A method, apparatus, storage medium, and electronic device for identifying incorrect question tags.
By analyzing the semantic relationships between failed and successful retrieval events in the intelligent error notebook, and automatically supplementing and expanding tags, the problem of the lack of standardization and dynamic updating of the tag system in the intelligent error notebook is solved, thereby improving the accuracy and efficiency of retrieval and meeting the user's error retrieval needs.
Patent Information
- Application Number
- CN202611636899.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-22
- Publication Date
- 2026-11-17
AI Technical Summary
The existing intelligent error correction notebooks have fixed preset static tag expressions, and user-defined tags are limited by personal expression habits. As a result, the tag system does not have the ability to standardize and dynamically update, and cannot identify the semantic relationship between the actual search tags and the preset and user-defined tags. This leads to low search accuracy and efficiency, which affects the user's review efficiency and the effect of consolidating knowledge points.
By acquiring the target users' incorrect question retrieval needs and failed retrieval events, the semantic relationship between failed retrieval tags and successful retrieval tags is analyzed, candidate related events are screened, and expanded tags are determined based on operational behavior data and tag confidence. Tags are automatically supplemented to improve retrieval accuracy.
It enables accurate filtering of matching incorrect questions when user-defined tags and preset tags have similar semantics but different literal meanings, improving the hit rate of the search, and enhancing the user's review efficiency and knowledge consolidation effect.
Smart Images

Figure CN123115397A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of educational technology, and in particular to a method, apparatus, storage medium, and electronic device for identifying incorrect question labels. Background Technology
[0002] With the development of educational information technology, intelligent error notebooks are widely used in users' scenarios of organizing and reviewing incorrect questions. Intelligent error notebooks can classify and label incorrect questions and perform conditional searches. Users can retrieve the incorrect questions they need to practice based on their own learning situation, and conduct targeted review and knowledge consolidation.
[0003] Currently, intelligent error notebooks generally adopt a preset static tag system, with tags categorized by subject, knowledge point, and error type. When entering errors, users can select the corresponding tags from the preset static tags to mark the errors, or they can customize error tags, manually enter the tag content, and bind it to the corresponding error. When searching for errors, users can manually select preset filtering conditions such as subject, knowledge point, and error type, as well as custom tags, to filter out errors that meet the criteria.
[0004] However, existing smart error notebooks use fixed, pre-defined static tags, while user-defined tags are limited by personal expression habits, resulting in a lack of standardization and dynamic updating capabilities. When the user's actual search tags are identical to the pre-defined and custom tags, the smart error notebook can filter out the errors matching the actual search tags. However, when the user has omitted or mislabeled tags, or when the actual search tags are semantically similar but literally different from the pre-defined and custom tags due to different expression habits, the smart error notebook cannot recognize the semantic relationship between the actual search tags and the pre-defined and custom tags. This results in the inability to filter out errors matching the actual search tags, requiring the user to change the actual search tags and perform multiple searches, or manually add tags to the errors. This method reduces the accuracy and efficiency of the search, failing to meet the user's error retrieval needs and consequently affecting the user's review efficiency and knowledge consolidation. Summary of the Invention
[0005] In view of this, this application provides a method, apparatus, storage medium, and electronic device for determining error tags. The main purpose is to improve upon existing intelligent error notebooks where the preset static tag expressions are fixed, and user-defined tags are limited by personal expression habits, resulting in a lack of standardization and dynamic updating capabilities. When the user's actual search tags are completely identical to the preset and user-defined tags, the intelligent error notebook can filter out the error questions matching the actual search tags. However, when the user has omitted tags, mislabeled tags, or uses different expression habits resulting in semantically similar but literally different actual search tags compared to the preset and user-defined tags, the intelligent error notebook cannot recognize the semantic relationship between the actual search tags and the preset and user-defined tags, thus failing to filter out the error questions matching the actual search tags. Users need to change the actual search tags and perform multiple re-searches, or manually add tags to the error questions. Using this method reduces the accuracy and efficiency of the search, failing to meet the user's error question retrieval needs, and consequently affecting the user's review efficiency and knowledge consolidation effect.
[0006] Firstly, this application provides a method for determining incorrect question tags, including: The system obtains the target user's target error retrieval needs in the intelligent error notebook, as well as the failed retrieval events of the target user's error retrieval according to the target retrieval tags corresponding to the target error retrieval needs. The failed retrieval events are retrieval events in which no error questions matching the error retrieval needs are found. Based on the semantic association between the target search tag corresponding to the failed search event and the successful search tag corresponding to the successful search event, candidate association events that are semantically related to the failed search event are selected from the successful search events. The successful search event is a search event in which the wrong question that meets the wrong question search requirements is retrieved by changing the search tag within the target time window after the failed search event. Obtain a set of candidate incorrect questions for the candidate related events. Based on the target user's operation behavior data in the candidate related events, analyze the degree of matching between each candidate incorrect question in the set of candidate incorrect questions and the target incorrect question retrieval requirement. Determine the relevance score between each candidate incorrect question and the target incorrect question retrieval requirement. Based on the relevance score, select target related questions that match the target incorrect question retrieval requirement from the set of candidate incorrect questions. Analyze the tag confidence of the target retrieval tag relative to the target associated question, and determine the target retrieval tag that meets the confidence condition as the extended tag of the target associated question. The extended tag is used to retrieve the target associated question.
[0007] Optionally, the step of obtaining the candidate incorrect question set of the candidate related events involves analyzing the matching degree between each candidate incorrect question in the candidate incorrect question set and the target incorrect question retrieval requirement based on the target user's operation behavior data in the candidate related events, determining the relevance score between each candidate incorrect question and the target incorrect question retrieval requirement, and filtering target related questions that match the target incorrect question retrieval requirement from the candidate incorrect question set based on the relevance score, including: Based on the learning behavior data of the target user in the candidate related events, evaluate whether the target user learns after retrieving the candidate set of incorrect questions, and obtain the learning intensity data of the target user learning the candidate set of incorrect questions; Based on the retrieval dimension information of the target retrieval tag, analyze whether the retrieval dimension of the target retrieval tag and each candidate wrong question in the candidate wrong question set are consistent, and obtain the dimension consistency data of the target retrieval tag and each candidate wrong question in the candidate wrong question set; Based on the time information of the candidate related events, the temporal sequence information of the candidate related events relative to the failed retrieval events is analyzed to obtain the temporal priority data of the candidate related events; Based on the semantic association, the learning intensity data, the dimensional consistency data, and the temporal priority data, the matching degree between each candidate wrong question in the candidate wrong question set and the target wrong question retrieval requirement is analyzed to obtain the relevance score between each candidate wrong question and the target wrong question retrieval requirement, and the candidate wrong questions that meet the relevance score condition are determined as target related questions that match the target wrong question retrieval requirement.
[0008] Optionally, the step of analyzing the tag confidence of the target search tag relative to the target associated question, and determining the target search tag that meets the confidence condition as the extended tag of the target associated question, includes: Based on the text content of the target-related question, the semantic similarity between the target search tag and the text content of the question is analyzed to obtain the first semantic similarity data between the target search tag and the target-related question; Based on the existing tag set corresponding to the target associated question, the semantic similarity between the target retrieval tag and each existing tag in the existing tag set is analyzed to obtain the second semantic similarity data between the target retrieval tag and the target associated question; Based on the target knowledge points contained in the target-related questions, analyze whether the search knowledge points contained in the target search tags are consistent with the target knowledge points, and obtain the consistency data of the knowledge points between the target search tags and the target-related questions; Based on the target error type of the target-related question, analyze whether the retrieval error type of the target retrieval tag is consistent with the target error type, and obtain the consistency data of the error type between the target retrieval tag and the target-related question; Based on historical search records, the historical correlation between the target search tag and the existing tags of the target associated questions is analyzed to obtain historical correlation data between the target search tag and the target associated questions; Based on the first semantic similarity data, the second semantic similarity data, the knowledge point consistency data, the error type consistency data, and the historical association data, the tag confidence of the target retrieval tag relative to the target associated question is determined, and the target retrieval tag whose tag confidence satisfies the confidence condition is determined as the extended tag of the target associated question.
[0009] Optionally, the step of filtering candidate related events that are semantically related to the failed search event from the successful search events based on the semantic association between the target search tag corresponding to the failed search event and the successful search tag corresponding to the successful search event includes: Within the target time window following the failed retrieval event, the retrieval behavior of the target user in the intelligent error notebook is monitored; In response to the target user finding an incorrect question that matches the incorrect question retrieval requirement through the target retrieval behavior of changing the search tags, the retrieval event corresponding to the target retrieval behavior is determined as the successful retrieval event; Obtain all successful search events of the target user within the target time window and all successful search tags of the successful search events, analyze the text semantic similarity between the target search tag and each of the successful search tags, and obtain the tag semantic similarity data between the target search tag and each of the successful search tags; Analyze the keyword similarity between the target search tag and each of the successful search tags, and obtain the keyword similarity data between the target search tag and each of the successful search tags; Based on the target academic attribute information corresponding to the target search tag, analyze the consistency between the target academic attribute information and the candidate academic attribute information corresponding to each successful search tag, and the academic attribute consistency data between the target search tag and each successful search tag; Based on the semantic similarity data of the tags, the keyword similarity data, and the academic attribute consistency data, the semantic association between the target retrieval tag and each of the successfully retrieved tags is determined, and candidate association events that meet the tag association conditions are selected from the successful retrieval events.
