Query error correction methods, devices, computer equipment, storage media and program products

Abstract

This application provides a query error correction method, apparatus, computer device, storage medium, and program product, relating to the technical fields of artificial intelligence, cloud technology, intelligent transportation, and assisted driving. By determining a first error indication information for the query information based on a target model matched with the user, personalized error judgment is performed based on the user's target model, making the judgment result more closely match the user's actual interests and needs. When the query information is incorrect, at least one candidate information is determined based on target error correction data matched with the user. Subsequently, based on the second error indication information of the candidate information determined by the target model, corrective information is further filtered out. The target error correction data includes at least the user's interest data. Therefore, personalized error correction of the query information based on the user's interests can meet the error correction needs of each individual, improving the actual accuracy of error judgment and correction, and thus improving the actual accuracy rate of query error correction.

Description

Technical Field

[0001] This application relates to the technical fields of artificial intelligence, cloud technology, smart transportation, and driver assistance, and specifically to a query and error correction method, apparatus, computer equipment, storage medium, and program product. Background Technology

[0002] Currently, search engines are one of the most important ways for people to obtain information. Users can simply enter keywords into the search box to find relevant information of interest. However, because users often enter incorrect keywords, query correction has emerged. Query correction identifies and corrects errors in the keywords entered by the user, so that the search engine can return the user's expected search results based on the corrected keywords.

[0003] In related technologies, a unified thesaurus of search engines is usually used for query correction. The thesaurus can collect correct words that appear frequently in search engines. For example, for any user input keyword "direct dial", the thesaurus can be used to uniformly correct "direct dial" to "live broadcast".

[0004] The aforementioned error correction process essentially uses a unified error correction standard to measure the query intent behind each user's keywords. However, there are still some users whose corrected keywords do not match their corresponding query intent, thus failing to obtain the expected results based on the corrected keywords. Therefore, the actual accuracy of the above query error correction is relatively low. Summary of the Invention

[0005] This application provides a method, apparatus, computer device, storage medium, and program product for query error correction, which can solve the problem of low accuracy in actual error correction in related technologies. The technical solution is as follows:

[0006] On the one hand, a query error correction method is provided, the method comprising:

[0007] In response to receiving query information from a user's input search box, a first error indication is determined based on a target model matching the user. The target model is used to determine whether the information input into the target model is incorrect based on the user's interest data in the target application.

[0008] In response to the first error indication information indicating that the query information is incorrect, at least one candidate information associated with the query information is determined based on the target error correction data matched with the user and the query information, wherein the target error correction data includes at least the user's interest data in the target application;

[0009] Based on the target model, a second error indication message is determined for the at least one candidate information;

[0010] Based on the second error indication information, correction information for the query information is determined from the at least one candidate information.

[0011] On the other hand, a query error correction apparatus is provided, the apparatus comprising:

[0012] The first determining module is used to respond to receiving query information from a user input search box, and determine first error indication information of the query information based on a target model matching the user. The target model is used to determine whether the information input into the target model is incorrect based on the user's interest data in the target application.

[0013] A candidate determination module is configured to, in response to the first error indication information indicating that the query information is incorrect, determine at least one candidate information associated with the query information based on target error correction data matching the user and the query information, wherein the target error correction data includes at least the user's interest data in the target application;

[0014] The second determining module is used to determine, based on the target model, the second error indication information of the at least one candidate information;

[0015] A correction module is used to determine correction information for the query information from the at least one candidate information based on the second error indication information.

[0016] On the other hand, a computer device is provided, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the above-described query and error correction method.

[0017] On the other hand, a computer-readable storage medium is provided on which a computer program is stored, which, when executed by a processor, implements the above-described query and error correction method.

[0018] On the other hand, a computer program product is provided, including a computer program that, when executed by a processor, implements the above-described query and error correction method.

[0019] The beneficial effects of the technical solution provided in this application are:

[0020] By using a target model matched to the user, a first error indication is determined for the query information. This target model determines whether the information input to the model is erroneous based on the user's interest data, thus enabling personalized error judgment based on the user's interest data, making the judgment result more closely match the user's actual interest needs. When the query information is erroneous, at least one candidate information associated with the query information is determined based on target error correction data matched to the user. Subsequently, a second error indication is obtained based on the candidate information determined by the target model to further filter out corrective information. The target error correction data includes at least the user's interest data. This allows for personalized error correction of the query information based on the user's interests, ensuring that the corrected information more accurately matches the user's true query intent, meeting the error correction needs of each individual user, improving the actual accuracy of the error judgment and correction process, and thus improving the actual accuracy rate of query error correction. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below.

[0022] Figure 1 A schematic diagram illustrating the implementation environment of a query error correction method provided in this application embodiment;

[0023] Figure 2a A flowchart illustrating a query error correction method provided in an embodiment of this application;

[0024] Figure 2b A flowchart illustrating a query error correction method provided in an embodiment of this application;

[0025] Figure 3 This is a schematic diagram of the structure of a user group model provided in an embodiment of this application;

[0026] Figure 4 A flowchart illustrating a query and error correction process provided in this application embodiment;

[0027] Figure 5 This is a schematic diagram of the structure of a query error correction device provided in an embodiment of this application;

[0028] Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0029] The embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the embodiments described below with reference to the accompanying drawings are exemplary descriptions for explaining the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions of the embodiments of this application.

[0030] Those skilled in the art will understand that, unless otherwise stated, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the terms “comprising” and “including” as used in embodiments of this application mean that the corresponding feature can be implemented as the presented feature, information, data, step, operation, element, and / or component, but do not exclude implementation as other features, information, data, step, operation, element, component, and / or combinations thereof supported by the art. It should be understood that when we say that an element is “connected” or “coupled” to another element, the one element can be directly connected or coupled to the other element, or it can mean that the one element and the other element establish a connection relationship through an intermediate element. Furthermore, “connected” or “coupled” as used herein can include wireless connection or wireless coupling. The term “and / or” as used herein indicates at least one of the items defined by the term; for example, “A and / or B” indicates implementation as “A,” or implementation as “A,” or implementation as “A and B.”

[0031] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0032] The query error correction method provided in this application involves technologies such as cloud computing, cloud storage, and big data. For example, it can utilize computing devices included in a cloud computing resource pool to call a model matched to the user to determine errors in the user-input query information. Alternatively, it can utilize cloud storage technology to store error correction data for each user, user group models for each user group, and general models and error correction data for the target application; upon receiving a user's query request, it can call the model and the matching error correction data. For example, it can also employ a distributed database approach in big data to distribute and store large amounts of data such as interest data from numerous users and interest datasets from user groups.

[0033] Cloud computing is a computing model that distributes computing tasks across a large pool of computers, enabling various application systems to access computing power, storage space, and information services as needed. The network providing these resources is called the "cloud." From the user's perspective, resources in the "cloud" appear infinitely scalable, readily available, on-demand, and expandable, with payment based on usage.

[0034] As a provider of fundamental cloud computing capabilities, a cloud resource pool (referred to as a cloud platform, generally called an IaaS (Infrastructure as a Service) platform) is established. Various types of virtual resources are deployed in the resource pool for external customers to choose from. The cloud resource pool mainly includes: computing devices (virtualized machines containing operating systems), storage devices, and network devices.

[0035] Based on logical function, a PaaS (Platform as a Service) layer can be deployed on top of the IaaS (Infrastructure as a Service) layer, and a SaaS (Software as a Service) layer can be deployed on top of the PaaS layer. Alternatively, SaaS can be deployed directly on top of IaaS. PaaS is a platform for running software, such as databases and web containers. SaaS refers to various types of business software, such as web portals and bulk SMS senders. Generally speaking, SaaS and PaaS are upper layers compared to IaaS.

[0036] Cloud storage is a new concept that extends and develops from the concept of cloud computing. A distributed cloud storage system (hereinafter referred to as a storage system) refers to a storage system that uses cluster applications, grid technology and distributed storage file systems to bring together a large number of storage devices of various types in the network (storage devices are also called storage nodes) to work together through application software or application interfaces to provide data storage and business access functions to the outside world.

[0037] Big data refers to data sets that cannot be captured, managed, and processed within a certain timeframe using conventional software tools. It represents massive, rapidly growing, and diverse information assets that require new processing models to achieve stronger decision-making, insightful discovery, and process optimization capabilities. With the advent of the cloud era, big data has attracted increasing attention. Big data requires specialized technologies to effectively process large amounts of data within a tolerable timeframe. Technologies suitable for big data include massively parallel processing databases, data mining, distributed file systems, distributed databases, cloud computing platforms, the internet, and scalable storage systems.

[0038] Figure 1 This is a schematic diagram illustrating the implementation environment of a query error correction method provided in this application. For example... Figure 1As shown, the implementation environment includes a server 101 and a terminal 102. The server 101 can be a backend server for an application. The terminal 102 has an application installed, and the terminal 102 and the server 102 can interact with each other based on the application.

[0039] This application can be configured with a search function, which returns search results related to the user's input query information. In this application, the application can also be configured with a query correction function, which identifies and corrects errors in the user's input query information, so that the server 101 returns the user's expected search results based on the corrected information. In this application, query correction can be personalized based on the user's interest data, enabling the application's query correction function to meet the personalized needs of each user.

[0040] Personalized query correction refers to the fact that different users have different error types and correction goals. This application can realize personalized error correction of the query information entered by users, meet the personalized error correction needs of users with different interests, and improve the search user experience of video platforms.

[0041] In one possible scenario, terminal 102 can send user-input query information to server 101, where server 101 will determine and correct any errors in the query information, obtaining corrected information. Server 101 will then continue searching based on the corrected information. In another possible scenario, terminal 102 will obtain user-input query information, determine and correct any errors in the query information, and send the corrected information to server 101. Server 102 will then perform a search based on the corrected information.

[0042] Of course, the error determination and correction process for query information can also be implemented through interaction between server 101 and terminal 102. For example, in another scenario, terminal 102 determines the query information as incorrect and sends the query information and the result of the error determination to server 101, whereby server 101 corrects the query information based on the result of the error determination. Alternatively, both server 101 and terminal 102 can determine and correct the query information as incorrect; that is, terminal 102 sends the information after error determination and correction to server 101, and server 101 performs further error determination and correction on the information sent by terminal 102. Of course, in this embodiment, the query correction process and its various steps can be executed by any computer device such as a terminal or server, or interactively between multiple computer devices. This embodiment does not limit the specific device used for query correction.

