Question and answer learning method and device based on artificial intelligence and storage medium

Abstract

The invention relates to the technical field of question and answer data processing, in particular to a question and answer learning method and device based on artificial intelligence and a storage medium. The method comprises the following steps: acquiring question and answer data, user data and question tracking data, and constructing a question and answer database; performing feature extraction on the problem tracking data and the user data to obtain a problem keyword set and an aging feature value; based on the question keyword set and the question and answer data, analyzing semantic similarity and keyword matching degree to extract effective question and answer content; analyzing a user discrimination factor based on the effective question and answer content, and constructing a quality evaluation model based on the user discrimination factor and the aging characteristic value to obtain an answer score; extracting candidate answers according to the answer scores, and pushing the candidate answers to the user; and collecting feedback data of the user to the candidate answers, and updating the quality evaluation model based on the feedback data. According to the invention, accurate matching and pushing of questions and user answer contents are realized.

Description

Technical Field

[0001] This invention relates to the field of question-and-answer data processing technology, and in particular to a question-and-answer learning method, apparatus and storage medium based on artificial intelligence. Background Technology

[0002] Question-answering learning systems originate from traditional information retrieval and knowledge reasoning techniques. Early systems relied on keyword matching to retrieve information, later evolving to include semantic analysis and rule-based reasoning. These technologies use logical rules and pattern matching to understand user questions and generate answers.

[0003] Current question-answering systems generally suffer from the following shortcomings: Lack of timeliness: Traditional methods fail to effectively track question modification records, resulting in outdated answers for time-sensitive questions such as technical ones; Insufficient personalization: User profiles and question matching rely solely on basic tags, ignoring dynamic factors such as activity level and professional consistency; Static evaluation model: Answer quality depends on preset rules and cannot be dynamically optimized based on feedback. When user adoption rates fluctuate, the system cannot automatically adjust weights, causing recommendation results to continuously deviate from actual needs; Noise interference: Keyword extraction fails to filter secondary words, causing semantic analysis to deviate from the core topic and increasing the probability of pushing invalid answers. Summary of the Invention

[0004] The purpose of this invention is to provide an artificial intelligence-based question-answering learning method, apparatus, and storage medium to solve at least one of the problems existing in the prior art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] An AI-based question-answering learning method includes:

[0007] Collect question-and-answer data, user data, and question tracking data to build a question-and-answer database;

[0008] Feature extraction is performed on issue tracking data and user data to obtain a set of issue keywords and timeliness feature values;

[0009] Based on the set of question keywords and question-answer data, semantic similarity and keyword matching are analyzed to extract effective question-answer content;

[0010] Based on the analysis of effective question and answer content, user discrimination factors are analyzed, and a quality assessment model is constructed based on user discrimination factors and timeliness feature values ​​to obtain answer scores;

[0011] Based on the answer rating, candidate answers are extracted and pushed to users;

[0012] Collect user feedback data on candidate answers and update the quality assessment model based on the feedback data.

[0013] Furthermore, the time interval between the current time and the last modification time corresponding to the question ID is used as the decay duration, and the question-answer decay factor is analyzed based on the decay duration.

[0014] Furthermore, the question and answer content are segmented into words to obtain multiple sets of question and answer terms. The question and answer terms are then matched with preset secondary keywords. Words that are the same as the preset secondary keywords in the question and answer terms are removed, and the removed question and answer terms are used as the question keyword set.

[0015] The number of times the modification time corresponding to the statistical problem ID is within 30 days before the current time is used as the timeliness count. The timeliness characteristic value is analyzed based on the timeliness count and the number of modifications. The expression of the timeliness characteristic value is: T(i)=α(i)×N1(i) / N2(i), where T(i) represents the timeliness characteristic value, N1(i) represents the timeliness count, and N2(i) represents the number of modifications.

[0016] Furthermore, the semantic similarity between the answer content and the standard answer is analyzed, and the question keywords are matched with related keywords. The number of keywords that are the same in the question keywords and related keywords is counted as the matching number. Based on the matching number and the question keywords and related keywords, the keyword matching degree is analyzed.

[0017] The answer content is extracted based on semantic similarity and keyword matching degree. Answer content that meets the requirements of semantic similarity greater than or equal to the similarity threshold and keyword matching degree greater than or equal to the matching degree threshold is selected as valid answer content.