[0010] Optionally, the step of obtaining the target user's target error retrieval needs in the intelligent error notebook, and the failed retrieval events of the target user performing error retrieval according to the target retrieval tags corresponding to the target error retrieval needs, includes: Obtain the retrieval language data of the target user when searching for incorrect questions in the intelligent error notebook, perform semantic analysis on the retrieval language data, and obtain the target user's target error search needs; The target user's target search intent is obtained by identifying the target user's search intent for the target incorrect question retrieval request. Based on the target search intent, the target user's search time intent, search subject intent, and incorrect question type intent are analyzed to obtain the target search tags corresponding to the target search intent. The wrong question is retrieved based on the target search tag, and the search event in which the target user does not find a wrong question that meets the wrong question search requirements is determined as the failed search event.
[0011] Optionally, the method further includes: In response to the addition of a new incorrect question in the intelligent error notebook, the multi-dimensional feature information of the new incorrect question is obtained; Based on the multi-dimensional feature information, feature analysis is performed on the historical wrong questions in the intelligent wrong question notebook, and at least one candidate historical wrong question with similar features to the newly added wrong question is selected from the intelligent wrong question notebook; Obtain a set of candidate error labels for the at least one candidate historical error question, analyze the matching degree between each candidate error label in the candidate error label set and the label of the newly added error question, and determine at least one candidate label that matches the newly added error question from the candidate error label set; Analyze the confidence level of the at least one candidate label relative to the label of the newly added incorrect question, and determine the candidate label that meets the confidence level condition as the new extended label of the newly added incorrect question. The new extended label is used to retrieve the newly added incorrect question.
[0012] Secondly, this application provides a device for determining incorrect question tags, comprising: The acquisition module is configured to acquire the target user's target error retrieval needs in the intelligent error notebook, and the failed retrieval events of the target user's error retrieval according to the target retrieval tags corresponding to the target error retrieval needs. The failed retrieval events are retrieval events in which no error questions matching the error retrieval needs are found. The filtering module is configured to filter candidate related events that are semantically related to the failed search event from the successful search events based on the semantic relationship between the target search tag corresponding to the failed search event and the successful search tag corresponding to the successful search event. The successful search event is a search event in which a wrong question that meets the wrong question search requirements is retrieved by changing the search tag within a target time window after the failed search event. The analysis module is configured to obtain a set of candidate incorrect questions for the candidate related events, analyze the matching degree between each candidate incorrect question in the set and the target incorrect question retrieval requirement based on the target user's operation behavior data in the candidate related events, determine the relevance score between each candidate incorrect question and the target incorrect question retrieval requirement, and filter out the target related questions that match the target incorrect question retrieval requirement from the set of candidate incorrect questions based on the relevance score. The determination module is configured to analyze the tag confidence of the target retrieval tag relative to the target associated question, and determine the target retrieval tag that meets the confidence condition as the extended tag of the target associated question, the extended tag being used to retrieve the target associated question.
[0013] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the error label determination method described in the first aspect.
[0014] Fourthly, this application provides an electronic device, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the computer program to implement the error label determination method described in the first aspect.
[0015] Fifthly, this application provides a computer program product, which includes a computer program that, when executed by a processor, implements the error label determination method described in the first aspect.
[0016] By means of the above technical solution, this application provides a method, apparatus, storage medium, and electronic device for determining incorrect question tags, comprising: acquiring the target incorrect question retrieval needs of a target user in an intelligent incorrect question notebook, and the failed retrieval events of the target user performing incorrect question retrieval according to the target retrieval tags corresponding to the target incorrect question retrieval needs, wherein a failed retrieval event is a retrieval event in which no incorrect question matching the incorrect question retrieval needs is found; based on the semantic association relationship between the target retrieval tags corresponding to the failed retrieval events and the successful retrieval tags corresponding to the successful retrieval events, selecting candidate association events that are semantically related to the failed retrieval events from the successful retrieval events, wherein the successful retrieval events are the target retrieval events following the failed retrieval events. Within a time window, the system retrieves search events where the search tags are changed to find incorrect questions that meet the search requirements. It then obtains a set of candidate incorrect questions for related events, analyzes the matching degree between each candidate incorrect question and the target incorrect question search requirement based on the target user's operational behavior data within these events, determines the relevance score between each candidate incorrect question and the target incorrect question search requirement, and filters out target related questions that match the target incorrect question search requirement from the candidate incorrect question set based on the relevance score. Finally, it analyzes the tag confidence of the target search tags relative to the target related questions, and determines the target search tags that meet the confidence criteria as extended tags for the target related questions. These extended tags are used to retrieve target related questions. Compared with existing technologies, this application filters candidate related events based on the semantic relationship between target retrieval tags and successful retrieval tags, establishes a connection between failed retrievals and subsequent successful retrievals, and provides a source of target questions for tag supplementation; it analyzes the matching degree between candidate incorrect questions and target incorrect question retrieval needs based on operational behavior data and determines the relevance score, accurately selecting target related questions that match the failed retrieval needs from the candidate incorrect question set; it analyzes the tag confidence of target retrieval tags relative to target related questions and determines extended tags, supplementing the tags used by users in failed retrievals as extended tags for target related questions, realizing automatic supplementation of extended tags, improving the hit rate of subsequent similar retrievals, meeting users' incorrect question retrieval needs, and thus improving users' review efficiency and knowledge point consolidation effect. Attached Figure Description
[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1A flowchart illustrating a method for determining incorrect question tags provided in an embodiment of this application is shown; Figure 2 A flowchart illustrating a method for determining incorrect question tags provided in an embodiment of this application is shown; Figure 3 This illustration shows a structural schematic diagram of a device for determining incorrect question tags according to an embodiment of this application; Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation
[0020] The embodiments of this application will now be described in more detail with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.
[0021] To address the limitations of existing intelligent error correction notebooks, where pre-set static tags have a fixed format and user-defined tags are restricted by personal expression habits, resulting in a lack of standardization and dynamic updating capabilities, this embodiment provides a method for determining error correction tags. This method addresses the issue that, while intelligent error correction notebooks can filter out errors when the user's actual search tags match the pre-set and custom tags, errors occur when the user has omitted or mislabeled tags, or when the actual search tags are semantically similar but literally different from the pre-set and custom tags due to different expression habits. In such cases, the intelligent error correction notebook cannot recognize the semantic relationship between the actual search tags and the pre-set and custom tags, failing to filter out errors that match the actual search tags. Users then need to change the actual search tags and perform multiple searches, or manually add tags to the errors. This reduces the accuracy and efficiency of the search, failing to meet users' error correction retrieval needs and consequently affecting their review efficiency and knowledge consolidation. Figure 1 As shown, the method includes: Step 101: Obtain the target user's target error retrieval needs in the intelligent error notebook, as well as the failed retrieval events of the target user's error retrieval according to the target retrieval tags corresponding to the target error retrieval needs.
[0022] Among them, a failed search event is a search event in which no incorrect question that meets the requirements for incorrect question retrieval is found.
[0023] In this embodiment, the target user can be a user who uses the smart error notebook to retrieve and learn from incorrect questions. The target user can initiate an error retrieval request, view the retrieval results, and perform learning operations on the retrieved errors. For example, the target user in this embodiment can be a student user, a teacher user, or a parent user.
[0024] In the embodiments of this application, the intelligent error notebook can be an educational information technology application or system that supports error entry, error management, and error retrieval. The intelligent error notebook can be used to store users' error data, manage error tags, and respond to users' error retrieval requests. For example, the intelligent error notebook in the embodiments of this application is specifically an application that supports classification and filtering by subject, time, tag, or error cause.
[0025] In this embodiment, error retrieval can be an operation where a target user enters search criteria in a smart error notebook to find error questions that meet their needs. Error retrieval can be used to filter error questions that meet specific conditions from the user's error question database. For example, error retrieval in this embodiment can specifically include various retrieval methods such as voice input retrieval, text input retrieval, and tag-based filtering retrieval.
[0026] In this embodiment of the application, the target error retrieval requirement can be the actual search intent and condition combination expressed by the target user when initiating an error retrieval. The target error retrieval requirement can be used to characterize all the constraints that the target user hopes to find for the error. For example, in this embodiment of the application, the target error retrieval requirement can specifically be the search intent corresponding to natural language expressions such as "help me find the math calculation errors from last week".
[0027] In this embodiment, the target retrieval tag can be a specific value of the tag dimension in the structured retrieval conditions obtained based on the target error retrieval requirement. The target retrieval tag can be used as one of the filtering conditions for error retrieval. For example, the target retrieval tag in this embodiment may specifically include standard tags such as calculation error and misreading the question, or it may include tag values corresponding to user-verbal expressions such as calculation error, carelessness, and incorrect formula.
