical browsing information according to the size relationship between the browsing frequency corresponding to each historical browsing information and the average browsing frequency;
[0195] The historical search words and each historical browsing information with set relevance associated with the historical search words are set as a training sample.
[0196] In one embodiment of the present application, the information retrieval device further includes a model training module: the model training module further includes:
[0197] For each text title, each reference historical search word in the reference search word cluster associated with the text title is input into a feature extraction module in the prediction model, and the reference historical search word features corresponding to each reference historical search word are output;
[0198] The feature mean of the reference historical search term features corresponding to the reference search term cluster is calculated, and the parameters of the feature extraction module are adjusted according to the feature mean.
[0199] In one embodiment of the present application, the information retrieval device further includes a statistical module, and the determination module 903 is used to: output target recommendation information corresponding to the target text title, specifically including:
[0200] Generate and display a recommendation page based on the target recommendation information corresponding to the text title;
[0201] The determination module 903 is used to: output the target recommendation information corresponding to the target text title, and the statistics module is used to:
[0202] Receive user actions on the recommended page, including clicks or swipes;
[0203] If the user operation triggers the hyperlink corresponding to any text field in the recommendation page, the target recommendation information corresponding to the hyperlink will be displayed;
[0204] When a recommendation page closing instruction triggered by a user operation is received, the recommendation page is controlled to be closed, and the displayed target recommendation information corresponding to the search term is obtained, the search term is set as a historical search term, and the displayed target recommendation information is set as historical browsing information.
[0205] In one embodiment of the present application, the determination module 903 is used to generate and display a recommendation page according to the target recommendation information corresponding to the text title, specifically including:
[0206] Extracting key information from the target recommendation information corresponding to each target text title, wherein the key information includes at least one of a key text segment, a key image, and a key audio segment;
[0207] Add each target text title and corresponding key information to each text column in the preset recommendation page template;
[0208] Create a hyperlink between each text column and the target recommendation information corresponding to the target text title in the text column, obtain the recommendation page and display it.
[0209] Based on the above implementation scheme, the embodiment of the present invention also provides an information retrieval device, such as Fig.10 As shown, Fig.10 It is a schematic diagram of the structure of an embodiment of the information retrieval device provided in the embodiments of the present application.
[0210] Information retrieval equipment includes:
[0211] one or more processors;
[0212] Memory; and
[0213] One or more applications, wherein the one or more applications are stored in the memory and are configured to execute, by the processor, the steps of the information retrieval method in any of the above-mentioned information retrieval method embodiments.
[0214] Specifically, the information retrieval device may include one or more processing core processors 1001, one or more computer-readable storage media memories 1002, a power supply 1003, an input unit 1004, and other components. Those skilled in the art will understand that Fig.10 The information retrieval device structure shown in the figure does not constitute a limitation on the information retrieval device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.
[0215] The processor 1001 is the information retrieval center, which uses various interfaces and lines to connect various parts of the entire information retrieval device. By running or executing software programs and / or modules stored in the memory 1002, and calling data stored in the memory 1002, it performs various functions of the information retrieval device and processes data, thereby monitoring the information retrieval device as a whole. It can be understood that the processor 1001 transmits signals with the controller. Optionally, the processor 1001 may include one or more processing cores; preferably, the processor 1001 can integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface and application programs, etc., and the modem processor mainly processes wireless communications. It can be understood that the above-mentioned modem processor may not be integrated into the processor 1001.
[0216] The memory 1002 can be used to store software programs and modules. The processor 1001 executes various functional applications and data processing by running the software programs and modules stored in the memory 1002. The memory 1002 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the information retrieval device, etc. In addition, the memory 1002 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices. Accordingly, the memory 1002 may also include a memory controller to provide the processor 1001 with access to the memory 1002.
[0217] In some embodiments of the present application, the information retrieval device can be implemented in the form of a computer program. The computer program can be used in Fig.10The information retrieval device may store various program modules constituting the information retrieval method device, such as: Fig. 9 The first acquisition module 901, the second acquisition module 902, and the determination module 903 are shown. The computer program composed of various program modules enables the processor to execute the steps of the information retrieval method of each embodiment of the present application described in this specification.