[0043] This application can be applied to various scenarios, including but not limited to big data search, cloud technology, artificial intelligence, smart transportation, and assisted driving. The application can be any application that supports query correction, such as video applications, live streaming applications, content interaction applications, social applications, search engines, shopping applications, game applications, and enterprise communication and office tool platforms. For example, in a scenario where a user enters query information in the search box of a video application, this application can first correct the query information to return the user's desired video list based on the corrected information. The application can be a standalone application or a program plugin installed in a standalone application, such as a standalone video application or a video playback mini-program installed in a social application, or a web-based video website in a browser.

[0044] Server 101 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server or server cluster providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The aforementioned networks can include, but are not limited to, wired networks and wireless networks. Wired networks include local area networks (LANs), metropolitan area networks (MANs), and wide area networks (WANs). Wireless networks include Bluetooth, Wi-Fi, and other networks that enable wireless communication. Terminal 102 can be a mobile phone, a smart voice interaction device, a smart home appliance, an in-vehicle terminal (e.g., an in-vehicle navigation terminal, an in-vehicle computer), a tablet computer, a laptop computer, a digital radio receiver, a MID (Mobile Internet Device), a desktop computer, a smart speaker, a smartwatch, etc. Terminal 102 and server 101 can be directly or indirectly connected via wired or wireless communication, but are not limited to these methods. Furthermore, the number of terminals 102 and servers 101 can be one or more. This application embodiment does not limit the specific values ​​for the number of terminals and servers. The specific numbers can be determined based on the actual application scenario requirements, and are not limited here.

[0045] Figure 2a This is a flowchart illustrating a query error correction method provided in an embodiment of this application. The execution subject of this method can be a computer device, which can be a server or a terminal. As shown in Figure 2, the method includes the following steps.

[0046] Step 201: In response to receiving the query information from the user's input search box, the computer device determines the first error indication information of the query information based on the target model that matches the user.

[0047] The target model is used to determine whether the information input into the target model is erroneous based on the user's interest data in the target application. The computer device inputs the query information into the target model, and through the error determination process of the target model, outputs first error indication information for the query information. This first error indication information at least indicates whether the query information is erroneous. For example, different user maturity levels may require different target models; the computer device can be configured with a correlation between user maturity and error determination models; the computer device can use this correlation to select the error determination model corresponding to the user's maturity level as the target model. For example, the first error indication information at least includes the probability that the query information is correct.

[0048] Step 202: In response to the first error indication information indicating that the query information is incorrect, the computer device determines at least one candidate information associated with the query information based on the target error correction data matching the user and the query information.

[0049] The target error correction data includes at least the user's interest data in the target application. In this application, the computer device can search for similar information in the target error correction data based on query information, and select similar information with a similarity exceeding a target similarity threshold as at least one candidate information. Of course, this candidate information is found in the target error correction data; therefore, it also includes the user's interest data in the target application.

[0050] For example, the computer device may use at least one of the user's error correction data or the error correction data of a user group as the target error correction data. The error correction data of the user group includes the user group's interest dataset, and the user's error correction data includes the user's interest data in the target application. Alternatively, the target error correction data may also include general error correction data, which includes the global user interest dataset of the target application.

[0051] Step 203: The computer device determines the second error indication information of the at least one candidate information based on the target model.

[0052] The computer device inputs each candidate information into the target model and obtains a second error indication information for each candidate information, which includes the probability of the candidate information being correct.

[0053] Step 204: The computer device determines the correction information for the query information from the at least one candidate information based on the second error indication information.

[0054] The computer device can filter out corrective information that meets a specified probability condition from at least one candidate information. For example, the specified probability condition can be that the correct probability of the candidate information is greater than 0.8, or that the correct probability of the candidate information is greater than the correct probability of the query information.

[0055] By using a target model matched to the user, a first error indication is determined for the query information. This target model determines whether the information input to the model is erroneous based on the user's interest data, thus enabling personalized error judgment based on the user's interest data, making the judgment result more closely match the user's actual interest needs. When the query information is erroneous, at least one candidate information associated with the query information is determined based on target error correction data matched to the user. Subsequently, a second error indication is obtained based on the candidate information determined by the target model to further filter out corrective information. The target error correction data includes at least the user's interest data. This allows for personalized error correction of the query information based on the user's interests, ensuring that the corrected information more accurately matches the user's true query intent, meeting the error correction needs of each individual user, improving the actual accuracy of the error judgment and correction process, and thus improving the actual accuracy rate of query error correction.

[0056] Figure 2b This is a flowchart illustrating a query error correction method provided in an embodiment of this application. The executing entity of this method can be a computer device, which can be a server or a terminal. Figure 2b As shown, the method includes the following steps.

[0057] Step 301: The computer device receives the query information entered by the user into the search box.

[0058] Users can enter query information into the search box of the target application. If the computer device is a terminal, it retrieves the query information when it detects that the search box contains query information, or when it detects a search command triggered by the user. If the computer device is a server, the terminal sends the query information to the server, and the server receives the query information sent by the terminal.

[0059] The query information may include at least one of the following: a query string entered by the user, a string extracted from the query image, or a text description of an emoticon icon. For example, the user may enter a string in the search box. Alternatively, the user may enter an image, emoticon icon, etc., in the search box, and the computer device can obtain the corresponding string. Accordingly, the computer device may receive the query information in at least the four ways shown in Methods 1 to 4 below.

[0060] Method 1: The computer device receives the query string entered by the user in the search box of the target application and uses the query string as the query information.

[0061] The strings can include, but are not limited to, text, pinyin, symbols, numbers, punctuation marks, operators, etc. The target application's page may include a search box, which can include a string input plugin. Users can trigger this plugin to enter a query string into the search box, and the computer device receives the user's input query string.

[0062] Method 2: The computer device receives the query image entered by the user through the search box, extracts the string included in the query image, and uses the string extracted from the query image as the query information.

[0063] The search box can also include an image input plugin, which users can trigger to input a query image. The computer device can then extract strings from the query image to obtain the strings contained within it. For example, the query image could be a movie or TV show cover image containing the title; the computer device can extract the title from the cover image. Alternatively, the computer device can perform image recognition on the query image to identify the text information corresponding to the image content. For instance, based on a celebrity's image identified in the image, the device can further retrieve the celebrity's name.

[0064] Method 3: The computer device receives the emoji icon entered by the user through the search box, obtains the text description information of the emoji icon, and uses the text description information of the emoji icon as the query information.

[0065] The search box can also include an emoji input plugin. Users can trigger the emoji input plugin to input emojis, and the computer device can obtain the text description information corresponding to the user-input emoji from the association between the emoji and the text description information. For example, when a user triggers the emoji input plugin in the search box of a target application's page, multiple candidate emoji icons can be displayed on the page. The computer device can then retrieve the emoji selected by the user from these candidate icons. For instance, the candidate emoji icons could include emojis corresponding to various characters in multiple movies and TV shows; for example, a user could select the emoji icon corresponding to a female lead from a movie or TV show they are interested in to search for that emoji.

[0066] Method 4: The computer device receives the query string and emoticon icon entered by the user through the search box, obtains the text description information of the emoticon icon, and uses the query string and the text description information of the emoticon icon as the query information.

[0067] Users can input two types of information: strings and emoji icons. This query can include both the query string and the textual description of the emoji icon. The method by which the computer device obtains the textual description of the emoji icon is the same as in Method 3 described above, and will not be repeated here.

[0068] It should be noted that the computer device can combine any two or three of the above methods one to three to obtain query information. This application only uses the above method four, which combines user input strings and emoticons, as an example. Of course, it is also possible to combine user input characters and images to obtain query information, or combine user input characters, images, and emoticons to obtain query information, etc. The embodiments of this application do not limit the specific combination method.

[0069] In this step, by configuring the search box, the string entered by the user in the search box can be accurately obtained as query information, and subsequent query correction can be performed based on the string entered by the user, which improves the accuracy of obtaining query information; at the same time, the string extracted from the query image or the text description information of the emoticon icon can be used as query information, which supports query correction of information entered by the user under multiple search methods, and improves the applicability of the query correction method in this application.

[0070] Step 302: In response to receiving the query information from the user's input search box, the computer device determines the target model that matches the user based on the user's user maturity level.

[0071] The user maturity level indicates the user's activity level in the target application; the target model is used to determine whether the information input into the target model is incorrect based on the user's interest data in the target application. Different user maturity levels may require different target models. In this step, the computer device can determine the user's user maturity level and then determine the target model based on that level. The computer device can be configured with a correlation between user maturity and error detection models; based on this correlation, the computer device can use the error detection model corresponding to the user's user maturity level as the target model.

[0072] In one possible implementation, this step may include steps 3021-3022 (not shown in the figure).

[0073] Step 3021: The computer device determines the user's maturity level.

[0074] The computer device can determine user maturity based on factors such as the user's active time and interaction patterns within the target application. Specifically, it can determine user maturity based on at least one of the user's active time in the target application or the number of the user's interest tags. In one possible implementation, user maturity can be the maturity level of the user's accumulated interest profile within the user's active time in the target application. The process by which the computer device determines user maturity can include: determining the user's active time in the target application; counting the number of interest tags for the user within the active time based on the operation object association tags of the user's interactive operations; and determining user maturity based on the active time and the number of interest tags. For example, the number of interest tags can be the total number of multiple interest tags included in the user's interest profile. For example, the computer device can use the operation object association tags of the interactive operations as the user's interest tags and count the user's interest tags based on the number of interactive operations. For instance, for each newly added interest tag, the interest tag count is configured as 1, and for each additional interactive operation on the interest tag, the interest tag count is incremented by 1. For example, the active time may include the duration of user interaction on the target application. Interactions include, but are not limited to, playing videos, liking, sending bullet comments, posting comments, posting videos, browsing posts in the community, downloading videos, interacting with other users on the target application, browsing product details pages or making product transactions, and any other operations related to the target application.