[0028] In this application embodiment, a failed search event can be a search event where the system fails to find any incorrect questions that meet the incorrect question search requirements after the target user submits a search request. A failed search event can be used as a trigger signal to identify new tag needs and optimize tag mapping relationships. A failed search does not indicate that the user's need is invalid, but rather reflects that the current incorrect question tagging system has not yet fully covered the user's actual expression. Therefore, a failed search can be considered an important trigger signal for identifying new tag needs and optimizing tag mapping relationships. Compared to users actively adding tags, failed searches can have stronger target orientation. Users only engage in failed search behavior when they have a genuine search need and existing tags cannot meet it. Therefore, this behavior can more accurately reveal the user's actual usage habits, including spoken expressions, personalized error descriptions, and potential tag gaps. This application can use failed searches as the starting point for the tag self-evolution process. For example, in this application embodiment, a failed search event can specifically be a search event where the search results are empty, or a search event where the number of search results is lower than a preset threshold.
[0029] In this embodiment, when the system determines a search to be a failed search, it can automatically generate a corresponding failed search record. The failed search record can include at least the query text, the parsed search dimensions and their values, at least one piece of contextual information from the subject, year, and question type of the query, a query timestamp, a user identifier, a session identifier, and a terminal identifier. The search dimensions and their values can be used to represent the user's search criteria, such as subject, time range, knowledge point, error type, question type, and answer status. By preserving the structured context at the time of the failed search, a basis can be provided for subsequent correlation analysis between failed and successful searches.
[0030] In the embodiments of this application, when a target user initiates a search for incorrect questions in the intelligent error notebook, the system can obtain the search language data input by the target user, perform semantic analysis on the search language data to obtain the target error question search requirements, and at the same time, parse the corresponding target search tags based on the target error question search requirements to obtain the corresponding target search tags. The system performs error question search according to the target search tags. If no error question matching the target error question search requirements is found, the search event is determined as a failed search event and recorded.
[0031] Step 102: Based on the semantic relationship between the target search tag corresponding to the failed search event and the successful search tag corresponding to the successful search event, select candidate related events that are semantically related to the failed search event from the successful search events.
[0032] Among them, a successful retrieval event is a retrieval event in which a retrieval event is completed within the target time window after a failed retrieval event, and the retrieval tag is changed to find a retrieval of a retrieval question that meets the requirements for retrieval of retrieval questions.
[0033] In this embodiment, a successful retrieval event can be a retrieval event in which a target user finds a correct incorrect question that meets the requirements of incorrect question retrieval after changing the search tags following a failed retrieval event. A successful retrieval event can be used to provide a source of relevant incorrect questions for a failed retrieval event. For example, in this embodiment, a successful retrieval event can specifically be a retrieval event in which a target user changes from a failed calculation error retrieval to a calculation error retrieval and successfully finds the correct incorrect question.
[0034] In the embodiments of this application, the target time window can be a preset time range extending backward from the occurrence time of the failed retrieval event. The target time window can be used to limit the time correlation range between successful retrieval events and failed retrieval events. For example, in the embodiments of this application, the target time window can be a 5-minute sliding window, or it can be other preset durations such as 10 minutes or 30 minutes.
[0035] In this embodiment, semantic association can be the similarity between two search tags in terms of semantic expression, structured features, or tag meaning. Semantic association can be used to measure whether failed search tags and successful search tags point to the same type of incorrect question requirement. Similarity can be used to characterize the degree of closeness between two objects in terms of semantic expression, structured features, or tag meaning, and can be determined based on at least one of text vector similarity, keyword overlap, knowledge point consistency, and error type consistency.
[0036] In the embodiments of this application, candidate related events can be retrieval events that are semantically related to failed retrieval events and are selected from successful retrieval events. Candidate related events can be used to provide a set of candidate incorrect questions for subsequent selection of target related questions.
[0037] It should be noted that the associated object is not the two queries themselves, but rather the relationship between the failed search record and the target incorrect question hit by the subsequent successful search. That is, the system can determine whether the incorrect question hit by the subsequent successful search is the actual target incorrect question corresponding to the aforementioned failed search by analyzing user consistency, temporal continuity, semantic similarity, and behavioral confirmation information between the failed and successful searches. For example, in this embodiment, the candidate associated event can specifically be a successful search event that corresponds to the same user as the failed search event, occurs in the same session, and whose query text semantic similarity meets a threshold condition.
[0038] In this embodiment of the application, after obtaining a failed search event, the system can monitor the target user's search behavior within the target time window after the failed search event. When it is detected that the target user has found a wrong question that meets the wrong question search requirements by changing the search tags, the search is determined as a successful search event. Then, the semantic association between the target search tags corresponding to the failed search event and the successful search tags corresponding to each successful search event is analyzed, and candidate association events that meet the tag association conditions are selected from the successful search events.
[0039] Step 103: Obtain the candidate incorrect question set of the candidate related events. Based on the target user's operation behavior data in the candidate related events, analyze the matching degree between each candidate incorrect question in the candidate incorrect question set and the target incorrect question retrieval requirement. Determine the relevance score between each candidate incorrect question and the target incorrect question retrieval requirement. Based on the relevance score, select the target related questions that match the target incorrect question retrieval requirement from the candidate incorrect question set.
[0040] In this embodiment of the application, the candidate incorrect question set can be a set of all incorrect questions matched by the candidate related event. The candidate incorrect question set can be used to provide candidate incorrect questions to be filtered to determine the target related question. For example, the candidate incorrect question set in this embodiment of the application may specifically include one or more candidate incorrect questions.
[0041] In this embodiment, the operational behavior data can be the behavioral record data generated by the target user's various operations on the candidate incorrect questions in the candidate related events. The operational behavior data can be used to characterize the target user's level of attention and confirmation of the candidate incorrect questions. For example, the operational behavior data in this embodiment may specifically include the user's behavior data of clicking on a certain incorrect question on the successful search results page, the user's browsing behavior data of staying on a certain incorrect question for more than a preset time, and the user's behavior data of performing subsequent operations such as viewing the analysis, practicing again, and adding to review for a certain incorrect question.
[0042] In this embodiment, the relevance score can be a quantitative score that measures the degree of matching between candidate incorrect questions and the target incorrect question retrieval requirement. The relevance score can be used to filter out the target related questions that truly correspond to the failed retrieval requirement from the candidate incorrect question set. If there are multiple candidate incorrect questions in the successful retrieval results, the system can select the final related question based on the relevance score. For example, the relevance score in this embodiment can specifically be a value between 0 and 1 obtained by weighting multiple dimensions of factors.
[0043] In this embodiment, the target related question can be a question selected from the candidate incorrect question set that meets the matching condition of the target incorrect question retrieval requirement. The target related question can be used as the attachment object of the extended tag. For example, in this embodiment, the target related question can specifically be a candidate incorrect question with a relevance score not lower than a preset relevance threshold.
[0044] In this embodiment of the application, after screening candidate related events, the system can obtain the set of candidate incorrect questions that the candidate related events hit, and analyze the degree of matching between each candidate incorrect question and the target incorrect question retrieval needs based on the target user's operation behavior data in the candidate related events, thereby determining the relevance score corresponding to each candidate incorrect question, and filtering the target related question from the set of candidate incorrect questions according to the relevance score.
[0045] Step 104: Analyze the confidence of the target search tags relative to the target related questions, and determine the target search tags that meet the confidence conditions as the extended tags of the target related questions.
[0046] Among them, extended tags are used to retrieve related questions.
[0047] In this embodiment, tag confidence can be a quantitative indicator representing the credibility of adding a target search tag to the target associated question tag set. Tag confidence can be used to determine whether a target search tag is suitable as an extended tag for a target associated question. The system can analyze the matching degree between candidate extended tags and target associated questions based on preset rules and generate a tag confidence standard. For example, in this embodiment, the tag confidence can specifically be a value between 0 and 1 obtained by weighting multiple matching factors.
[0048] In the embodiments of this application, the confidence level condition can be a preset threshold condition used to determine whether the confidence level of the tag meets the requirements for generating extended tags. The confidence level condition can be used to filter candidate tags with low confidence. The tag confidence level standard can be calibrated through historical retrieval logs, manually verified samples, and online operation feedback, and the first confidence level threshold and the second confidence level threshold can be determined by combining the accuracy, false positive rate, and recall rate after tag supplementation.
[0049] In this embodiment, tags can be divided into two categories: standard tags and extended tags. Standard tags can be used to represent system-preset or standardized tags. For example, standard tags in this embodiment can specifically include calculation errors, misreading the question, etc. Extended tags can be used to represent colloquial expressions, synonyms, or personalized expressions used by users in natural language retrieval. Extended tags can be used to improve the hit rate of natural language retrieval without disrupting the original standard tag system. For example, extended tags in this embodiment can specifically include calculation errors, carelessness, incorrect formulas, etc. By establishing a mapping relationship between extended tags and standard tags, the hit rate of natural language retrieval can be improved without disrupting the original standard tag system, achieving a self-evolving effect of becoming more accurate with each search.
[0050] In this embodiment of the application, after determining the target related topic, the system can extract the semantic expression that the user did not match from the failed search records, and use the semantic expression as a candidate extended tag. It can then establish a candidate mapping relationship with the existing standard tags of the target related topic, analyze the tag confidence of the target search tag relative to the target related topic, and determine the target search tag that meets the confidence condition as the extended tag of the target related topic, so that the target user can retrieve the target related topic through the extended tag.