[0018] Furthermore, the adoption rate factor is analyzed based on the adoption rate and question-answer decay factor; the number of times a user logs in within a month is counted as the recent login count, the time interval between the current time and the user's last active time is counted as the user's validity period, the number of times a user answers different question types is counted as the user's answer type count, the number of answers in the user's historical answer records is counted as the user's total answer count, the proportion of the user's answer type count for the question type corresponding to the current analysis question ID in the user's total answer count is counted as the type consistency parameter, and the credibility factor is analyzed based on the recent login count, user validity period, and type consistency parameter.

[0019] Furthermore, the adoption rate factor, credibility factor, and timeliness feature value are weighted to obtain the answer score. The expression for the answer score is: W = w1 × S(j) + w2 × U(i,j) + w3 × T(i), where W represents the answer score, w1 represents the adoption rate weight, w2 represents the credibility weight, w3 represents the timeliness weight, S(j) represents the adoption rate factor, and U(i,j) represents the credibility factor.

[0020] Furthermore, the answer scores are sorted in descending order, and the answer content corresponding to the top X scores is taken as candidate answers, where X represents the extracted parameters.

[0021] Furthermore, when the feedback data indicates acceptance, the analysis process for the acceptance rate factor in the next analysis of the current issue is updated;

[0022] The model optimization rules are triggered based on the adoption rate and type consistency parameters. These rules include: when a user's adoption rate decreases by more than Q1 within a week, the statistical time range for the number of timeliness events is increased to 60 days, and the timeliness feature value is recalculated; when a user's type consistency parameter fluctuates by more than Q2 within a week, the consistency weight is reduced, and the activity weight and timeliness weight are increased; where Q1 represents the first amplitude threshold and Q2 represents the second amplitude threshold.

[0023] On the other hand, the present invention also provides an artificial intelligence-based question-answering learning device, comprising:

[0024] The data collection module is used to collect question-and-answer data, user data, and question tracking data, and to build a question-and-answer database.

[0025] The feature analysis module is used to extract features from issue tracking data and user data to obtain a set of issue keywords and timeliness feature values;

[0026] The similarity matching module is used to analyze semantic similarity and keyword matching based on the question keyword set and question-answer data in order to extract effective question-answer content;

[0027] The quality assessment module is used to analyze user discrimination factors based on effective question and answer content, and to build a quality assessment model based on user discrimination factors and timeliness feature values ​​to obtain answer scores;

[0028] The candidate push module is used to extract candidate answers based on answer scores and push them to users.

[0029] The feedback analysis module is used to collect user feedback data on candidate answers and update the quality assessment model based on the feedback data.

[0030] On the other hand, the present invention also provides a storage medium characterized in that it stores instructions that, when run on a computer, cause the computer to execute the artificial intelligence-based question-answering learning method as described in any of the preceding claims.

[0031] The beneficial effects of this invention are as follows: By integrating multi-dimensional data and dynamic evaluation mechanisms, the system achieves accurate matching and timely response in the question-and-answer system. By combining semantic analysis and user profiling, it ensures that the answers are deeply aligned with the question requirements and user background. By dynamically quantifying the freshness of content through question modification frequency and decay factors, it prioritizes pushing updated and iterated effective answers. Based on user feedback, it optimizes the weight strategy in real time, enabling the model to adapt to changes in requirements. Overall, the system significantly reduces the cost of manual screening and improves the efficiency of reaching high-value answers, making it particularly suitable for time-sensitive scenarios such as technical support and knowledge bases. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a flowchart of the question-answering learning method based on artificial intelligence in this embodiment.

[0034] Figure 2 This is a flowchart of the data feature extraction method in this embodiment.

[0035] Figure 3 This is a flowchart of the method for constructing the quality assessment model in this embodiment.

[0036] Figure 4 This is a schematic diagram of the artificial intelligence-based question-answering learning device in this embodiment. Detailed Implementation

[0037] To more clearly illustrate the present invention, the following description, in conjunction with preferred embodiments and accompanying drawings, further explains the invention. Similar components in the drawings are indicated by the same reference numerals. Those skilled in the art should understand that the specific description below is illustrative rather than restrictive and should not be construed as limiting the scope of protection of the present invention.