[0218] For example, Fig.10 The information retrieval device shown can be Fig. 9 The first acquisition module 901 in the information retrieval method device shown executes step S201. The information retrieval device can execute step S202 through the second acquisition module 902. The information retrieval device can execute step S203 through the determination module 903. The information retrieval device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the information retrieval device is used to provide computing and control capabilities. The memory of the information retrieval device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the information retrieval device is used to communicate with an external information retrieval device through a network connection. When the computer program is executed by the processor, an information retrieval method is implemented.
[0219] The information retrieval device also includes a power supply 1003 for supplying power to each component. Preferably, the power supply 1003 can be logically connected to the processor 1001 through a power management system, so as to manage charging, discharging, and power consumption through the power management system. The power supply 1003 can also include any components such as one or more DC or AC power supplies, recharging systems, power failure detection circuits, power converters or inverters, and power status indicators.
[0220] The information retrieval device may further include an input unit 1004, which may be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function controls.
[0221] Although not shown, the information retrieval device may also include a display unit, etc., which will not be described in detail here. Specifically in this embodiment, the processor 1001 in the information retrieval device will load the executable files corresponding to the processes of one or more application programs into the memory 1002 according to the following instructions, and the processor 1001 will run the application programs stored in the memory 1002, thereby realizing various functions, as follows:
[0222] Get the search term and the text title associated with the search term;
[0223] Obtaining reference search term clusters associated with each text title, where the reference search term clusters are constructed based on historical browsing information and historical search terms corresponding to each historical browsing information;
[0224] According to the search words, each text title, and the reference search word cluster corresponding to each text title, the target text title corresponding to the search word is determined, and the target recommendation information corresponding to the target text title is output.
[0225] A person of ordinary skill in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be completed by instructions, or by controlling related hardware through instructions. The instructions may be stored in a computer-readable storage medium and loaded and executed by a processor.
[0226] To this end, an embodiment of the present invention provides a computer-readable storage medium (hereinafter referred to as a storage medium), which may include: a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc. A computer program is stored thereon, and the computer program is loaded by a processor to execute the steps in any one of the information retrieval methods provided in the embodiments of the present invention. For example, the computer program may be loaded by a processor to execute the following steps:
[0227] Get the search term and the text title associated with the search term;
[0228] Obtaining reference search term clusters associated with each text title, where the reference search term clusters are constructed based on historical browsing information and historical search terms corresponding to each historical browsing information;
[0229] According to the search words, each text title, and the reference search word cluster corresponding to each text title, the target text title corresponding to the search word is determined, and the target recommendation information corresponding to the target text title is output.
[0230] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the detailed description of other embodiments above, and will not be repeated here.
[0231] In specific implementation, the above units or structures can be implemented as independent entities, or can be arbitrarily combined to be implemented as the same or several entities. The specific implementation of the above units or structures can refer to the previous method embodiments, which will not be repeated here.
[0232] The specific implementation of the above operations can be found in the previous embodiments, which will not be described in detail here.
[0233] The above is a detailed introduction to an information retrieval method, device, prediction model training method, information retrieval equipment and storage medium provided in the embodiments of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for technical personnel in this field, according to the idea of the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.