[0051] Compared with existing technologies, this embodiment filters candidate related events based on the semantic relationship between the target retrieval tag and the successful retrieval tag, establishes a connection between failed retrieval and subsequent successful retrieval, and provides a source of target questions for tag supplementation; it analyzes the matching degree between candidate incorrect questions and target incorrect question retrieval needs based on operation behavior data and determines the relevance score, accurately selecting target related questions that match the failed retrieval needs from the candidate incorrect question set; it analyzes the tag confidence of the target retrieval tag relative to the target related question and determines extended tags, supplementing the tags used by the user in failed retrieval as extended tags for the target related question, realizing automatic supplementation of extended tags, improving the hit rate of subsequent similar retrievals, meeting the user's incorrect question retrieval needs, and thus improving the user's review efficiency and knowledge point consolidation effect.
[0052] As an optional approach, when performing the following steps: "obtaining a set of candidate incorrect questions for candidate related events, analyzing the matching degree between each candidate incorrect question in the set and the target incorrect question retrieval requirement based on the target user's operational behavior data in the candidate related events, determining the relevance score between each candidate incorrect question and the target incorrect question retrieval requirement, and selecting target related questions that match the target incorrect question retrieval requirement from the candidate incorrect question set based on the relevance score," the following methods can be used, but are not limited to: Figure 2 As shown, the method includes: Step 201: Based on the learning behavior data of the target user in the candidate related events, evaluate whether the target user learns after retrieving the candidate wrong question set, and obtain the learning intensity data of the target user learning the candidate wrong question set.
[0053] In this embodiment, learning behavior data can be recorded data of learning-related operations performed by a target user on candidate incorrect questions after retrieving the candidate incorrect question set. This learning behavior data can be used to assess the target user's level of engagement in learning the candidate incorrect questions. After meeting the basic screening conditions for candidate related events, the system can further combine user behavior to determine the final related objects. User behavior can include at least the following: the user clicked on a specific incorrect question on the successful search results page; the user browsed a specific incorrect question for more than a preset dwell time; and the user performed subsequent operations such as viewing the explanation, practicing again, or adding to review for a specific incorrect question. For example, the learning behavior data in this embodiment can specifically include viewing explanation behavior data, practicing again behavior data, adding to review behavior data, and dwell browsing behavior data.
[0054] In this embodiment of the application, the learning intensity data can be data representing the level of engagement of the target user in learning the set of candidate incorrect questions, obtained by quantifying learning behavior data. The learning intensity data can be used to measure the target user's degree of confirmation of the candidate incorrect questions. For example, the learning intensity data in this embodiment of the application can be a quantitative score calculated by assigning different weights to different learning behaviors, wherein clicking behavior can be assigned a value of 0.5, viewing the explanation behavior can be assigned a value of 0.8, and practicing again or adding to the review behavior can be assigned a value of 1.
[0055] In this embodiment of the application, the system can evaluate whether the target user has studied and the depth of study after retrieving the candidate set of incorrect questions based on the target user's learning behavior data in the candidate related events, and thus obtain the learning intensity data of the target user in studying the candidate set of incorrect questions.
[0056] Step 202: Based on the retrieval dimension information of the target retrieval tag, analyze whether the retrieval dimension of the target retrieval tag and each candidate incorrect question in the candidate incorrect question set is consistent, and obtain the dimension consistency data of the target retrieval tag and each candidate incorrect question in the candidate incorrect question set.
[0057] In this embodiment of the application, the retrieval dimension information can be the structured retrieval dimensions and corresponding values contained in the target retrieval tag. The retrieval dimension information can be used to characterize the retrieval conditions of the target user. For example, the retrieval dimension information in this embodiment of the application may specifically include dimensions such as subject, time range, knowledge point, error type, question type, and answer status, as well as their corresponding values.
[0058] In this embodiment, dimensional consistency data can be data representing the degree of consistency between the search dimension values of the target search tag and the corresponding attribute values of the candidate incorrect question. Dimensional consistency data can be used to measure the degree of matching between the structured attributes of the candidate incorrect question and the target search requirements. For example, the dimensional consistency data in this embodiment can specifically be a numerical value determined based on the degree of consistency of dimensions such as subject, grade, question type, knowledge point, and error type. Complete consistency can be recorded as 1, partial consistency can be recorded as 0.5, and inconsistency can be recorded as 0.
[0059] In this embodiment of the application, the system can analyze whether the values of the target search tag and each candidate wrong question in the candidate wrong question set are consistent in each search dimension based on the search dimension information of the target search tag, thereby obtaining the dimension consistency data between the target search tag and each candidate wrong question.
[0060] Step 203: Based on the time information of the candidate related events, analyze the time sequence information of the candidate related events relative to the failed retrieval events to obtain the time sequence priority data of the candidate related events.
[0061] In this embodiment, the time information may be the occurrence time of the candidate related event and the time interval between it and the failed retrieval event. The time information can be used to analyze the degree of correlation between the candidate related event and the failed retrieval event in the time series. For example, the time information in this embodiment may specifically include the timestamp of the candidate related event and the difference between the timestamp of the candidate related event and the timestamp of the failed retrieval event.
[0062] In this embodiment, the time-series priority data can be data characterizing the priority of candidate related events relative to failed search events in a time series. The time-series priority data can be used to measure the strength of intent continuity between successful and failed searches. The time-series priority of the first successful hit after a failed search can refer to the system prioritizing the incorrect question hit by the first successful search as the target related object when the same user experiences multiple consecutive failed searches within a preset time window, and then a successful search occurs for the first time. This is because the first successful search after consecutive failures usually has stronger intent continuity with the aforementioned failed searches. As the number of queries increases, users may gradually modify their search expressions or switch search targets, weakening the correspondence between subsequent successful searches and the original failed searches. Therefore, the first success better represents the first fulfillment of the aforementioned failed search needs. For example, the time-series priority data in this embodiment can specifically be a value determined based on the position of the successful search in the sequence of consecutive failed searches, where the first successful search after consecutive failures can be assigned a value of 1, and subsequent successful searches can be assigned a value of 0.5.
[0063] In this embodiment of the application, the system can analyze the temporal position of the candidate related events relative to the failed retrieval events based on the time information of the candidate related events, and then obtain the temporal priority data.
[0064] Step 204: Based on semantic association, learning intensity data, dimensional consistency data, and temporal priority data, analyze the degree of matching between each candidate wrong question in the candidate wrong question set and the target wrong question retrieval requirement, obtain the relevance score between each candidate wrong question and the target wrong question retrieval requirement, and determine the candidate wrong questions that meet the relevance score condition as the target related questions that match the target wrong question retrieval requirement.
[0065] For the embodiments of this application, the formula for calculating the relevance score can be as shown in Formula 1, where Score represents the relevance score, Simq represents the semantic relevance, Sdim represents dimensional consistency data, Sbeh represents learning intensity data, Stime represents temporal priority data, and b1, b2, b3, and b4 can be corresponding weights. As one implementation, b1=0.4, b2=0.2, b3=0.2, and b4=0.2 can be set. When the Score is greater than or equal to the relevance threshold, the system can identify the corresponding successfully retrieved incorrect question as the target relevance question.
[0066] Score=b1×Simq+b2×Sdim+b3×Sbeh+b4×Stime (Formula 1) For example, the semantic association Simq can be used to characterize the semantic similarity between failed and successful search texts; the dimensional consistency data Sdim can be used to characterize the overlap between failed and successful searches in dimensions such as subject, grade, question type, knowledge point, and error type; the learning intensity data Sbeh can be used to characterize the user's confirmation of the successful search result, with a value of 0.5 for clicking, 0.8 for viewing the explanation, and 1.0 for practicing again or adding to the review; and the temporal priority data Stime can be used to characterize the position of the successful search in a sequence of consecutive failed searches, with a value of 1 for the first success after consecutive failures and 0.5 for subsequent successes.
[0067] In this embodiment of the application, the system can comprehensively analyze the degree of matching between each candidate wrong question in the candidate wrong question set and the target wrong question retrieval requirement based on semantic association, learning intensity data, dimensional consistency data and temporal priority data, calculate the relevance score corresponding to each candidate wrong question, and determine the candidate wrong questions whose relevance scores meet the relevance score conditions as target relevance questions.
[0068] Optionally, when performing the task of "analyzing the tag confidence of the target search tag relative to the target associated question, and determining the target search tags that meet the confidence conditions as the extended tags of the target associated question," the following methods can be used, but are not limited to these: Based on the question text content of the target associated question, analyze the semantic similarity between the target search tag and the question text content to obtain the first semantic similarity data between the target search tag and the target associated question; based on the existing tag set corresponding to the target associated question, analyze the semantic similarity between the target search tag and each existing tag in the existing tag set to obtain the second semantic similarity data between the target search tag and the target associated question; based on the target knowledge points contained in the target associated question, analyze whether the search knowledge points contained in the target search tag are consistent with the target knowledge points, to obtain... The system obtains consistency data on knowledge points between the target search tag and the target associated question; based on the target error type of the target associated question, it analyzes whether the retrieval error type of the target search tag is consistent with the target error type, thus obtaining consistency data on error type between the target search tag and the target associated question; based on historical search records, it analyzes the historical association degree between the target search tag and the existing tags of the target associated question, thus obtaining historical association data between the target search tag and the target associated question; based on the first semantic similarity data, the second semantic similarity data, knowledge point consistency data, error type consistency data, and historical association data, it determines the tag confidence of the target search tag relative to the target associated question, and identifies the target search tag whose tag confidence meets the confidence condition as the extended tag of the target associated question.