[0038] It should be noted that although the terms first, second, third, etc., may be used in the embodiments of this application for description, these descriptions should not be limited to these terms. These terms are only used to distinguish the descriptions. For example, without departing from the scope of the embodiments of this application, first can also be referred to as second, and similarly, second can also be referred to as first.

[0039] Please see Figure 1 As shown, this is the question-answering learning method based on artificial intelligence in this embodiment, including:

[0040] Step S1: Collect question-and-answer data, user data, and question tracking data to construct a question-and-answer database. The question-and-answer data includes question ID, standard answer, related keywords, preset timeliness coefficient, and question type. The standard answer is a preset online search answer related to the key content of the question. In this embodiment, the setting of the standard answer is not specifically limited; it can be preset content or open-source question-and-answer datasets, etc. The related keywords are keywords in the question content and answer content. For example, if the question is "What are the core technologies in the field of artificial intelligence?" and the standard answer is "The core technologies in the field of artificial intelligence include machine learning, deep learning, natural language processing, and computer vision," then the related keywords for the question and answer content are "artificial intelligence field, core technologies, machine learning." The "learning, deep learning, natural language processing, computer vision" categories are mentioned. The preset timeliness coefficient is a preset timeliness coefficient for each question type. In this embodiment, the setting of the preset timeliness coefficient is not specifically limited. The preset timeliness coefficient should be set to be within (0,1]. For example, when the question type is a technical question, the preset timeliness coefficient can be set to 0.8, and when the question type is a theoretical question, the preset timeliness coefficient can be set to 0.3, etc. The user data includes user ID, answer content, historical answer records, adoption rate, and active time. The question tracking data includes question ID, question content, historical version, modification time, and modification number. The question and answer data, user data, and question tracking data are collected by importing data from the backend of the question and answer learning management system.

[0041] Specifically, in step S1 of this embodiment, the time interval between the current time and the last modification time corresponding to the question ID is used as the decay duration, and the question-and-answer decay factor is analyzed based on the decay duration. The expression for the question-and-answer decay factor is: λ(i)=1 / (D1(i)+1), where λ(i) represents the question-and-answer decay factor, i represents the question ID, and D1(i) represents the decay duration. The unit of the decay duration is days.

[0042] Specifically, in step S1 of this embodiment, question-and-answer data, user data, question tracking data, and question-and-answer decay factors are stored as data content in the question-and-answer database.

[0043] Specifically, in step S1 of this embodiment, by integrating issue tracking data, user behavior data, and question-and-answer data, a complete data foundation is provided for multi-dimensional analysis, avoiding the problem of information silos.

[0044] Please continue reading. Figure 1 As shown, the AI-based question-answering learning method further includes:

[0045] Step S2 involves extracting features from the problem tracking data and user data to obtain a set of problem keywords and timeliness feature values.

[0046] Please see Figure 2 As shown, this is a method for extracting data features, including:

[0047] Step S21: Extract keywords from the question content and the answer content to obtain a set of question keywords.

[0048] Specifically, in step S21 of this embodiment, the question content and the answer content are segmented into words to obtain multiple sets of question and answer words, and the question and answer words are matched with preset secondary keywords. Words that are the same as preset secondary keywords in the question and answer words are removed, and the removed question and answer words are used as a set of question keywords.

[0049] It is understood that this embodiment does not specifically limit the word segmentation method for the question and answer content. Those skilled in the art can freely set it, such as using Chinese word segmentation tools like jieba and SnowNLP to segment the question and answer content.

[0050] It is understood that this embodiment does not specifically limit the setting of preset secondary keywords. The preset secondary keywords are keywords in the question content and answer content that do not reflect the theme meaning, such as "beautiful", "is it", "what", etc., which are keywords with low relevance to the question content and answer content. Those skilled in the art can set them freely according to the type of question.

[0051] Please continue reading. Figure 2 As shown, the data feature extraction method further includes:

[0052] Step S22: Analyze the timeliness characteristic value based on the modification time and the number of modifications.

[0053] Specifically, in step S21 of this embodiment, the number of times the modification time corresponding to the problem ID is within 30 days before the current time is used as the timeliness count, and the timeliness characteristic value is analyzed based on the timeliness count and the number of modifications. The expression of the timeliness characteristic value is: T(i)=α(i)×N1(i) / N2(i), where T(i) represents the timeliness characteristic value, N1(i) represents the timeliness count, and N2(i) represents the number of modifications.