Claims
1. An information retrieval method, characterized in that: include: Obtain a search term and a text title associated with the search term; Acquire a reference search word cluster associated with each of the text titles, wherein the reference search word cluster is constructed based on historical browsing information and a historical search word corresponding to each of the historical browsing information; Determine the target text title corresponding to the search term according to the search term, each of the text titles, and the reference search term cluster corresponding to each of the text titles, and output target recommendation information corresponding to the target text title; The construction step of the reference search word cluster includes: According to the user's historical behavior data, obtain the user's browsing history information; For each piece of historical browsing information, obtaining a historical search term corresponding to the historical browsing information, counting the repetition frequency of each of the historical search terms, determining a reference historical search term corresponding to the historical browsing information, and constructing a reference search term cluster corresponding to the historical browsing information according to the reference historical search term; The repetition frequency indicates the same number of the same search terms in all the historical search terms that have been searched for each of the historical browsing information; The counting of the repetition frequencies of the historical search terms, determining the reference historical search terms corresponding to the historical browsing information, and constructing the reference search term cluster corresponding to the historical browsing information according to the reference historical search terms, comprises: taking the first plurality of the historical search terms with the highest repetition frequencies as the reference historical search terms for the historical browsing information, and aggregating the reference historical search terms into the reference search term cluster corresponding to the historical browsing information; Wherein, determining the target text title corresponding to the search term according to the search term, each of the text titles, and the reference search term cluster corresponding to each of the text titles includes: Extract features from the search term and each of the reference historical search terms in the reference search term cluster respectively by using a feature extraction module in the prediction model to obtain search term features and reference historical search term features; Performing vector distance calculation on the search term feature and the reference historical search term feature through a calculation module in the prediction model to obtain a vector distance, and determining the relevance between the search term and the reference search term cluster according to the magnitude of the vector distance; Extracting a target reference search word cluster from the reference search word cluster according to the relevance corresponding to the reference search word cluster, and setting the text title corresponding to the target reference search word cluster as the target text title corresponding to the search word; The training of the feature extraction module includes the following steps: For each text title, each reference historical search word in the reference search word cluster associated with the text title is input into the feature extraction module in the prediction model, and the reference historical search word features corresponding to each reference historical search word are output; the feature mean of the reference historical search word features corresponding to the reference search word cluster is calculated, and the parameters of the feature extraction module are adjusted according to the feature mean.
2. The information retrieval method according to claim 1, characterized in that: The feature extraction module in the prediction model extracts features from the search term and each reference historical search term in the reference search term cluster, and obtains the search term features and the reference historical search term features, and also includes: The search term, each of the text titles, and the reference search term cluster associated with each of the text titles are input into a prediction model, which is a BERT model.
3. The information retrieval method according to claim 2, characterized in that: The prediction model is trained by the following method: Constructing a training sample set based on the user's historical search terms and the historical recommendation information corresponding to the historical search terms; According to each training sample in the training sample set, the preset BERT model is trained to obtain a prediction model; The training sample construction steps are as follows: For each historical search term, execute: Counting the browsing frequency of each of the historical recommendation information corresponding to the historical search terms to obtain an average browsing frequency; Determining the relevance between each of the historical recommendation information and the historical search term according to the magnitude relationship between the browsing frequency corresponding to each of the historical recommendation information and the average browsing frequency; The historical search word and each of the historical browsing information with set relevance associated with the historical search word are set as a training sample.
4. The information retrieval method according to any one of claims 1 to 3, characterized in that: The outputting the target recommendation information corresponding to the target text title includes: Extracting key information from the target recommendation information corresponding to each of the target text titles, wherein the key information includes at least one of a key text segment, a key image, and a key audio segment; Adding each of the target text titles and the corresponding key information to each text column in a preset recommendation page template; A hyperlink is created between each of the text columns and the target recommendation information corresponding to the target text title in the text column, and a recommendation page is obtained and displayed.
5. The information retrieval method according to claim 4, characterized in that: After creating a hyperlink between each of the text columns and the target recommendation information corresponding to the target text title in the text column, obtaining a recommendation page and displaying it, the method further includes: Receiving a user operation on the recommendation page, the operation including clicking or sliding; If the user operation triggers a hyperlink corresponding to any text field in the recommendation page, the target recommendation information corresponding to the hyperlink is displayed; When the recommendation page closing instruction triggered by the user operation is received, the recommendation page is controlled to be closed, and the displayed target recommendation information corresponding to the search term is obtained, the search term is set as a historical search term, and the displayed target recommendation information is set as historical browsing information.