[0069] In this embodiment, the question text content can be a collection of text content such as the question stem, options, standard answer, and explanation text of the target-related question. The question text content can be used for semantic similarity analysis with the target retrieval tag. For example, the question text content in this embodiment can specifically include the question stem, solution steps, answer explanation, and explanation text of a math problem.
[0070] In the embodiments of this application, the first semantic similarity data may be data that characterizes the semantic closeness between the target search tag and the text content of the target associated question. The first semantic similarity data may be used to measure whether the target search tag matches the semantic content of the question itself.
[0071] In this embodiment, the existing tag set can be a collection of all tags currently bound to the target-related question. The existing tag set may include standard tags and existing extended tags. For example, in this embodiment, the existing tag set may specifically include standard tags such as "calculation error".
[0072] In the embodiments of this application, the second semantic similarity data may be data that characterizes the semantic similarity between the target retrieval tag and each existing tag in the existing tag set of the target associated question. The second semantic similarity data can be used to measure whether the target retrieval tag has a semantic association with the existing tags of the question.
[0073] In this embodiment of the application, the target knowledge point can be the knowledge point tested by the target-related question, and the target knowledge point can be used to verify whether the knowledge point pointed to by the target search tag is consistent with the content tested by the question.
[0074] In the embodiments of this application, knowledge point consistency data can be data that characterizes whether the search knowledge points contained in the target search tag are consistent with the target knowledge points of the target-related question. Knowledge point consistency data can be used to verify the degree of matching between the tag and the question from the knowledge point dimension.
[0075] In this embodiment, the target error type can be the error cause classification corresponding to the target associated question. The target error type can be used to verify whether the error type implied by the target search tag is consistent with the actual error type of the question. For example, the target error type in this embodiment may specifically include calculation error, question reading error, formula misuse, etc.
[0076] In the embodiments of this application, error type consistency data can be data that characterizes whether the retrieval error type of the target retrieval tag is consistent with the target error type of the target associated question. Error type consistency data can be used to verify the degree of matching between the tag and the question from the error type dimension.
[0077] In this embodiment, historical search records can be historical log records generated by users performing search operations in the intelligent error notebook. These historical search records can be used to statistically analyze the historical association between the target search tag and existing tags of the target related questions. The co-occurrence frequency of the candidate extended tag and the target standard tag in the historical search log can be used to characterize the frequency with which the candidate extended tag and the target standard tag appear together in the historical log; the number of times the same mapping relationship is repeatedly verified can be used to characterize the stability of the same mapping relationship being repeatedly verified in different search sessions.
[0078] In the embodiments of this application, historical association data can be data characterizing the degree of association between the target search tag and the existing tags of the target related question in historical search logs. Historical association data can be used to verify the universality and stability of the tag mapping relationship from a historical statistical perspective. For example, the historical association data in the embodiments of this application may specifically include the normalized result of historical co-occurrence frequency and the normalized result of repeated verification count.
[0079] In this embodiment, the formula for calculating the label confidence score is as shown in Formula 2, where Conf represents the label confidence score, Stext represents the comprehensive semantic similarity between the failed retrieval text and the target associated question content (i.e., the first semantic similarity data in this embodiment), Slabel represents the semantic similarity between the candidate extended label and the existing standard label of the target associated question (i.e., the second semantic similarity data in this embodiment), Sknow represents the knowledge point consistency data, Serr represents the error type consistency data, Sco represents the normalized result of historical co-occurrence frequency, Sver represents the normalized result of repeated verification count, and the historical associated data includes the normalized result of historical co-occurrence frequency and the normalized result of repeated verification count. c1, c2, c3, c4, c5, and c6 can be the corresponding weights, and c1+c2+c3+c4+c5+c6=1.
[0080] Conf = c1 × Stext + c2 × Slabel + c3 × Sknow + c4 × Serr + c5 × Sco + c6 × Sver (Formula 2) Optionally, Stext can be determined by the cosine similarity between the failed retrieval text and the question stem and parsed text vector representations; Slabel can be determined by the semantic vector similarity between the candidate extended label and the existing standard label description; Sknow can be used to characterize whether the knowledge point pointed to by the candidate extended label is consistent with the knowledge point of the target associated question; Serr can be used to characterize whether the error type implied by the candidate extended label is consistent with the error type of the target associated question; Sco can be used to characterize the frequency of co-occurrence of the candidate extended label and the target standard label in the historical log; Sver can be used to characterize the stability of the same mapping relationship being repeatedly verified in different retrieval sessions.
[0081] In this embodiment, the system can set a three-level judgment mechanism: when Conf is greater than or equal to the first confidence threshold and there is no tag conflict, the candidate extended tag can be automatically added to the tag set of the target related question, or a mapping relationship between the candidate extended tag and the target standard tag can be established; when Conf is between the first confidence threshold and the second confidence threshold, the candidate extended tag can be written into the confirmation set, and a decision can be made on whether to formally add it after further verification by more user behaviors; when Conf is lower than the second confidence threshold, the candidate extended tag can be discarded.
[0082] In the embodiments of this application, a label conflict can refer to a situation where the semantics pointed to by a candidate extended label contradict the semantics of the existing labels of the target associated question, or violate the preset label constraint relationship. For example, in the embodiments of this application, a label conflict can specifically be a situation where a candidate extended label corresponds to a correct answer after semantic normalization, but the target associated question already has a label that clearly indicates it is a calculation error, in which case it can be determined as a conflicting label and its addition is rejected.
[0083] In this application embodiment, the tag supplementation method may include the following two: if the semantic consistency between the candidate extended tag and the existing standard tag of the target associated question reaches a threshold, the candidate extended tag can be written into the extended tag set of the target associated question; if a stable mapping relationship is formed between the candidate extended tag and a certain standard tag, an extended tag-standard tag mapping table can be established, and the extended tag can be normalized to the corresponding standard tag during subsequent retrieval.
[0084] Optionally, when performing the task of "selecting candidate related events that are semantically related to failed search events from successful search events based on the semantic association between the target search tags corresponding to failed search events and the successful search tags corresponding to successful search events," the following methods can be used, but are not limited to: monitoring the target user's search behavior in the intelligent error notebook within the target time window after a failed search event; determining the search event corresponding to the target search behavior as a successful search event in response to the target user finding an error question that meets the error question search requirements by changing the target search tags; obtaining all successful search events of the target user within the target time window and the successful search tags of all successful search events, and analyzing the relationship between the target search tags and each successful search event. The semantic similarity between tags is analyzed to obtain the semantic similarity data between the target search tag and each successfully retrieved tag. Keyword similarity between the target search tag and each successfully retrieved tag is also analyzed, yielding keyword similarity data. Based on the target academic attribute information corresponding to the target search tag, the consistency between the target academic attribute information and the candidate academic attribute information corresponding to each successfully retrieved tag is analyzed, yielding academic attribute consistency data. Based on the tag semantic similarity data, keyword similarity data, and academic attribute consistency data, the semantic association between the target search tag and each successfully retrieved tag is determined, and candidate association events that meet the tag association criteria are selected from the successful retrieval events.
[0085] In this embodiment of the application, the retrieval behavior can be various retrieval operations initiated by the target user in the intelligent error notebook. The retrieval behavior can be used to monitor whether the target user has performed a subsequent successful retrieval after a failed retrieval event. For example, the retrieval behavior in this embodiment of the application may specifically include voice input retrieval behavior, text input retrieval behavior, and tag filtering retrieval behavior, etc.
[0086] In this embodiment, the target retrieval behavior can be a retrieval behavior initiated by a target user after a failed retrieval event by changing the search tag. The target retrieval behavior can be used to determine a successful retrieval event. For example, in this embodiment, the target retrieval behavior can specifically be a retrieval behavior initiated by a target user after changing the search tag from "calculation error" to "calculation error".
[0087] In this embodiment of the application, the successful retrieval tag can be the retrieval tag corresponding to the successful retrieval event, and the successful retrieval tag can be used for semantic association analysis with the target retrieval tag. For example, in this embodiment of the application, the successful retrieval tag can specifically be the tag dimension value parsed from the successful retrieval event.
[0088] In this embodiment of the application, the tag semantic similarity data can be data that characterizes the degree of textual semantic similarity between the target retrieval tag and the successfully retrieved tag. The tag semantic similarity data can be used to measure the degree of association between two tags in semantic expression.
[0089] In the embodiments of this application, keyword similarity data can be data that characterizes the degree of keyword overlap between the target search tag and the successfully searched tag. Keyword similarity data can be used to supplement the measurement of the correlation between two tags from the keyword dimension.
[0090] In this embodiment, the target academic attribute information may be academic-related attribute information corresponding to the target search tag, and the target academic attribute information can be used to verify consistency with the corresponding attributes of the successfully searched tag. For example, the target academic attribute information in this embodiment may specifically include subject information and grade information.
[0091] In this embodiment of the application, the candidate academic attribute information may be academic-related attribute information corresponding to a successfully retrieved tag, and the candidate academic attribute information may be used for consistency comparison with the target academic attribute information. For example, the candidate academic attribute information in this embodiment may specifically include subject information and grade information corresponding to the successful retrieval event.