[0075] The process by which the computer device determines the user's maturity based on the active time and the number of interest tags may include: when the active time does not exceed a target time threshold, the computer device determines the user's maturity as a first maturity level; when the active time exceeds the target time threshold, if the number of interest tags does not exceed a target tag number threshold, the computer device determines the user's maturity as a second maturity level; if the active time exceeds the target time threshold and the number of interest tags exceeds the target tag number threshold, the computer device determines the user's maturity as a third maturity level. For example, the target time threshold and the target tag number threshold can be configured as needed, and this application embodiment does not specifically limit them. For example, the target time threshold can be 5 days, 10 days, etc.; the target tag number threshold can be 300, 500, etc.

[0076] In one possible example, the computer device can be configured with three user maturity levels, arranged in ascending order of user maturity. These three user maturity levels can include new and inexperienced, user with immature interest profile, and user with mature interest profile. For example, the user's maturity level can be determined through the following processes (1)-(3):

[0077] (1) Determine if a user is a new cold user: whether the user's active time is less than T (e.g., T is 5 days). If it is less than T, the user is a new cold user.

[0078] (2) Determine if the user's interest profile is not mature enough: If it is not a new cold user, continue to determine if the number of user interest tags is less than or equal to K (e.g., K is 300). If it is less than or equal to K, then the user's interest profile is not mature enough.

[0079] (3) Determine whether the user is a mature user in the user interest profile: If the user is not a new cold user and the number of user interest tags is greater than K, then the user is a mature user in the user interest profile.

[0080] In one possible example, the associated tags of the operation object may include, but are not limited to: tags associated with the operation object, tags associated with the associated objects of the operation object, and interest tags configured by the user through tag configuration operations. For example, the associated objects of the operation object may be: a movie or TV series associated with a post published or replied to by the user, a certain creator, etc. For example, the target application may be a video application, which is configured with a query and error correction function. The user's interactive operations in the video application may include, but are not limited to: video playback operations, video interaction operations, post interaction operations in the community, interaction operations with creators in the video application, tag configuration operations, etc. For example, the computer device may count the number of interest tags through the following processes (1)-(3):

[0081] (1) When a user completes a valid playback operation on a video (playback completion rate is greater than a certain threshold; for example, playback duration exceeds 30 minutes) or an interactive operation on a video (like, forward, comment, bullet screen, share, etc.), the tags of the video are accumulated in the user's user interest profile; if there are no identical interest tags in the user interest profile, the number of interest tags for that interest tag is configured to 1; if there are already identical interest tags in the user interest profile, the number of interest tags for the identical interest tags in the user interest profile is incremented by 1.

[0082] (2) Accumulation of user interest in post interaction in the community: When a post or reply in the community is associated with a person, a movie or TV series, or a movie or TV series character, the associated tags such as actor, character, character tag, and feature description tag of the corresponding movie or TV series are accumulated into the user interest profile; if there is no same interest tag in the user interest profile, the number of interest tags for that interest tag is configured to 1; if there is already a same interest tag in the user interest profile, the number of interest tags for the same interest tag in the user interest profile is incremented by 1.

[0083] (3) When a user completes an interactive operation on a creator (following operation, liking operation, sharing the creator's work operation, sharing the creator's homepage operation, etc.), the tags of all the creator's videos are accumulated into the user's interest profile. If there are no identical interest tags in the user's interest profile, the number of interest tags for that interest tag is configured to 1; if there are already identical interest tags in the user's interest profile, the number of interest tags for the identical interest tags in the user's interest profile is incremented by 1.

[0084] In one possible example, the interest tags in a user's interest profile can also be decayed. For example, if the number of interest tags for a user's interest tag does not increase within a time period D (e.g., one week, one month, six months, etc.), the number of interest tags for that interest tag will be decayed. For example, a half-life decay method can be used, where the number of interest tags is reduced to half of its previous value every time time D passes. For example, if the original number of interest tags is 100 and time D is one month, the number of interest tags will decrease to 50 after the first month, to 25 after the second month, and so on.

[0085] Step 3022: When the user's maturity exceeds a maturity threshold, the computer device determines at least one target model that matches the user based on a user group with similar interests.

[0086] The target model includes at least a user group model corresponding to a user group with similar interests to the user. This user group model is used to determine whether the information input to the user group model is incorrect based on the user group's interest dataset. The user group includes at least two users with similar interests to the user. The user group's interest dataset includes the user's interest data in the target application. In one possible example, this step may include: when the user's maturity exceeds a maturity threshold, the computer device may determine the user group model corresponding to the user group with similar interests as the target model matching the user. In another possible example, the target model may include a user group model and a general model. The general model is used to determine whether the information input to the general model is incorrect based on the global user dataset of the target application. Accordingly, this step may include: when the user's maturity exceeds a maturity threshold, the computer device may also determine the general model and the user group model as the target model matching the user based on the user group with similar interests. For example, the maturity threshold may be a first maturity level, that is, when the user's maturity is a second or third maturity level, the user's maturity exceeds the maturity threshold.

[0087] It should be noted that this computer device can divide multiple users of a target application into multiple user groups based on user interest tags, with each user group comprising multiple users with similar interests. In one possible example, the computer device can cluster users at the third level of user maturity, for example, clustering mature users by interest similarity to construct user groups with similar interests. The interest similarity between two users is calculated as the cumulative count of duplicate interest tags in the two users' interest tag sets divided by the quotient of the total number of interest tags in both users' sets. For example, the KMeans (k-means clustering algorithm) clustering method can be used to cluster mature users to obtain multiple user groups with similar interests, and the cluster center for each user group. The cluster center for each user group can be the center point of that user group; for example, the cluster center can include the number of interest tags corresponding to each interest tag at the center point. For example, the goal of using the KMeans clustering method is to ensure that users within the same cluster have similar interests, while users in different clusters have significantly different interests. When using KMeans clustering, the target number of clusters can be C, where C can be the quotient of the number of mature users with interests divided by the number of standard users U. For example, U can be configured to 100,000.

[0088] By building user groups with similar interests, it's possible to assist users with immature interests in query correction. For example, by mapping immature users to user groups with mature interests, the richness of personalized data for immature users can be improved. Furthermore, it avoids the need for a separate error detection model for each user, thus avoiding the high training and storage costs associated with such models. By building error detection models for a user group, training and storage costs remain controllable while simultaneously meeting users' personalized error correction needs based on their interests.

[0089] For example, after clustering users at the third maturity level into multiple user groups, the user group to which a user with the third maturity level belongs is the user group whose interests are similar to that user. For users with the second maturity level, the similarity between the user and each user group can be calculated based on the cluster center of each user group and the number of each interest tag of the user at the second maturity level. The user group with the highest similarity is taken as the user group whose interests are similar to that user.

[0090] In one possible example, when the user's maturity does not exceed a maturity threshold, for example, for a new, cold user with a user maturity of level one, the computer device can determine a generic model as the target model that matches the user.

[0091] It should be noted that by using the aforementioned methods for accumulating user interest tags and determining user maturity, users can be categorized into different personalized levels based on multiple dimensions such as active time and the number of interest tags, thus improving the accuracy of user maturity classification. Furthermore, different error detection models are matched to users at different maturity levels, and different error correction data is used subsequently, facilitating personalized query error correction for different maturity levels and improving the accuracy of query error correction.

[0092] Step 303: The computer device determines the first error indication information for the query information based on the target model matching the user.

[0093] The computer device inputs the query information into the target model, and through the error determination process of the target model, outputs first error indication information for the query information. This first error indication information at least indicates whether the query information is incorrect. For example, the first error indication information includes at least the probability that the query information is correct.

[0094] The first error indication information is used to indicate whether the query information is incorrect and the type of error in the query information. In one possible implementation, the first error indication information includes the probability of the query information being correct and the probability of the error type, whereby the error type probability indicates the likelihood that the query information contains at least one type of error. Then, step 303 may include: the computer device extracting at least one descriptive information from the query information using the target model; the at least one descriptive information includes at least one of the following: the query information's pinyin, stroke count, positional encoding, keywords included in the query information, or phrases included in the query information; and the computer device, based on the at least one descriptive information of the query information, determining whether the query information is correct and obtaining the probability of the query information being correct and the probability of the error type.

[0095] Example 1: For cases where the target model includes either a user group model or a general model, for instance, when the user maturity level is second or third, the user group model can output the probability of the query being correct. When the user maturity level is first, the general model can output the probability of the query being correct.

[0096] Example 2: When the target model includes a user group model and a general model, the correct probability of the query information is the average of the two correct probabilities output by the two models respectively. For example, when the user maturity is the second maturity or the third maturity, the average of the first correct probability output by the user group model and the second correct probability output by the general model is determined as the correct probability of the query information.

[0097] It should be noted that the method of obtaining the correct probability based on the error type output by the target model is similar to the process of obtaining the correct probability shown in Examples 1 and 2 above. As in Example 1, one can obtain the error type probability output by the target model; or, as in Example 2, one can obtain the average between the first error type probability output by the user group model and the second error type probability output by the general model as the error type probability of the query information. These will not be elaborated on here.

[0098] The target model can be a BERT (Bidirectional Encoder Representations from Transformers) model, such as... Figure 3 As shown, the query information can be represented as a user query. The user query is input into the trained BERT model, and through the BERT model, multiple descriptive information such as word representation, character representation, character position encoding, character stroke representation, and character pinyin representation in the user query are obtained. Figure 3 In the CLS position, a placeholder is used to represent the contextual features of the entire sentence. The CLS position can be placed at the beginning of the sentence. The BERT model outputs deep features representing each position based on features obtained from multiple descriptive information. These features can be, for example, 768-dimensional feature vectors. Then, max pooling is performed on the feature vectors corresponding to the deep features at each position to obtain the maximum depth value for each position. The vectors representing the maximum depth values ​​at each position are concatenated with the contextual feature vectors output from the CLS position. This concatenated vector is then passed through a fully connected layer and the activation function of the output layer (e.g., a softmax normalized exponential function) to output the correct probability and the error type probability.