6. A prediction model training method, characterized in that: include: Get historical search terms and historical browsing information that the user has browsed; Obtaining a text title associated with the historical search term, and obtaining a search term cluster associated with the text title; Setting each of the historical search terms and each of the historical recommendation information with set relevance associated with each of the historical search terms as a training sample to obtain a training sample set; For each piece of historical browsing information, obtaining a historical search term corresponding to the historical browsing information, counting the repetition frequency of each of the historical search terms, determining a reference historical search term corresponding to the historical browsing information, and constructing a search term cluster corresponding to the historical browsing information according to the reference historical search term; According to each of the training samples in the training sample set and the search word cluster, a preset training model is trained to obtain a prediction model, and a feature extraction module in the prediction model is used to extract features of the search word and each of the reference historical search words in the reference search word cluster to obtain search word features and reference historical search word features; Performing vector distance calculation on the search term feature and the reference historical search term feature through a calculation module in the prediction model to obtain a vector distance, and determining the relevance between the search term and the reference search term cluster according to the magnitude of the vector distance; Extracting a target reference search word cluster from the reference search word cluster according to the relevance corresponding to the reference search word cluster, and setting the text title corresponding to the target reference search word cluster as the target text title corresponding to the search word; The training of the feature extraction module includes the following steps: For each text title, each of the reference historical search terms in the reference search term cluster associated with the text title is input into the feature extraction module in the prediction model, and the reference historical search term features corresponding to each of the reference historical search terms are output; the feature mean of the reference historical search term features corresponding to the reference search term cluster is calculated, and the parameters of the feature extraction module are adjusted according to the feature mean; The repetition frequency indicates the same number of the same search terms in all the historical search terms that have been searched for each of the historical browsing information; Among them, the counting of the repetition frequency of each of the historical search words, determining the reference historical search words corresponding to the historical browsing information, and constructing the search word cluster corresponding to the historical browsing information based on the reference historical search words includes: taking the first multiple historical search words with the highest repetition frequency as the reference historical search words for the historical browsing information, and aggregating the reference historical search words into a search word cluster corresponding to the historical browsing information.
7. The prediction model training method according to claim 6, characterized in that: The historical recommendation information with set relevance is obtained by the following steps: For each of the historical search terms, execute: Counting the browsing frequency of each of the historical recommendation information corresponding to the historical search terms to obtain an average browsing frequency; Determining the relevance between each of the historical recommendation information and the historical search term according to the magnitude relationship between the browsing frequency corresponding to each of the historical recommendation information and the average browsing frequency; The relevance of the historical recommendation information to the historical text title is set.
8. An information retrieval device, characterized in that: The device comprises: A first acquisition module: used to acquire a search term and a text title associated with the search term; A second acquisition module: used for acquiring reference search word clusters associated with each of the text titles, wherein the reference search word clusters are constructed based on historical browsing information and historical search words corresponding to each of the historical browsing information; A determination module: used to determine the target text title corresponding to the search term according to the search term, each of the text titles, and the reference search term cluster corresponding to each of the text titles, and output target recommendation information corresponding to the target text title; The second acquisition module is used to: acquire historical browsing information browsed by the user according to the user's historical behavior data; for each piece of historical browsing information, acquire the historical search words corresponding to the historical browsing information, count the repetition frequency of each of the historical search words, determine the reference historical search words corresponding to the historical browsing information, and construct the reference search word cluster corresponding to the historical browsing information according to the reference historical search words; The repetition frequency indicates the same number of the same search terms in all the historical search terms that have been searched for each of the historical browsing information; The second acquisition module is specifically configured to use the first plurality of the historical search terms with the highest repetition frequency as reference historical search terms for the historical browsing information, and aggregate the reference historical search terms into a reference search term cluster corresponding to the historical browsing information; Among them, the determination module is specifically used to perform feature extraction on the search word and each of the reference historical search words in the reference search word cluster through the feature extraction module in the prediction model to obtain search word features and reference historical search word features; perform vector distance calculation on the search word features and the reference historical search word features through the calculation module in the prediction model to obtain vector distance, and determine the relevance between the search word and the reference search word cluster according to the size of the vector distance; extract a target reference search word cluster from the reference search word cluster according to the relevance corresponding to the reference search word cluster, and set the text title corresponding to the target reference search word cluster to the target text title corresponding to the search word; The information retrieval device also includes: a model training module, which is used to input each reference historical search term in the reference search term cluster associated with the text title into a feature extraction module in the prediction model for each text title, and output the reference historical search term features corresponding to each reference historical search term; calculate the feature mean of the reference historical search term features corresponding to the reference search term cluster, and adjust the parameters of the feature extraction module according to the feature mean.
9. An information retrieval device, characterized in that: The information retrieval device comprises: one or more processors; Memory; and One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the steps in the information retrieval method according to any one of claims 1 to 5.
10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program is loaded by a processor to execute the steps in the information retrieval method according to any one of claims 1 to 5.