[0092] In this embodiment of the application, academic attribute consistency data can be data that characterizes the degree of consistency between the target academic attribute information of the target retrieval tag and the candidate academic attribute information of the successfully retrieved tag. Academic attribute consistency data can be used to verify the degree of correlation between two retrieval events from the perspective of academic attributes.
[0093] For the embodiments of this application, the formula for calculating semantic association can be as shown in Formula 3, where Simq can represent semantic association, Ssem can represent tag semantic similarity data, Skey can represent keyword similarity data, Sstr can represent academic attribute consistency data, and a1, a2, and a3 can be corresponding weights, with a1+a2+a3=1. As an optional approach, a1=0.6, a2=0.2, and a3=0.2 can be set.
[0094] Simq=a1×Ssem+a2×Skey+a3×Sstr (Formula 3) For example, the semantic similarity data Ssem of tags can be obtained by calculating the cosine similarity between the vector representations of failed and successful search texts; the keyword similarity Skey can be obtained by the ratio of the number of common keywords to the total number of keywords; the academic attribute consistency data Sstr can be determined based on the degree of consistency of structured dimensions such as subject, grade, question type, and knowledge point. Complete consistency can be recorded as 1, partial consistency can be recorded as 0.5, and inconsistency can be recorded as 0.
[0095] For the embodiments of this application, the system can filter candidate related events according to the following conditions: failed retrieval and successful retrieval can correspond to the same user; failed retrieval and successful retrieval can occur in the same session or in consecutive sessions on the same terminal; the time interval between failed retrieval and successful retrieval can be less than a preset duration threshold; the two retrievals can meet at least one consistency condition in the dimensions of subject, grade or question type; the semantic similarity of the query texts of the two retrievals can be greater than the first similarity threshold.
[0096] Optionally, when executing the process of "obtaining the target user's target error retrieval needs in the intelligent error notebook, and the failed retrieval events of the target user's error retrieval based on the target retrieval tags corresponding to the target error retrieval needs," the following methods may be used, but are not limited to: obtaining the retrieval language data of the target user's error retrieval in the intelligent error notebook; performing semantic analysis on the retrieval language data to obtain the target user's target error retrieval needs; identifying the retrieval intent of the target error retrieval needs to obtain the target user's target retrieval intent; based on the target retrieval intent, analyzing the target user's retrieval time intent, retrieval subject intent, and error type intent to obtain the target retrieval tags corresponding to the target retrieval intent; performing error retrieval based on the target retrieval tags, and determining the retrieval event where the target user does not find any error retrieval questions that meet the error retrieval needs as a failed retrieval event.
[0097] In this embodiment, the retrieval language data can be the raw language data input by the target user when initiating a wrong question retrieval. The retrieval language data can be used for semantic analysis to extract the target wrong question retrieval needs. For example, the retrieval language data in this embodiment can specifically include voice input data and text input data, such as "Please help me find the wrong math calculation questions from last week."
[0098] In the embodiments of this application, retrieval intent recognition can be a process of classifying and extracting slots from the target incorrect question retrieval needs. Retrieval intent recognition can be used to convert natural language expressions into structured retrieval conditions. Retrieval intent recognition can include two steps: intent recognition and slot extraction. Intent recognition can employ a classification model, including but not limited to a classification model fine-tuned using Bidirectional Encoder Representations from Transformers (BERT) or a robustly optimized BERT pre-training approach (RoBERTa). Entity extraction can employ a Named Entity Recognition (NER) model, including but not limited to a NER model composed of a BERT encoding layer and a Conditional Random Field (CRF) decoding layer. Domain Specific Language (DSL) generation can employ rule templates, including but not limited to using regular expressions with slot filling.
[0099] In this embodiment of the application, the system can perform intent recognition and slot extraction on the retrieved language data, extract various dimensions such as subject, time, and tags from the query, and generate a structured search condition DSL.
[0100] In this embodiment of the application, the search time intent can be the time range constraint intent expressed by the target user in the search request. The search time intent can be used to parse and obtain the time dimension value in the search conditions. For example, in this embodiment of the application, the search time intent can specifically be the time range of the most recent 7 days corresponding to last week.
[0101] In this embodiment, the subject search intent can be the subject constraint intent expressed by the target user in their search request. The subject search intent can be used to parse and obtain the subject dimension values in the search conditions. For example, in this embodiment, the subject search intent can specifically be mathematics corresponding to the mathematics discipline.
[0102] In this embodiment of the application, the error type intent can be the error type or error type constraint intent expressed by the target user in the search request. The error type intent can be used to parse and obtain the tag dimension values in the search conditions. For example, in this embodiment of the application, the error type intent can specifically be a calculation error tag corresponding to a calculation error.
[0103] Optionally, but not limited to, the following methods may be used: in response to the addition of a new incorrect question in the intelligent error notebook, obtain multi-dimensional feature information of the new incorrect question; based on the multi-dimensional feature information, perform feature analysis on historical incorrect questions in the intelligent error notebook, and select at least one candidate historical incorrect question from the intelligent error notebook that has similar features to the new incorrect question; obtain a set of candidate incorrect question tags for at least one candidate historical incorrect question, analyze the matching degree between each candidate incorrect question tag in the candidate incorrect question tag set and the tag of the new incorrect question, and determine at least one candidate tag that matches the new incorrect question from the candidate incorrect question tag set; analyze the tag confidence of at least one candidate tag relative to the new incorrect question, and determine the candidate tags that meet the confidence conditions as new extended tags for the new incorrect question, and use the new extended tags to retrieve the new incorrect question.
[0104] In this embodiment, the newly added incorrect questions can be newly entered incorrect question data into the intelligent incorrect question notebook. These newly added incorrect questions can be used to automatically tag questions based on historical tag knowledge when there is no manual annotation available. For example, the newly added incorrect questions in this embodiment can specifically be incorrect questions newly photographed and entered by the user, or incorrect question data newly imported by the system.
[0105] In the embodiments of this application, when new incorrect question data is generated, the system can first extract the basic information of the incorrect question. The basic information may include at least one of the following: question stem, options, standard answer, explanation text, subject, grade, and student's answer result.
[0106] In this embodiment, the multi-dimensional feature information can be a set of multi-dimensional features obtained by processing the basic information of newly added incorrect questions through structured parsing and semantic representation. This multi-dimensional feature information can be used for feature similarity matching with historical incorrect questions. Structured parsing can include at least identifying question types, extracting knowledge points, identifying the examination object, determining error types, and extracting problem-solving step features or answer deviation features. Error types can include calculation errors, misreading the question, conceptual confusion, misuse of formulas, and missing steps; problem-solving step features can be used to characterize typical deviations students make during the problem-solving process, such as incorrect formulation, omission of units, and incorrect symbol writing. Through structured parsing, the original question content can be converted into a comparable, filterable, and computable set of structured features.
[0107] In this embodiment of the application, candidate historical incorrect questions can be historical incorrect questions selected from the historical incorrect question database of the intelligent incorrect question notebook that have similar features to the newly added incorrect question. Candidate historical incorrect questions can be used to provide the source of candidate tags. After obtaining the structured features of the new incorrect question, the system can retrieve candidate incorrect questions similar to the new incorrect question from the historical incorrect question database, and generate candidate tags by comprehensively considering the textual semantic similarity between the new incorrect question and the historical incorrect question, the structured consistency between the new incorrect question and the historical incorrect question in terms of question type, knowledge point, and error type, the frequency of occurrence of existing tags for similar historical incorrect questions, and the historical mapping relationship between extended tags and standard tags.
[0108] In this embodiment, the candidate incorrect question tag set can be a collection of all tags already bound to candidate historical incorrect questions. This candidate incorrect question tag set can be used to generate candidate tags for newly added incorrect questions. For example, the candidate incorrect question tag set in this embodiment may specifically include standard tags and extended tags.
[0109] In the embodiments of this application, the tag matching degree can be a quantitative indicator that characterizes the degree of fit between candidate wrong question tags and newly added wrong questions. The tag matching degree can be used to filter out candidate tags that match newly added wrong questions from the candidate wrong question tag set.
[0110] In this embodiment, the newly added extended tags can be automatically generated tags for newly added incorrect questions. These extended tags can improve the retrieval performance during the cold start phase when newly added incorrect questions have not yet been manually labeled. For candidate tags, the system can further combine tag confidence and tag conflict rules to filter the final tags, and write the standard tags and extended tags that meet the conditions into the new incorrect question tag set. In this way, even when new incorrect questions have not yet been manually labeled, the system can automatically tag them based on the tag knowledge accumulated from historical failed searches, thereby improving the retrieval performance during the cold start phase.
[0111] Compared with existing technologies, this embodiment improves the accuracy of target related question selection by analyzing the matching degree and determining the relevance score based on learning behavior data, dimensional consistency data, and time-series priority data; it improves the reliability of extended tag generation by determining tag confidence based on question text semantic similarity, existing tag semantic similarity, knowledge point consistency, error type consistency, and historical related data; it improves the accuracy of candidate related event selection by monitoring retrieval behavior within the target time window and determining semantic relationships by combining tag semantic similarity, keyword similarity, and academic attribute consistency; it transforms user natural language retrieval needs into structured retrieval conditions by performing semantic analysis and retrieval intent recognition on retrieval language data, parsing target retrieval tags, and identifying failed retrieval events; and it automatically generates new extended tags for new incorrect questions by filtering candidate historical incorrect questions based on multi-dimensional feature information of newly added incorrect questions and analyzing tag matching degree and tag confidence, thereby improving the retrieval accuracy of new incorrect questions.