[0099] In one possible implementation, the training process of the user group model includes: a computer device acquiring historical query information of at least two users in the user group, and generating negative sample data including noise data by performing a noise generation operation on a portion of the historical query information; the computer device using the initial model to predict the negative sample data to obtain predicted sample data; the computer device adjusting the model parameters of the initial model based on the similarity between the predicted sample data and the positive sample data of the historical query information, until the initial model meets the model training conditions and the adjustment stops, thus obtaining the user group model. The noise generation operation includes at least one of word fragment replacement, word fragment deletion, or word fragment addition; the word fragment replacement operation includes at least one of replacing with pinyin-similar word fragments, replacing with character-similar word fragments, or replacing with semantically similar word fragments.

[0100] For example, for any user group model, a set of historical query information (e.g., query data) previously entered by multiple users in that user group can be obtained, and a portion of the input historical query information set can be randomly selected. Noise generation is performed on this selected portion of the historical query information set to automatically construct negative sample data including the noise data. For example, noise generation can be performed in the following ways (1)-(3):

[0101] (1) Noise data with substitution errors is generated by replacing word fragments, and words in the query data are randomly replaced with other words, including the following four methods (a)-(d):

[0102] (a) Randomly replace words in certain positions with alternative words;

[0103] (b) Randomly replace words in certain positions with other words that have similar shapes;

[0104] (c) Randomly replace words in certain positions with other words that sound similar;

[0105] (d) Randomly replace words in certain positions with other words that have similar content.

[0106] (2) Noise data with missing errors is generated by deleting word fragments, and words in certain positions in the Query data are randomly deleted.

[0107] (3) Generate redundant and erroneous noise data by adding word fragment operations, and randomly add some other words in certain positions in the Query data.

[0108] In this context, the random noise data generated by the methods described in (1)-(3) above is used as the noise data included in the negative sample query data of the error class. The original form of the query data can be manually verified to serve as the positive sample query data of the correct class. Then, the initial error judgment model is trained based on the positive and negative samples. In one possible example, the computer device can predict the correct form of each query data in the negative sample. For example, it can predict the correct data corresponding to each query data in the negative sample to obtain the predicted sample data. Based on the similarity between the predicted sample data and the positive sample data, the initial error judgment model is trained. After the initial error judgment model training converges, the user group model is obtained. In another possible example, the computer device can also predict the sample indication information of each query data in the negative sample. The sample indication information can include the sample correctness probability, the sample error type probability, and the truth value label of each query data obtained during the noise generation process of the negative sample. The truth value label can include the truth value of whether each query data in the negative sample is wrong and the error type. Based on the similarity between the sample indication information and the truth value label, the initial error judgment model is trained. After the initial error judgment model training converges, the user group model is obtained.

[0109] For example, for a general model, a global historical query information set (e.g., query data) previously input by all users on the target application can be obtained. A portion of this historical query information set is randomly selected, and noise is generated on this selected portion to automatically construct negative sample data including the noise data. Positive sample data can be a global historical query information set that has undergone some manual verification. The method for generating negative sample data using the global historical query information set is similar to the method for generating negative sample data corresponding to the user group model described above, and will not be elaborated further here. The computer device can predict negative samples using an initial general model. Based on the similarity between the predicted sample data and the positive sample data, the initial general model is trained. After the initial general model training converges, a general model is obtained. The training method for the general model is similar to the training method for the user group model described above, and will not be elaborated further here.

[0110] In this step, a portion of the historical query data input by users in the user group is randomly selected and noise is generated in this portion of the data to automatically construct query error data. Based on positive and negative samples, a user group model is trained, which facilitates the subsequent personalized error judgment for users with different interests based on the user group model, thereby improving the accuracy of error judgment and further improving the accuracy of query error correction.

[0111] Furthermore, when the first error indication information indicates that the query information is incorrect, the query information is corrected based on the following step 304. When the first error indication information indicates that the query information is not incorrect, the subsequent search can be performed directly based on the query information to return the search results, and the subsequent step 304 does not need to be executed.

[0112] Step 304: In response to the first error indication information indicating that the query information is incorrect, the computer device determines the target error correction data that matches the user based on the user's maturity level.

[0113] The target error correction data includes at least the user's interest data in the target application; for example, the computer device may use at least one of the user's error correction data or the error correction data of a user group as the target error correction data. The error correction data of the user group includes the user group's interest dataset, and the user's error correction data includes the user's interest data in the target application.

[0114] In one possible implementation, the target error correction data may further include general error correction data, which includes the global user interest dataset of the target application; based on several possible scenarios of the target error correction data, the implementation of this step may include the following four methods.

[0115] In the first method, the computer device determines the error correction data of the user group as the target error correction data that matches the user, based on the user group of the user.

[0116] In one possible example, when a user's maturity level is second maturity, the error correction data of a user group with similar interests can be used as the target error correction data for that user. The error correction data for that user group can include the user group's interest dataset for the target application.

[0117] The user group's interest dataset may include the user group's behavioral data within the target application. In one possible example, the user group's behavioral data within the target application includes, but is not limited to: video titles and descriptions of creators that the user group has effectively watched or followed; video comments and bullet comments from the user group; comments on posts or replies made by the user group in the community; and the user group's historical query information. It should be noted that behavioral data can be configured as needed; this application only uses the above-described data examples for illustration. Of course, behavioral data may also include the names of products collected or purchased by users, the titles of articles collected or forwarded, the titles of live streams watched, and the titles of downloaded videos, etc. This application embodiment does not limit the data that may be included in the behavioral data.

[0118] In one possible implementation, the target error correction data further includes a query index for the interest data, which is used to index similar segments within the interest data. The computer device can generate error correction data for a user group based on its interest dataset. This process can include: the computer device performing word segmentation on the user group's interest dataset; generating a query index for indexing similar segments based on the similarity between multiple segments obtained from the word segmentation; and generating the error correction data for the user group based on the multiple segments and the query index of similar segments. In one example, the computer device performs word segmentation on the interest dataset and can filter out segments in the segmentation results that appear more frequently than a target frequency threshold. A query index is generated based on the similarity between the filtered segments. The target frequency threshold can be configured as needed, and this embodiment does not limit it; for example, the target frequency threshold can be 5 times, 3 times, 10 times, etc.

[0119] In one possible example, error correction data can be constructed based on the pinyin similarity between multiple segments. In this case, the query index includes a pinyin query index, which is used to index word segments with similar pinyin in the interest data. In another possible example, error correction data can also be constructed based on the glyph similarity between multiple segments. In this case, the query index includes a stroke query index, which is used to index word segments with similar strokes in the interest data. In yet another possible example, error correction data can also be constructed based on the semantic similarity between multiple segments. In this case, the query index includes a semantic query index, which is used to index word segments with semantically similar keywords in the interest data. Accordingly, (1)-(3) below respectively show the methods of constructing error correction data based on pinyin similarity, glyph similarity, or semantic similarity.

[0120] (1) Constructing error correction data based on pinyin similarity: The interest dataset is segmented into words, and a user group vocabulary is constructed based on multiple word segments whose frequency exceeds the target frequency threshold. The word segments in the user vocabulary are annotated with pinyin. Based on the pinyin of multiple word segments, pinyin-similar word segments are obtained from multiple word segments. The pinyin is used as the pinyin index of the pinyin-similar word segments to obtain error correction data for the user group that includes multiple word segments and pinyin index. For example, the pinyin of a word can be used as the query index for similarity retrieval in Elasticsearch (used for distributed full-text search). Then, by inputting the pinyin of a word, one or more similar words with the pinyin of that word can be obtained, as well as the pinyin similarity between the word and the similar words.

[0121] (2) Construct error correction data based on glyph similarity: Decompose the word segments in the user group word list that has been constructed into strokes; and based on the stroke sequences of multiple word segments, obtain the glyph-similar word segments among the multiple word segments, and use the stroke sequence as the glyph index of the glyph-similar word segment, so as to obtain the error correction data of the user group including multiple word segments and glyph indexes. For example, use the stroke sequence of the entry as the query index for similarity retrieval in Elasticsearch; then input the stroke sequence of a certain entry, and one or more similar entries that are glyph-similar to the entry can be obtained, and the glyph similarity between the entry and the similar entries can also be obtained.

[0122] (3) Construct error correction data based on semantic similarity: Split the word segments in the user group word list that has been constructed at the character granularity, and based on the split keywords and the similar keywords that are semantically similar to the keywords, determine multiple semantically similar word segments including the similar keywords; and construct a semantic query index for the multiple semantically similar word segments. For example, use the semantically similar word segment as the semantic query index; for example, semantic similarity can be that the character content of the keywords included in the word segment is similar, and it can be realized that when inputting a certain query entry, other entries with similar character content of the entry are returned. For example, when inputting "The whole city is covered with golden armor", entries with similar character content such as "All over the city are clad in golden armor" can be queried according to character similarity; and the character content similarity between the entry and the similar entries can also be obtained.

[0123] Among them, the methods of constructing the user group word list in (2) and (3) above are the same as those in (1), and will not be elaborated one by one in (2) and (3).

[0124] The second method: The computer device determines the error correction data of the user and the error correction data of the user group as the target error correction data based on the user group of the user.

[0125] In a possible example, when the user maturity is the third maturity level, the error correction data of the user group that is similar to the user's interest can be used as the target error correction data of the user. The error correction data of the user can include the interest data of the user in the target application.

[0126] The user's error correction data may include a query index. The computer device can also obtain error correction data including multiple segments and a query index by segmenting the user's interest data. The user's interest data may also include the user's behavioral data within the target application, which can include examples of behavioral data shown in the first approach. Furthermore, the query index may include at least one of a pinyin index, a character shape index, or a semantic index. The method of constructing the user's error correction data based on the user's interest data, and specifically the method of constructing the user's error correction data based on pinyin similarity, character shape similarity, or semantic similarity, is similar to the process of constructing the user group's error correction data in the first approach described above. The specific implementation methods for constructing the user's error correction data will not be elaborated upon here.

[0127] The third method involves the computer device determining the target error correction data based on the user's user group and general error correction data.

[0128] In one possible example, when a user's maturity level is second maturity, the general error correction number and the error correction data of a user group with similar interests can be used as the user's target error correction data.