[0112] Furthermore, as Figure 1 and Figure 2 The specific implementation of the method shown in this embodiment provides a device for determining incorrect question tags, such as... Figure 3 As shown, the device includes: an acquisition module 31, a filtering module 32, an analysis module 33, and a determination module 34.
[0113] The acquisition module 31 is configured to acquire the target user's target error retrieval needs in the intelligent error notebook, and the failed retrieval events of the target user's error retrieval according to the target retrieval tags corresponding to the target error retrieval needs. The failed retrieval event is a retrieval event in which no error retrieval meeting the error retrieval needs is found. The filtering module 32 is configured to filter candidate related events that are semantically related to the failed search events from the successful search events based on the semantic relationship between the target search tag corresponding to the failed search event and the successful search tag corresponding to the successful search event. The successful search event is the search event that finds the wrong question that meets the wrong question search requirements within the target time window after the failed search event by changing the search tag. Analysis module 33 is configured to obtain a set of candidate incorrect questions for candidate related events, analyze the matching degree between each candidate incorrect question in the set and the target incorrect question retrieval requirement based on the target user's operation behavior data in the candidate related events, determine the relevance score between each candidate incorrect question and the target incorrect question retrieval requirement, and filter out the target related questions that match the target incorrect question retrieval requirement from the set of candidate incorrect questions based on the relevance score. The determination module 34 is configured to analyze the tag confidence of the target retrieval tag relative to the target associated question, and determine the target retrieval tag that meets the confidence condition as the extended tag of the target associated question. The extended tag is used to retrieve the target associated question.
[0114] In some examples of this embodiment, the analysis module 33 is specifically configured to: evaluate whether the target user learns after retrieving the candidate incorrect question set based on the target user's learning behavior data in the candidate related events, and obtain learning intensity data of the target user learning the candidate incorrect question set; analyze whether the retrieval dimension of the target retrieval tag and each candidate incorrect question in the candidate incorrect question set is consistent based on the retrieval dimension information of the target retrieval tag, and obtain dimension consistency data of the target retrieval tag and each candidate incorrect question in the candidate incorrect question set; analyze the temporal sequence information of the candidate related events relative to the failed retrieval events based on the time information of the candidate related events, and obtain temporal priority data of the candidate related events; and analyze the degree of matching between each candidate incorrect question in the candidate incorrect question set and the target incorrect question retrieval requirement based on the semantic relationship, learning intensity data, dimension consistency data, and temporal priority data, and obtain the relevance score of each candidate incorrect question and the target incorrect question retrieval requirement, and determine the candidate incorrect questions that meet the relevance score condition as target related questions that match the target incorrect question retrieval requirement.
[0115] In some examples of this embodiment, the determining module 34 is specifically configured to: analyze the semantic similarity between the target search tag and the question text content based on the question text content of the target associated question, to obtain first semantic similarity data between the target search tag and the target associated question; analyze the semantic similarity between the target search tag and each existing tag in the existing tag set corresponding to the target associated question, to obtain second semantic similarity data between the target search tag and the target associated question; analyze whether the search knowledge points contained in the target search tag are consistent with the target knowledge points based on the target knowledge points contained in the target associated question, to obtain knowledge point consistency data between the target search tag and the target associated question; and, based on the target... To determine the target error type for the associated question, analyze whether the error type of the target search tag matches the target error type, thus obtaining consistency data between the target search tag and the target associated question's error type. Based on historical search records, analyze the historical association degree between the target search tag and the existing tags of the target associated question, thus obtaining historical association data between the target search tag and the target associated question. Based on the first semantic similarity data, the second semantic similarity data, knowledge point consistency data, error type consistency data, and historical association data, determine the tag confidence of the target search tag relative to the target associated question, and identify the target search tag whose tag confidence meets the confidence condition as the extended tag of the target associated question.
[0116] In some examples of this embodiment, the filtering module 32 is specifically configured to monitor the target user's search behavior in the intelligent error notebook within a target time window after a failed search event; in response to the target user finding an error question that meets the error question search requirements by changing the search tags, the search event corresponding to the target search behavior is determined as a successful search event; all successful search events of the target user within the target time window and the successful search tags of all successful search events are obtained, and the textual semantic similarity between the target search tag and each successful search tag is analyzed to obtain the tag semantic similarity between the target search tag and each successful search tag. Data analysis includes: analyzing keyword similarity between the target search tag and each successfully retrieved tag; analyzing the consistency between the target academic attribute information and the candidate academic attribute information corresponding to each successfully retrieved tag; and determining the semantic association between the target search tag and each successfully retrieved tag based on tag semantic similarity data, keyword similarity data, and academic attribute consistency data. Finally, it involves selecting candidate association events that meet the tag association criteria from the successful retrieval events.
[0117] In some examples of this embodiment, the acquisition module 31 is specifically configured to acquire the retrieval language data of the target user in the intelligent error notebook for searching for incorrect questions; perform semantic analysis on the retrieval language data to obtain the target user's target error search needs; identify the search intent of the target error search needs to obtain the target user's target search intent; based on the target search intent, analyze the target user's search time intent, search subject intent, and error question type intent to obtain the target search tags corresponding to the target search intent; perform error search according to the target search tags, and determine the search event in which the target user does not find an error question that meets the error search needs as a failed search event.
[0118] In some examples of this embodiment, the determining module 34 is specifically configured to, in response to the addition of a new incorrect question in the intelligent error notebook, obtain multi-dimensional feature information of the new incorrect question; based on the multi-dimensional feature information, perform feature analysis on the historical incorrect questions in the intelligent error notebook, and select at least one candidate historical incorrect question from the intelligent error notebook that has similar features to the new incorrect question; obtain a set of candidate incorrect question tags for at least one candidate historical incorrect question, analyze the matching degree between each candidate incorrect question tag in the candidate incorrect question tag set and the tag of the new incorrect question, and determine at least one candidate tag that matches the new incorrect question from the candidate incorrect question tag set; analyze the tag confidence of at least one candidate tag relative to the tag of the new incorrect question, and determine the candidate tag that meets the confidence condition as the new extended tag of the new incorrect question, and the new extended tag is used to retrieve the new incorrect question.
[0119] It should be noted that other corresponding descriptions of the functional units involved in the error label determination device provided in this embodiment can be found in [reference needed]. Figure 1 and Figure 2 The corresponding descriptions in [the document] will not be repeated here.
[0120] Based on the above, Figure 1 and Figure 2 Accordingly, this embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. Figure 1 and Figure 2 The method shown.
[0121] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.
[0122] like Figure 4 The diagram shown is a hardware structure schematic of an electronic device according to the present invention, comprising: At least one processor 401; and, Memory 402 is communicatively connected to at least one processor 401; wherein, The memory 402 stores instructions that can be executed by at least one processor, such that the at least one processor can perform the error label determination method described above.
[0123] Figure 4 Take a processor 401 as an example.
[0124] The electronic device may also include an input device 403 and an output device 404.
[0125] The processor 401, memory 402, input device 403, and output device 404 can be connected via a bus or other means. Figure 4 Taking the example of a connection between China and Israel via a bus.
[0126] Memory 402, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the error label determination method in the embodiments of this application, for example, Figure 1 and Figure 2The method flow is shown. The processor 401 executes various functional applications and data processing by running non-volatile software programs, instructions, and modules stored in the memory 402, thereby implementing the error label determination method in the above embodiments.
[0127] The memory 402 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the error tagging method. Furthermore, the memory 402 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 402 may optionally include memory remotely located relative to the processor 401, and these remote memories may be connected via a network to the apparatus performing the error tagging method. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
[0128] The input device 403 can receive user clicks and generate signal inputs related to user settings and function control for the error label determination method. The output device 404 may include a display device such as a screen.
[0129] One or more modules are stored in memory 402, and when run by one or more processors 401, the error label determination method in any of the above method embodiments is executed.
[0130] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.
[0131] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.
[0132] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.
[0133] Through the above description of the implementation methods, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms, or it can be implemented by hardware. By applying the solution of this embodiment, compared with the existing technology, this embodiment filters candidate related events based on the semantic relationship between the target retrieval tag and the successful retrieval tag, establishes a relationship between failed retrieval and subsequent successful retrieval, and provides the source of target questions for tag supplementation; by analyzing the matching degree between candidate wrong questions and target wrong question retrieval needs based on operation behavior data and determining the relevance score, the target related questions matching the failed retrieval needs are accurately filtered from the candidate wrong question set; by analyzing the tag confidence of the target retrieval tag relative to the target related question and determining the extended tag, the tags used by the user in the failed retrieval are supplemented as extended tags of the target related question, realizing the automatic supplementation of extended tags, improving the hit rate of subsequent similar retrievals, meeting the user's wrong question retrieval needs, and thus improving the user's review efficiency and knowledge point consolidation effect; by using learning behavior data, dimensional consistency data and time sequence priority... The system analyzes data matching to determine relevance scores, improving the accuracy of target related question selection. It determines tag confidence based on question text semantic similarity, existing tag semantic similarity, knowledge point consistency, error type consistency, and historical related data, enhancing the reliability of extended tag generation. It monitors retrieval behavior within the target time window and combines tag semantic similarity, keyword similarity, and academic attribute consistency to determine semantic relationships, improving the accuracy of candidate related event selection. It performs semantic analysis and retrieval intent recognition on retrieval language data to parse target retrieval tags and identify failed retrieval events, transforming users' natural language retrieval needs into structured retrieval conditions. Finally, it filters candidate historical incorrect questions based on multi-dimensional feature information of newly added incorrect questions and analyzes tag matching and tag confidence, automatically generating new extended tags for new incorrect questions, improving the retrieval accuracy of new incorrect questions.
[0134] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.
[0135] The above are merely specific embodiments of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to these embodiments, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.
Claims
1. A method for determining incorrect question tags, characterized in that, include: The system obtains the target user's target error retrieval needs in the intelligent error notebook, as well as the failed retrieval events of the target user's error retrieval according to the target retrieval tags corresponding to the target error retrieval needs. The failed retrieval events are retrieval events in which no error questions matching the error retrieval needs are found. Based on the semantic association between the target search tag corresponding to the failed search event and the successful search tag corresponding to the successful search event, candidate association events that are semantically related to the failed search event are selected from the successful search events. The successful search event is a search event in which the wrong question that meets the wrong question search requirements is retrieved by changing the search tag within the target time window after the failed search event. Obtain a set of candidate incorrect questions for the candidate related events. Based on the target user's operation behavior data in the candidate related events, analyze the degree of matching between each candidate incorrect question in the set of candidate incorrect questions and the target incorrect question retrieval requirement. Determine the relevance score between each candidate incorrect question and the target incorrect question retrieval requirement. Based on the relevance score, select target related questions that match the target incorrect question retrieval requirement from the set of candidate incorrect questions. Analyze the tag confidence of the target retrieval tag relative to the target associated question, and determine the target retrieval tag that meets the confidence condition as the extended tag of the target associated question. The extended tag is used to retrieve the target associated question.
2. The method according to claim 1, characterized in that, The process of obtaining a set of candidate incorrect questions for the candidate related events involves analyzing the matching degree between each candidate incorrect question in the set and the target incorrect question retrieval request based on the target user's operational behavior data in the candidate related events, determining the relevance score between each candidate incorrect question and the target incorrect question retrieval request, and filtering target related questions that match the target incorrect question retrieval request from the set based on the relevance score. This includes: Based on the learning behavior data of the target user in the candidate related events, evaluate whether the target user learns after retrieving the candidate set of incorrect questions, and obtain the learning intensity data of the target user learning the candidate set of incorrect questions; Based on the retrieval dimension information of the target retrieval tag, analyze whether the retrieval dimension of the target retrieval tag and each candidate wrong question in the candidate wrong question set are consistent, and obtain the dimension consistency data of the target retrieval tag and each candidate wrong question in the candidate wrong question set; Based on the time information of the candidate related events, the temporal sequence information of the candidate related events relative to the failed retrieval events is analyzed to obtain the temporal priority data of the candidate related events; Based on the semantic association, the learning intensity data, the dimensional consistency data, and the temporal priority data, the matching degree between each candidate wrong question in the candidate wrong question set and the target wrong question retrieval requirement is analyzed to obtain the relevance score between each candidate wrong question and the target wrong question retrieval requirement, and the candidate wrong questions that meet the relevance score condition are determined as target related questions that match the target wrong question retrieval requirement.
3. The method according to claim 1, characterized in that, The step of analyzing the tag confidence of the target search tag relative to the target associated question, and determining the target search tag that meets the confidence condition as the extended tag of the target associated question, includes: Based on the text content of the target-related question, the semantic similarity between the target search tag and the text content of the question is analyzed to obtain the first semantic similarity data between the target search tag and the target-related question; Based on the existing tag set corresponding to the target associated question, the semantic similarity between the target retrieval tag and each existing tag in the existing tag set is analyzed to obtain the second semantic similarity data between the target retrieval tag and the target associated question; Based on the target knowledge points contained in the target-related questions, analyze whether the search knowledge points contained in the target search tags are consistent with the target knowledge points, and obtain the consistency data of the knowledge points between the target search tags and the target-related questions; Based on the target error type of the target-related question, analyze whether the retrieval error type of the target retrieval tag is consistent with the target error type, and obtain the consistency data of the error type between the target retrieval tag and the target-related question; Based on historical search records, the historical correlation between the target search tag and the existing tags of the target associated questions is analyzed to obtain historical correlation data between the target search tag and the target associated questions; Based on the first semantic similarity data, the second semantic similarity data, the knowledge point consistency data, the error type consistency data, and the historical association data, the tag confidence of the target retrieval tag relative to the target associated question is determined, and the target retrieval tag whose tag confidence satisfies the confidence condition is determined as the extended tag of the target associated question.
4. The method according to claim 1, characterized in that, The step of filtering candidate related events that are semantically related to the failed search events from the successful search events, based on the semantic association between the target search tag corresponding to the failed search event and the successful search tag corresponding to the successful search event, includes: Within the target time window following the failed retrieval event, the retrieval behavior of the target user in the intelligent error notebook is monitored; In response to the target user finding a wrong question that matches the wrong question retrieval requirement through the target retrieval behavior of changing the search tags, the retrieval event corresponding to the target retrieval behavior is determined as the successful retrieval event; Obtain all successful search events of the target user within the target time window and all successful search tags of the successful search events, analyze the text semantic similarity between the target search tag and each of the successful search tags, and obtain the tag semantic similarity data between the target search tag and each of the successful search tags; Analyze the keyword similarity between the target search tag and each of the successful search tags, and obtain the keyword similarity data between the target search tag and each of the successful search tags; Based on the target academic attribute information corresponding to the target search tag, analyze the consistency between the target academic attribute information and the candidate academic attribute information corresponding to each successful search tag, and the academic attribute consistency data between the target search tag and each successful search tag; Based on the semantic similarity data of the tags, the keyword similarity data, and the academic attribute consistency data, the semantic association between the target retrieval tag and each of the successfully retrieved tags is determined, and candidate association events that meet the tag association conditions are selected from the successful retrieval events.
5. The method according to claim 1, characterized in that, The acquisition of the target user's target error retrieval needs in the intelligent error notebook, and the failed retrieval events of the target user performing error retrieval according to the target retrieval tags corresponding to the target error retrieval needs, includes: Obtain the retrieval language data of the target user when searching for incorrect questions in the intelligent error notebook, perform semantic analysis on the retrieval language data, and obtain the target user's target error search needs; The target user's target search intent is obtained by identifying the target user's search intent for the target incorrect question retrieval request. Based on the target search intent, the target user's search time intent, search subject intent, and incorrect question type intent are analyzed to obtain the target search tags corresponding to the target search intent. The wrong question is retrieved based on the target search tag, and the search event in which the target user does not find a wrong question that meets the wrong question search requirements is determined as the failed search event.
6. The method according to any one of claims 1 to 5, characterized in that, The method further includes: In response to the addition of a new incorrect question in the intelligent error notebook, the multi-dimensional feature information of the new incorrect question is obtained; Based on the multi-dimensional feature information, feature analysis is performed on the historical wrong questions in the intelligent wrong question notebook, and at least one candidate historical wrong question with similar features to the newly added wrong question is selected from the intelligent wrong question notebook; Obtain a set of candidate error labels for the at least one candidate historical error question, analyze the matching degree between each candidate error label in the candidate error label set and the label of the newly added error question, and determine at least one candidate label that matches the newly added error question from the candidate error label set; Analyze the confidence level of the at least one candidate label relative to the label of the newly added incorrect question, and determine the candidate label that meets the confidence level condition as the new extended label of the newly added incorrect question. The new extended label is used to retrieve the newly added incorrect question.
7. A device for identifying incorrect question tags, characterized in that, include: The acquisition module is configured to acquire the target user's target error retrieval needs in the intelligent error notebook, and the failed retrieval events of the target user's error retrieval according to the target retrieval tags corresponding to the target error retrieval needs. The failed retrieval events are retrieval events in which no error questions matching the error retrieval needs are found. The filtering module is configured to filter candidate related events that are semantically related to the failed search event from the successful search events based on the semantic relationship between the target search tag corresponding to the failed search event and the successful search tag corresponding to the successful search event. The successful search event is a search event in which a wrong question that meets the wrong question search requirements is retrieved by changing the search tag within a target time window after the failed search event. The analysis module is configured to obtain a set of candidate incorrect questions for the candidate related events, analyze the matching degree between each candidate incorrect question in the set and the target incorrect question retrieval requirement based on the target user's operation behavior data in the candidate related events, determine the relevance score between each candidate incorrect question and the target incorrect question retrieval requirement, and filter out the target related questions that match the target incorrect question retrieval requirement from the set of candidate incorrect questions based on the relevance score. The determination module is configured to analyze the tag confidence of the target retrieval tag relative to the target associated question, and determine the target retrieval tag that meets the confidence condition as the extended tag of the target associated question, the extended tag being used to retrieve the target associated question.
8. 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 method of any one of claims 1 to 6.
9. An electronic device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 6.
10. A computer program product, the computer program product comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 6.