[0129] The general error correction data includes at least the global user interest dataset of the target application, that is, the interest data of all users in the target application. Of course, the global user error correction data may also include a query index. The computer device can also obtain error correction data including multiple segments and a query index by segmenting the global user interest dataset. The global user interest dataset may also include the behavioral data of all users in the target application; the query index may include at least one of a pinyin index, a character shape index, or a semantic index. The method of constructing general error correction data based on the global user interest dataset, and the specific methods of constructing general error correction data based on pinyin similarity, character shape similarity, or semantic similarity, are similar to the process of constructing error correction data for the user group in the first method described above. The specific implementation methods for constructing general error correction data will not be elaborated here.

[0130] The fourth method involves the computer device determining the target error correction data based on the user's user group, the user's error correction data, the user group's error correction data, and general error correction data.

[0131] In one possible example, when a user's maturity level is third, the target error correction data for that user can be the general error correction count, the user's error correction data, and the error correction data of a user group with similar interests. The methods for constructing the user's error correction data, the general error correction data, and the user group's error correction data are the same as the three methods mentioned above, and will not be elaborated upon here.

[0132] In this step, when the query information is incorrect, target error correction data matching the user is obtained based on the user's maturity level. This target error correction data includes the user's interest data, which facilitates subsequent correction of the query information based on the user's interest direction, meeting the personalized error correction needs of users with different interests, further improving the actual error correction accuracy of the query and enhancing the user search experience.

[0133] Step 305: The computer device determines at least one candidate information associated with the query information based on the target error correction data matching the user and the query information.

[0134] The target error correction data includes at least the user's interest data in the target application. In this application, the computer device can, based on query information, search for similar information in the target error correction data that is similar to the query information, and use similar information with a similarity exceeding a target similarity threshold as at least one candidate information. Of course, this candidate information is information searched from the target error correction data; therefore, the candidate information also includes the user's interest data in the target application.

[0135] In one possible implementation, the query information includes a query string, which can be used to find related word fragments based on the word fragments included in the string, in order to construct candidate information based on the related word fragments. Accordingly, this step may include: the computer device obtaining at least one related word fragment corresponding to the at least one word fragment included in the query string from the target error correction data; the computer device replacing the at least one word fragment corresponding to the at least one related word fragment in the query information based on the at least one related word fragment to obtain the at least one candidate information. For example, for the first word fragment "AB" in the string "ABXXX", the related word fragments associated with "AB" in the target error correction data include "CD" and "EF". After replacing "AB", the candidate information "CDXXX" and "EFXXX" for "ABXXX" are obtained; where "AB" can represent various forms such as English words, Chinese phrases, one or more Chinese characters, symbols, etc.

[0136] In a possible implementation, the target error correction data further includes a query index of the interest data. For each word segment included in the query string, the computer device queries at least one associated word segment similar to the word segment from the target error correction data based on the query index. The associated word segment may include at least one of a pinyin similar word segment, a glyph similar word segment, or a semantic similar word segment. Based on the pinyin query index, glyph query index, and voice query index shown in step 304 above, in this step, for each word segment, the computer device can obtain the associated word segment of the word segment through at least one of the following methods 1 to 3.

[0137] Method 1: The query index includes a pinyin query index. The computer device queries at least one pinyin similar word segment of the word segment from the target error correction data based on the pinyin of the word segment and the pinyin query index.

[0138] The computer device can extract the pinyin of the word segment and query at least one pinyin similar word segment indexed by the pinyin from the target error correction data. For example, for a certain word segment "优惠卷" (coupon) in the user input information, the corresponding pinyin "you hui juan" can be input, and the pinyin similar word segment "优惠券" (coupon) queried based on the pinyin index is output.

[0139] Method 2: The query index includes a stroke query index. The computer device queries at least one glyph similar word segment of the word segment from the target error correction data based on the stroke sequence of the word segment and the stroke query index.

[0140] The computer device can extract the stroke sequence of the word segment and query at least one glyph similar word segment indexed by the stroke sequence from the target error correction data. For example, for a certain word segment "钩鱼" (fishing) in the user input information, the corresponding stroke sequence can be input, and the glyph similar word segment "钓鱼" (fishing) queried based on the stroke sequence index is output.

[0141] Method 3: The query index includes a semantic query index. The computer device queries at least one semantic similar word segment of the word segment from the target error correction data based on the word segment and the semantic query index.

[0142] The computer device can query at least one semantic similar word segment indexed by the word segment from the target error correction data according to the word segment and the semantic query index. For example, when inputting "满城全带黄金甲" (All over the city, wearing golden armor), the entry "满城尽带黄金甲" (All over the city, wearing golden armor) with similar characters can be queried; for example, when inputting "忽然" (suddenly), the semantic similar word segments "突然" (suddenly), "猛然" (suddenly) queried based on "忽然" are output.

[0143] It should be noted that, corresponding to the four methods of obtaining the target error correction data shown in step 304, this step can query the at least one related word fragment from each type of error correction data included in the target error correction data, based on one or more error correction data included in the target error correction data. Taking the target error correction data including error correction data of user groups and error correction data of users as an example, for method one in step 305, the process of the computer device obtaining at least one pinyin similar word fragment may include: the computer device may query at least one first pinyin similar word fragment of the word fragment from the error correction data of user groups based on the pinyin index of the word fragment and the error correction data of user groups, respectively; and, based on the pinyin index of the word fragment and the error correction data of users, query at least one second pinyin similar word fragment of the word fragment from the error correction data of users. Of course, when the target error correction data includes one type of error correction data, three types of error correction data, or other cases of two types of error correction data, and for cases such as character shape similar word fragments or semantic similar word fragments, the method of obtaining related word fragments is the same as the example of obtaining related word fragments based on two types of error correction data, and will not be listed and elaborated here.

[0144] It should be noted that for each word fragment in the query information, similar word fragments (such as the three types of similar word fragments in methods one to three above) can be used to replace that word fragment in the query information to construct a candidate information. For example, if one word fragment in the query information is replaced, the resulting candidate information will include one similar word fragment. Alternatively, two or more word fragments in the query information can be replaced simultaneously, resulting in a candidate information that includes two or more similar word fragments. In this embodiment, candidate information can be constructed by replacing one or more word fragments at a time. The above description only uses replacing one word fragment at a time as an example; this application does not limit the specific method of constructing candidate information.

[0145] Step 306: The computer device determines the second error indication information of the at least one candidate information based on the target model.

[0146] The computer device inputs each candidate information into the target model to obtain a second error indication for each candidate information, which includes the correct probability of the candidate information. Similar to the process of obtaining the correct probability of the query information in step 303, this step can obtain the correct probability of the candidate information output by the user group model based on the situation in Example 1 of Step 303; or, obtain the correct probability of the candidate information output by the general model. Alternatively, based on the situation in Example 2 of Step 303, the average of the third correct probability output by the user group model and the fourth correct probability output by the general model can be obtained as the correct probability of the candidate information. The specific implementation of obtaining the correct probability of the candidate information is similar to the process in Example 1 or Example 2 of Step 303, and will not be elaborated here.

[0147] In steps 305-306, multiple candidate information is constructed based on similarity to the query information, increasing the likelihood of the candidate information matching the correct information. Furthermore, candidate information can be constructed using various methods such as pinyin query index, character shape query index, or term query index, enriching the sources of candidate information. This makes subsequent correction of the user's query information more accurate, reducing the cost of deletion and re-entry when the user enters incorrect query information, and improving the actual search accuracy and efficiency of subsequent searches.

[0148] Step 307: The computer device determines the correction information for the query information from the at least one candidate information based on the second error indication information.

[0149] The computer device can filter out corrective information that meets a specified probability condition from at least one candidate information. For example, the specified probability condition can be that the correct probability of the candidate information is greater than 0.8, or that the correct probability of the candidate information is greater than the correct probability of the query information.

[0150] In one possible implementation, the computer device may first correct the correctness probability of the candidate information based on the error type probability, and then determine the correction information based on the corrected correctness probability. Accordingly, this step can be implemented through the following steps 3071-3072 (not shown in the figure).

[0151] Step 3071: The computer device corrects the correct probability of at least one candidate information based on the error type probability of the query information, and obtains the correct score of the at least one candidate information.

[0152] In one possible implementation, the candidate information includes related word fragments that replace word fragments in the query information. The computer device can, based on the similarity type between the related word fragment and the replaced word fragment in the query information, obtain the type probability of a matching error type that matches the similarity type from the probabilities of at least one error type included in the error type probability. The computer device determines the correct score of the candidate information based on the correct probability of the candidate information and the type probability. For example, the similarity type includes at least one of: phonetic similarity, character shape similarity, and semantic similarity. The phonetic similarity matching error type includes phonetic similarity errors, the character shape similarity matching error type includes character shape similarity errors, and the semantic similarity matching error type includes at least one of character content similarity errors, missing word errors, or redundant word errors. Specifically, if the number of characters in the related word fragment is greater than the number of characters in the replaced word fragment at the original replacement position, it is a missing word error; if the number of characters in the related word fragment is less than the number of characters in the replaced word fragment at the original replacement position, it is a redundant word error. The similarity type between the related word fragment and the replaced word fragment in the query information indicates the source of the related word fragment. For example, if the related word fragment is a pinyin similar word fragment obtained based on the pinyin query index, then the similarity type between the corresponding related word fragment and the replaced word fragment is phonetic similarity; if the related word fragment is a character shape similar word fragment, the corresponding similarity type is character shape similarity; if the related word fragment is a semantic similar word fragment, the object similarity type is semantic similarity.

[0153] In one possible example, the computer device can determine the correct score of the candidate information as the product of the type probability and the correct probability. For example, if the correct probability of the candidate information is P3 based on the target model, the error type probability of the query information is P2, and the type probability of matching the error type is extracted from P2, then the correct score P4 can be expressed as: P4 = P3 × P2[matching error type corresponding to the candidate information], where P2[matching error type corresponding to the candidate information] represents the type probability of matching the error type in P2.

[0154] The candidate information may include related word fragments and the replaced word fragments, which may have one or more similarity types. When multiple similarity types exist, at least two of the pinyin similar word fragments, character shape similar word fragments, or semantic similar word fragments of the replaced word fragment overlap. For example, both the pinyin similar word fragments and character shape similar word fragments of the replaced word fragment are related word fragments M; for example, the pinyin similar word fragments and character shape similar word fragments of the query information "discount coupon" both include "coupon".

[0155] In one possible example, the candidate information includes a related word fragment. The pinyin-similar word fragments, character-similar word fragments, and semantic-similar word fragments of the replaced word fragment do not overlap. The candidate information corresponds to a matching error type, so the probability of the matching error type for the query information is directly obtained. The related word fragment is one of the pinyin-similar word fragments, character-similar word fragments, or semantic-similar word fragments of the replaced word fragment. For example, for the query information "ABXXX" and its corresponding candidate information "CDXXX", the source of "CD" only includes pinyin-similar word fragments obtained based on the pinyin query index, that is, "CD" is a pinyin-similar word fragment of the first word fragment "AB". Therefore, the probability of a character-similar error is obtained from the error type probability of the query information "ABXXX".

[0156] In another possible example, a candidate message includes a related word fragment. This related word fragment and the replaced word fragment share at least two similarity types, meaning that at least two of the replaced word fragment's pinyin similarity, character shape similarity, or semantic similarity overlap. The computer device then determines at least two similarity levels corresponding to these at least two similarity types based on these at least two similarity levels, and obtains the type probability of a matching error type that matches the target similarity type based on the target similarity type that meets a threshold condition. The threshold condition may include, but is not limited to: similarity higher than a certain threshold, maximum similarity, and similarity higher than a minimum threshold but not exceeding a maximum threshold. For example, a similarity higher than a certain threshold of 60%.

[0157] For example, for the query information “ABXXX” and its corresponding candidate information “JKXXX”, “JK” is the pinyin similarity segment and the character shape similarity segment of the first word segment “AB”. Among them, the pinyin similarity of “JK” and “AB” is 91% and the character shape similarity is 80%. Based on the pinyin similarity of 91%, the probability of the type of pinyin similarity error matching the pinyin similarity is obtained.

[0158] The two examples above illustrate the case where the candidate information includes only one related word fragment. Of course, a candidate information can also include at least two related word fragments, and these at least two related word fragments share at least one or both similarity types with the replaced word fragment. The type probability can then be determined by combining the at least one or both similarity types between these at least two related word fragments and the replaced word fragment. In one possible example, the type probability P0 can be calculated based on Formula 1:

[0159] Formula 1: P0 = ;

[0160] Where P0 is the type probability; i is the i-th associated word fragment among the at least two associated word fragments of the candidate information; x i p represents the similarity score (e.g., pinyin similarity, character shape similarity, or character content similarity) between the i-th related word fragment and the replaced word fragment, corresponding to the similarity type. i The type probability of a matching error type when matching the similarity type corresponding to the i-th associated word fragment.

[0161] In another possible implementation, a correction factor can be used to calculate the correct score, since each user's error habits are different; for example, a user may frequently make certain fixed types of errors. Therefore, a personalized target association can be configured for each user based on their error habits, and this target association includes the correspondence between error types and correction factors. This step may further include: the computer device can, based on the matched error type, obtain the correction factor corresponding to the matched error type from the target association matching the user, and determine the product of the type probability, the correction factor, and the correct probability as the correct score for the candidate information.

[0162] Step 3072: The computer device, based on the correct score of the at least one candidate information and the first error indication information, filters out correction information whose correct score meets the target conditions from the at least one candidate information.

[0163] The target condition may include, but is not limited to: the correct score being greater than the correct probability of the query information, the correct score exceeding the target score threshold, the maximum correct score, and the correct probability corresponding to the error type probability of the correct score being greater than the query information (i.e., (1-P2)). The computer device can use candidate information whose final correct score meets the target condition as correction information, and then perform subsequent searches based on the correction information.

[0164] To more clearly illustrate the execution flow of the embodiments of this application, the following will use... Figure 4 Let's take an example to illustrate. Figure 4 This is a flowchart of a query and error correction method provided in an embodiment of this application, such as... Figure 4As shown, taking a video application as an example, when a user searches on a video platform, the user enters a query. The computer device performs personalized error judgment on the user's query, for example, using a user group model to determine the probability of the query being correct and the probability of the error type. When the user's query contains errors, personalized error correction is performed on the user's query, and the corrected user query is returned for searching. Personalized data can be used to build the user group model, for example, by training the model using the user's historical query information. During personalized error correction, for both immature and mature users, at least one of the personalized error correction data from the user group or the user's personalized error correction data can be used for correction; of course, general error correction data can also be combined. Additionally, for users with a "new" or "cold" maturity level, general error correction data can be used for correction.

[0165] It should be noted that this application improves the ability of target applications such as video platforms, search engines, social applications, and live streaming applications to meet users' personalized query correction needs by constructing personalized query correction models, correction data, and correction strategies for users with different interests. This allows for more intelligent and accurate correction of user-input query information, further reducing the cost of re-entering incorrect query information when retrieving videos, articles, products, and other information, and improving the usability of the target application's search function. Especially for video platforms, with their large user base and diverse interests among different users or user groups, there are significant differences in their favorite celebrities, preferred video types, and topics of discussion. Users' search behavior on video platforms is a subjective means of discovering video-related information, inevitably influenced by their subjective interests. However, some existing technologies for correcting user video search queries fail to fully leverage the personalized interests of video platform users, resulting in an inability to handle error correction issues individually for users with different interests. The personalized video search query error correction method proposed in this application, based on the interests of video platform users, constructs personalized error judgment models and correction data for users with different interests. This enables the video platform's query error correction function to meet the personalized needs of each individual user, more accurately judging and correcting errors in user-inputted video search queries, greatly reducing the need for deletion and re-entry when users incorrectly input video search queries, improving the efficiency of personalized searches and enhancing the actual accuracy of subsequent searches.

[0166] Step 307 above enables personalized error detection and correction of query information for users with different personalization levels, improving the user experience of video search. Furthermore, by performing similarity type matching on related words and replaced words based on error type probability, the correctness probability of candidate information can be targeted and personalized based on the type probability of the matched error type, improving the accuracy of the corrected information and further enhancing query accuracy.

[0167] By using a target model matched to the user, a first error indication is determined for the query information. This target model determines whether the information input to the model is erroneous based on the user's interest data, thus enabling personalized error judgment based on the user's interest data, making the judgment result more closely match the user's actual interest needs. When the query information is erroneous, at least one candidate information associated with the query information is determined based on target error correction data matched to the user. Subsequently, a second error indication is obtained based on the candidate information determined by the target model to further filter out corrective information. The target error correction data includes at least the user's interest data. This allows for personalized error correction of the query information based on the user's interests, ensuring that the corrected information more accurately matches the user's true query intent, meeting the error correction needs of each individual user, improving the actual accuracy of the error judgment and correction process, and thus improving the actual accuracy rate of query error correction.

[0168] Figure 5 This is a schematic diagram of a query error correction device provided in an embodiment of this application. Figure 5 As shown, the device includes:

[0169] The first determining module 501 is used to respond to receiving query information from a user input search box, and determine first error indication information for the query information based on a target model that matches the user. The target model is used to determine whether the information input into the target model is incorrect based on the user's interest data in the target application.

[0170] The candidate determination module 502 is configured to, in response to the first error indication information indicating that the query information is incorrect, determine at least one candidate information associated with the query information based on the target error correction data matching the user and the query information, wherein the target error correction data includes at least the user's interest data in the target application;

[0171] The second determining module 503 is used to determine the second error indication information of the at least one candidate information based on the target model;

[0172] The correction module 504 is used to determine correction information for the query information from the at least one candidate information based on the second error indication information.

[0173] In one possible implementation, the candidate determination module 502 includes:

[0174] The related word acquisition unit is used to acquire at least one related word fragment corresponding to the at least one word fragment included in the query string from the target error correction data;

[0175] The candidate acquisition unit is used to replace at least one corresponding word fragment in the query information based on the at least one related word fragment to obtain the at least one candidate information.

[0176] In one possible implementation, the target error correction data also includes a query index of the interest data, which is used to index similar segments in the interest data;

[0177] This related term acquisition unit is used for at least one of the following:

[0178] For each word segment, based on the pinyin of the word segment and the pinyin query index, at least one pinyin-similar word segment of the word segment is queried from the target error correction data. The query index includes a pinyin query index, which is used to index pinyin-similar word segments in the interest data.

[0179] Based on the stroke sequence and stroke query index of the word fragment, at least one word fragment with similar character shape is queried from the target error correction data. The query index includes a stroke query index, which is used to index word fragments with similar stroke shape in the interest data.

[0180] Based on the word fragment and the semantic query index, at least one semantically similar word fragment is queried from the target error correction data. The query index includes a semantic query index, which is used to index word fragments with semantically similar keywords in the interest data.

[0181] In one possible implementation, the first error indication information includes the correctness probability of the query information and the error type probability, wherein the error type probability is used to indicate the possibility that the query information contains an error of at least one type; the second error indication information includes the correctness probability of the candidate information.

[0182] The correction module 504 includes:

[0183] The correction unit is used to correct the correct probability of at least one candidate information based on the error type probability of the query information, so as to obtain the correct score of at least one candidate information.

[0184] A filtering unit is used to filter out correction information whose correct scores meet the target conditions from the at least one candidate information based on the correct scores of the at least one candidate information and the first error indication information.

[0185] In one possible implementation, the candidate information includes related word fragments that replace word fragments in the query information;

[0186] The correction unit is used to obtain the type probability of the matching error type that matches the similarity type from the probability of at least one error type included in the error type probability, based on the similarity type between the related word fragment and the replaced word fragment in the query information; and to determine the correct score of the candidate information based on the correct probability of the candidate information and the type probability.

[0187] In one possible implementation, the similarity type includes at least one of: phonetic similarity, character shape similarity, and semantic similarity;

[0188] Among them, the matching error type of phonetic similarity includes phonetic similarity error, the matching error type of character shape similarity includes character shape similarity error, and the matching error type of semantic similarity includes at least one of character content similarity error, missing error, or redundancy error.

[0189] In one possible implementation, the correction unit is used for any of the following:

[0190] The product of the type probability and the correct probability is determined as the correct score for the candidate information;

[0191] Based on the matching error type, the correction factor corresponding to the matching error type is obtained from the target association relationship matching the user. The product value between the type probability, the correction factor and the correct probability is determined as the correct score of the candidate information. The target association relationship includes the correspondence between the error type and the correction factor.

[0192] In one possible implementation, the first error indication information includes the correctness probability and the error type probability of the query information; the first determining module is used to extract at least one descriptive information of the query information through the target model, the at least one descriptive information including at least one of the following: the pinyin, strokes, position code, keywords included in the query information, or phrases included in the query information; based on the at least one descriptive information of the query information, the query information is judged to be correct, and the correctness probability and the error type probability of the query information are obtained.

[0193] In one possible implementation, the target model includes a user group model, which is used to determine whether the information input to the user group model is incorrect based on the user group's interest dataset, wherein the user group includes at least two users with similar interests to the user.

[0194] The device also includes a model training module, which is used to acquire historical query information of at least two users in the user group, and generate negative sample data including noise data by performing a noise generation operation on some information in the historical query information; predict the negative sample data through the initial model to obtain predicted sample data; and adjust the model parameters of the initial model based on the similarity between the predicted sample data and the positive sample data of the historical query information until the initial model meets the model training conditions and the adjustment stops, thus obtaining the user group model.

[0195] The noise generation operation includes at least one of the following: replacing word fragments, deleting word fragments, or adding word fragments. The replacing word fragment operation includes at least one of the following: replacing with pinyin-similar word fragments, replacing with character-similar word fragments, or replacing with semantically similar word fragments.

[0196] In one possible implementation, the device further includes:

[0197] The user maturity determination module is used to determine the user maturity level in response to receiving query information from the user's input search box. The user maturity level is used to indicate the user's level of activity in the target application.

[0198] The target model determination module is used to determine at least one target model that matches the user based on a user group with similar interests when the user's maturity exceeds a maturity threshold.

[0199] In one possible implementation, the target model determination module is used for at least one of the following:

[0200] Based on a user group with similar interests to the user, the user group model corresponding to the user group is determined as the target model that matches the user.

[0201] Based on a user group with similar interests, the general model and the user group model are identified as the target models that match the user.

[0202] The general model is used to determine whether the information input to the general model is incorrect based on the global user dataset of the target application.

[0203] In one possible implementation, the user maturity determination module is used to determine the user's active time in the target application; based on the operation object association tags of the user's interactive operations, count the number of interest tags of the user during the active time; and determine the user maturity based on the active time and the number of interest tags.

[0204] In one possible implementation, the device further includes at least one of the following:

[0205] The first query module is used to receive the query string entered by the user in the search box of the target application and use the query string as the query information.

[0206] The second query module is used to receive the query image entered by the user through the search box, extract the string included in the query image, and use the string extracted from the query image as the query information.

[0207] The third query module is used to receive the emoji icon entered by the user through the search box, obtain the text description information of the emoji icon, and use the text description information of the emoji icon as the query information;

[0208] The fourth query module is used to receive the query string and emoji icon entered by the user through the search box, obtain the text description information of the emoji icon, and use the query string and the text description information of the emoji icon as the query information.

[0209] In one possible implementation, the device further includes an error correction data determination module, used for any of the following:

[0210] Based on the user's user group, the error correction data of the user group is determined as the target error correction data that matches the user. The error correction data of the user group includes the interest dataset of the user group.

[0211] Based on the user's user group, the user's error correction data and the error correction data of the user group are determined as the target error correction data. The user's error correction data includes the user's interest data.

[0212] Based on the user's user group, the error correction data of the user group and the general error correction data are determined as the target error correction data. The general error correction data includes the global user interest dataset of the target application.

[0213] Based on the user's user group, the user's error correction data, the user group's error correction data, and general error correction data are identified as the target error correction data.

[0214] By using a target model matched to the user, a first error indication is determined for the query information. This target model determines whether the information input to the model is erroneous based on the user's interest data, thus enabling personalized error judgment based on the user's interest data, making the judgment result more closely match the user's actual interest needs. When the query information is erroneous, at least one candidate information associated with the query information is determined based on target error correction data matched to the user. Subsequently, a second error indication is obtained based on the candidate information determined by the target model to further filter out corrective information. The target error correction data includes at least the user's interest data. This allows for personalized error correction of the query information based on the user's interests, ensuring that the corrected information more accurately matches the user's true query intent, meeting the error correction needs of each individual user, improving the actual accuracy of the error judgment and correction process, and thus improving the actual accuracy rate of query error correction.

[0215] The query correction device in this embodiment can execute the query correction method shown in the above embodiments of this application. The implementation principle is similar and will not be described again here.

[0216] Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. For example... Figure 6 As shown, the computer device includes: a memory and a processor; at least one program, stored in the memory, for execution by the processor, which, compared to related technologies, enables:

[0217] By using a target model matched to the user, a first error indication is determined for the query information. This target model determines whether the information input to the model is erroneous based on the user's interest data, thus enabling personalized error judgment based on the user's interest data, making the judgment result more closely match the user's actual interest needs. When the query information is erroneous, at least one candidate information associated with the query information is determined based on target error correction data matched to the user. Subsequently, a second error indication is obtained based on the candidate information determined by the target model to further filter out corrective information. The target error correction data includes at least the user's interest data. This allows for personalized error correction of the query information based on the user's interests, ensuring that the corrected information more accurately matches the user's true query intent, meeting the error correction needs of each individual user, improving the actual accuracy of the error judgment and correction process, and thus improving the actual accuracy rate of query error correction.

[0218] In one alternative embodiment, a computer device is provided, such as Figure 6 As shown, Figure 6The computer device 600 shown includes a processor 601 and a memory 603. The processor 601 and the memory 603 are connected, for example, via a bus 602. Optionally, the computer device 600 may further include a transceiver 604, which can be used for data interaction between the computer device and other computer devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 604 is not limited to one type, and the structure of the computer device 600 does not constitute a limitation on the embodiments of this application.

[0219] Processor 601 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 601 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0220] Bus 602 may include a pathway for transmitting information between the aforementioned components. Bus 602 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 602 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 6 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0221] The memory 603 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0222] The memory 603 stores application code (computer program) that executes the solution of this application, and its execution is controlled by the processor 601. The processor 601 executes the application code stored in the memory 603 to implement the content shown in the foregoing method embodiments.

[0223] The computer equipment includes, but is not limited to: servers, terminals, or any electronic devices that support query and error correction functions.

[0224] This application provides a computer-readable storage medium storing a computer program that, when run on a computer device, enables the computer device to execute the corresponding content of the error correction method described in the foregoing method embodiments.

[0225] This application provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the aforementioned query and error correction method.

[0226] The terms "first," "second," "third," "fourth," "1," "2," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in a sequence other than that shown in the illustrations or text descriptions.

[0227] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0228] The above description is only a partial embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A query error correction method, characterized in that, The method includes: In response to receiving query information from a user's input search box, based on a target model matching the user, a first error indication information is determined for the query information. The first error indication information includes the correctness probability and the error type probability of the query information. The error type probability is used to indicate the possibility that the query information contains at least one type of error. The target model is used to determine whether the information input into the target model is erroneous based on the user's interest data in the target application. The target model includes a user group model. The user group model is used to determine whether the information input into the user group model is erroneous based on the user group's interest dataset. The user group includes at least two users with similar interests to the user. In response to the first error indication information indicating that the query information is incorrect, based on the target error correction data matched with the user and the query information, at least one candidate information associated with the query information is determined, wherein the target error correction data includes at least the user's interest data in the target application, and the candidate information includes related word fragments that replace word fragments in the query information; Based on the target model, a second error indication information is determined for the at least one candidate information, wherein the second error indication information includes the correct probability of the candidate information; Based on the second error indication information, correction information for the query information is determined from the at least one candidate information; The step of determining the correction information for the query information from the at least one candidate information based on the second error indication information includes: Based on the similarity type between the associated word fragment and the replaced word fragment in the query information, the type probability of the matching error type that matches the similar type is obtained from the probability of at least one error type included in the error type probability; Based on the correctness probability and the type probability of the candidate information, the correct score of the candidate information is determined; Based on the correct score of the at least one candidate information and the first error indication information, correction information whose correct score meets the target condition is selected from the at least one candidate information.

2. The query error correction method according to claim 1, characterized in that, The query information includes a query string. The step of determining at least one candidate piece of information associated with the query information based on the target error correction data matched with the user and the query information includes: Based on at least one word fragment included in the query string, at least one related word fragment corresponding to the at least one word fragment is obtained from the target error correction data; Based on the at least one related word fragment, replace the corresponding at least one word fragment in the query information to obtain the at least one candidate information.

3. The query error correction method according to claim 2, characterized in that, The target error correction data also includes a query index for the interest data, which is used to index similar segments in the interest data; The step of obtaining at least one related word fragment corresponding to the at least one word fragment from the target error correction data based on the at least one word fragment included in the query string includes at least one of the following: For each word segment, based on the pinyin of the word segment and the pinyin query index, at least one pinyin-similar word segment of the word segment is queried from the target error correction data. The query index includes a pinyin query index, which is used to index pinyin-similar word segments in the interest data. Based on the stroke sequence and stroke query index of the word segment, at least one word segment with similar character shape is queried from the target error correction data. The query index includes a stroke query index, which is used to index word segments with similar stroke shape in the interest data. Based on the word fragment and the semantic query index, at least one semantically similar word fragment is queried from the target error correction data. The query index includes a semantic query index, which is used to index word fragments with semantically similar keywords in the interest data.

4. The query error correction method according to claim 1, characterized in that, The similarity types include at least one of the following: similar pronunciation, similar character shape, and similar meaning; The matching error types for phonetic similarity include phonetic similarity errors, the matching error types for character shape similarity include character shape similarity errors, and the matching error types for semantic similarity include at least one of character content similarity errors, missing errors, or redundant errors.

5. The query error correction method according to claim 1, characterized in that, The determination of the correct score of the candidate information based on the correctness probability and the type probability includes any one of the following: The product of the type probability and the correctness probability is determined as the correctness score of the candidate information; Based on the matching error type, a correction factor corresponding to the matching error type is obtained from the target association relationship matched with the user. The product value between the type probability, the correction factor, and the correct probability is determined as the correct score of the candidate information. The target association relationship includes the correspondence between the error type and the correction factor.

6. The query error correction method according to claim 1, characterized in that, The determination of the first error indication information of the query information based on the target model matching the user includes: Using the target model, at least one descriptive information of the query information is extracted, and the at least one descriptive information includes at least one of the following: the pinyin, stroke count, position code, keywords included in the query information, or phrases included in the query information. Based on at least one descriptive information of the query information, the query information is judged to be correct, and the probability of the query information being correct and the probability of the error type are obtained.

7. The query error correction method according to claim 6, characterized in that, The training process of the user group model includes: Obtain historical query information from at least two users in the user group, and generate negative sample data including noise data by performing a noise generation operation on a portion of the historical query information. The noise generation operation includes at least one of the following: replacing word fragments, deleting word fragments, or adding word fragments. The replacing word fragment operation includes at least one of the following: replacing with pinyin-similar word fragments, replacing with character-similar word fragments, or replacing with semantically similar word fragments. The negative sample data is predicted using the initial model to obtain the predicted sample data; Based on the similarity between the predicted sample data and the positive sample data of the historical query information, the model parameters of the initial model are adjusted until the initial model meets the model training conditions, and the adjustment is stopped to obtain the user group model.

8. The query error correction method according to claim 1, characterized in that, Before determining the first error indication information of the query information based on the target model matching the user, the method further includes: In response to receiving query information from a user's search box, the user maturity level is determined, which indicates the user's activity level in the target application. When the user's maturity exceeds a maturity threshold, at least one target model matching the user is determined based on a user group with similar interests.

9. The query error correction method according to claim 8, characterized in that, The step of determining at least one target model matching the user based on a user group with similar interests includes at least one of the following: Based on a user group with similar interests to the user, the user group model corresponding to the user group is determined as the target model that matches the user. Based on a user group with similar interests to the user, the general model and the user group model are determined as the target models that match the user. The general model is used to determine whether the information input to the general model is incorrect based on the global user dataset of the target application.

10. The query error correction method according to claim 8, characterized in that, Determining the user maturity level of the user includes: Determine the user's active time in the target application; Based on the operation object associated with the user's interactive operation, the number of interest tags of the user during the active time is counted. The user maturity level is determined based on the active time and the number of interest tags.

11. The query error correction method according to claim 1, characterized in that, Before determining the first error indication information of the query information based on the target model matching the user, in response to receiving query information from the user's input search box, the method further includes at least one of the following: Receive the query string entered by the user in the search box of the target application, and use the query string as the query information; Receive the query image input by the user through the search box, extract the string included in the query image, and use the string extracted from the query image as the query information; The system receives an emoji icon input by the user through the search box, obtains the text description information of the emoji icon, and uses the text description information of the emoji icon as the query information. The system receives a query string and an emoji icon input by the user through the search box, obtains the text description information of the emoji icon, and uses the query string and the text description information of the emoji icon as the query information.

12. The query error correction method according to claim 1, characterized in that, Before determining at least one candidate information of the query information based on the query information and the target error correction data matching the user, the method further includes any one of the following: Based on the user's user group, the error correction data of the user group is determined as the target error correction data that matches the user. The error correction data of the user group includes the interest dataset of the user group. Based on the user's user group, the error correction data of the user and the error correction data of the user group are determined as the target error correction data, wherein the error correction data of the user includes the user's interest data; Based on the user group of the user, the error correction data of the user group and the general error correction data are determined as the target error correction data, and the general error correction data includes the interest dataset of the global users of the target application. Based on the user's user group, the error correction data of the user, the error correction data of the user group, and the general error correction data are determined as the target error correction data.

13. A query error correction device, characterized in that, The device includes: A first determining module is configured to, in response to receiving query information from a user input search box, determine first error indication information of the query information based on a target model matching the user. The first error indication information includes the correctness probability and error type probability of the query information. The error type probability indicates the possibility that the query information contains at least one type of error. The target model is configured to determine whether the information input into the target model is erroneous based on the user's interest data in the target application. The target model includes a user group model. The user group model is configured to determine whether the information input into the user group model is erroneous based on the user group's interest dataset. The user group includes at least two users with similar interests to the user. A candidate determination module is configured to, in response to the first error indication information indicating that the query information is incorrect, determine at least one candidate information associated with the query information based on target error correction data matching the user and the query information, wherein the target error correction data includes at least the user's interest data in the target application, and the candidate information includes related word fragments that replace word fragments in the query information; The second determining module is used to determine, based on the target model, second error indication information of the at least one candidate information, wherein the second error indication information includes the correct probability of the candidate information; The correction module is configured to determine correction information for the query information from the at least one candidate information based on the second error indication information; The correction module includes: The correction unit is configured to, based on the similarity type between the associated word fragment and the replaced word fragment in the query information, obtain the type probability of the matching error type that matches the similarity type from the probability of at least one error type included in the error type probability; and determine the correct score of the candidate information based on the correct probability of the candidate information and the type probability. A filtering unit is configured to filter out correction information whose correct scores meet the target conditions from the at least one candidate information based on the correct scores of the at least one candidate information and the first error indication information.

14. The apparatus according to claim 13, characterized in that, The query information includes a query string, and the candidate determination module includes: The associated word acquisition unit is used to acquire at least one associated word fragment corresponding to the at least one word fragment included in the query string from the target error correction data; The candidate acquisition unit is used to replace at least one corresponding word fragment in the query information based on the at least one associated word fragment to obtain the at least one candidate information.

15. The apparatus according to claim 14, characterized in that, The target error correction data also includes a query index for the interest data, which is used to index similar segments in the interest data; The associated word acquisition unit is used for at least one of the following: For each word segment, based on the pinyin of the word segment and the pinyin query index, at least one pinyin-similar word segment of the word segment is queried from the target error correction data. The query index includes a pinyin query index, which is used to index pinyin-similar word segments in the interest data. Based on the stroke sequence and stroke query index of the word segment, at least one word segment with similar character shape is queried from the target error correction data. The query index includes a stroke query index, which is used to index word segments with similar stroke shape in the interest data. Based on the word fragment and the semantic query index, at least one semantically similar word fragment is queried from the target error correction data. The query index includes a semantic query index, which is used to index word fragments with semantically similar keywords in the interest data.

16. The apparatus according to claim 13, characterized in that, The similarity types include at least one of the following: similar pronunciation, similar character shape, and similar meaning; The matching error types for phonetic similarity include phonetic similarity errors, the matching error types for character shape similarity include character shape similarity errors, and the matching error types for semantic similarity include at least one of character content similarity errors, missing errors, or redundant errors.

17. The apparatus according to claim 13, characterized in that, The correction unit is used for any of the following: The product of the type probability and the correctness probability is determined as the correctness score of the candidate information; Based on the matching error type, a correction factor corresponding to the matching error type is obtained from the target association relationship matched with the user. The product value between the type probability, the correction factor, and the correct probability is determined as the correct score of the candidate information. The target association relationship includes the correspondence between the error type and the correction factor.

18. The apparatus according to claim 13, characterized in that, The first determining module is used for: Using the target model, at least one descriptive information of the query information is extracted, and the at least one descriptive information includes at least one of the following: the pinyin, stroke count, position code, keywords included in the query information, or phrases included in the query information. Based on at least one descriptive information of the query information, the query information is judged to be correct, and the probability of the query information being correct and the probability of the error type are obtained.

19. The apparatus according to claim 18, characterized in that, Also includes: The model training module is used for: Obtain historical query information from at least two users in the user group, and generate negative sample data including noise data by performing a noise generation operation on a portion of the historical query information. The noise generation operation includes at least one of the following: replacing word fragments, deleting word fragments, or adding word fragments. The replacing word fragment operation includes at least one of the following: replacing with pinyin-similar word fragments, replacing with character-similar word fragments, or replacing with semantically similar word fragments. The negative sample data is predicted using the initial model to obtain the predicted sample data; Based on the similarity between the predicted sample data and the positive sample data of the historical query information, the model parameters of the initial model are adjusted until the initial model meets the model training conditions, and the adjustment is stopped to obtain the user group model.

20. The apparatus according to claim 13, characterized in that, Also includes: The user maturity determination module is used to determine the user maturity level in response to receiving query information from a user's input search box. The user maturity level is used to indicate the user's activity level in the target application. The target model determination module is used to determine at least one target model that matches the user based on a user group with similar interests when the user's maturity exceeds a maturity threshold.

21. The apparatus according to claim 20, characterized in that, The target model determination module is used for: Based on a user group with similar interests to the user, the user group model corresponding to the user group is determined as the target model that matches the user. Based on a user group with similar interests to the user, the general model and the user group model are determined as the target models that match the user. The general model is used to determine whether the information input to the general model is incorrect based on the global user dataset of the target application.

22. The apparatus according to claim 20, characterized in that, The user maturity determination module is used for: Determine the user's active time in the target application; Based on the operation object associated with the user's interactive operation, the number of interest tags of the user during the active time is counted. The user maturity level is determined based on the active time and the number of interest tags.

23. The apparatus according to claim 13, characterized in that, The device further includes at least one of the following: The first query module is used to receive the query string entered by the user in the search box of the target application, and use the query string as the query information; The second query module is used to receive the query image input by the user through the search box, extract the string included in the query image, and use the string extracted from the query image as the query information; The third query module is used to receive the emoji icon entered by the user through the search box, obtain the text description information of the emoji icon, and use the text description information of the emoji icon as the query information; The fourth query module is used to receive the query string and emoticon icon entered by the user through the search box, obtain the text description information of the emoticon icon, and use the query string and the text description information of the emoticon icon as the query information.

24. The apparatus according to claim 13, characterized in that, The device further includes an error correction data determination module, used for any of the following: Based on the user's user group, the error correction data of the user group is determined as the target error correction data that matches the user. The error correction data of the user group includes the interest dataset of the user group. Based on the user's user group, the error correction data of the user and the error correction data of the user group are determined as the target error correction data, wherein the error correction data of the user includes the user's interest data; Based on the user group of the user, the error correction data of the user group and the general error correction data are determined as the target error correction data, and the general error correction data includes the interest dataset of the global users of the target application. Based on the user's user group, the error correction data of the user, the error correction data of the user group, and the general error correction data are determined as the target error correction data.

25. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the query error correction method according to any one of claims 1 to 12.

26. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the query and error correction method according to any one of claims 1 to 12.

27. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the query and error correction method according to any one of claims 1 to 12.

Images