[0054] Specifically, in step S2 of this embodiment, secondary words are filtered to improve keyword purity, and the timeliness of questions is quantified by the frequency of modifications in the past 30 days, thereby reducing the participation of outdated answers from the source.

[0055] Please continue reading. Figure 1 As shown, the AI-based question-answering learning method further includes:

[0056] Step S3 involves analyzing semantic similarity and keyword matching based on the question keyword set and question-answer data to extract effective question-answer content.

[0057] Specifically, in step S3 of this embodiment, a statistical method based on the vector space model is used to analyze the semantic similarity between the answer content and the standard answer, and the question keywords are matched with related keywords. The number of keywords that are the same in the question keywords and related keywords is counted as the matching number. Based on the matching number, the question keywords and related keywords are analyzed to determine the keyword matching degree. The expression for the keyword matching degree is: G(i,j)=NG(i,j)×2 / [NG1(i)+NG2(j|i)], where G(i,j) represents the keyword matching degree, NG(i,j) represents the matching number, NG1(i) represents the number of keywords in the related keywords, NG2(j|i) represents the number of keywords in the question keywords, and j represents the user ID.

[0058] It is understood that this embodiment does not specifically limit the semantic similarity analysis process, and those skilled in the art can freely set it, such as setting it to use methods such as edit distance and Hamming distance to analyze similarity.

[0059] Specifically, in step S3 of this embodiment, the answer content is extracted based on semantic similarity and keyword matching degree. The answer content that satisfies the semantic similarity greater than or equal to the similarity threshold and the keyword matching degree greater than or equal to the matching degree threshold is extracted as the valid answer content.

[0060] Specifically, in this embodiment, the similarity threshold is set to 0.7 and the matching threshold is set to 0.6. This embodiment does not impose specific limitations on the setting of the similarity threshold and the matching threshold. Those skilled in the art can set them freely. The setting of the similarity threshold and the matching threshold should satisfy the condition of being within [0.6, 0.8].

[0061] Specifically, in step S3 of this embodiment, irrelevant or low-relevance answers are accurately eliminated through dual verification of semantic similarity and keyword matching.

[0062] Please continue reading. Figure 1 As shown, the AI-based question-answering learning method further includes:

[0063] Step S4: Analyze user discrimination factors based on effective question and answer content, and construct a quality assessment model based on user discrimination factors and timeliness feature values ​​to obtain answer scores.

[0064] Please see Figure 3 As shown, this is a method for constructing a quality assessment model, including:

[0065] Step S41: Analyze user discrimination factors based on valid question and answer content. The user discrimination factors include adoption rate factor and credibility factor.

[0066] Specifically, in step S41 of this embodiment, the adoption rate factor is analyzed based on the adoption rate and the question-answer decay factor. The expression for the adoption rate factor is:

[0067] S(j)=C(j)×100×e -λ(i)×30 In the formula, S(j) represents the adoption rate factor, and C(j) represents the adoption rate.

[0068] Specifically, in step S41 of this embodiment, the number of times a user logs in within a month is counted as the recent login count; the time interval between the current time and the user's last active time is counted as the user's validity period; the number of times the user answers questions of different question types is counted as the user's answer type count; the number of answers in the user's historical answer records is counted as the user's total answer count; the proportion of the user's answer type count for the question type corresponding to the current analysis question ID in the user's total answer count is counted as the type consistency parameter; and a credibility factor is analyzed based on the recent login count, user validity period, and type consistency parameter. The expression for the credibility factor is:

[0069] U(i,j)=u1×min(1,N3(j) / 20)+u2 / (1+D2(j))+u3×P(i,j), where U(i,j) represents the credibility factor, u1 represents the activity weight, u2 represents the timeliness weight, u3 represents the consistency weight, u1+u2+u3=1, N3(j) represents the recent login count, D2(j) represents the user's validity period, and P(i,j) represents the type consistency parameter. The unit of the user's validity period is days.

[0070] Specifically, in this embodiment, the activity weight is set to 0.5, the timeliness weight is set to 0.3, and the consistency weight is set to 0.2. This embodiment does not impose specific limitations on the setting of the activity weight, timeliness weight, and consistency weight, and those skilled in the art can set them freely.

[0071] Please continue reading Figure 3 As shown, the method for constructing the quality assessment model further includes:

[0072] Step S42: Assign weights to the discriminant factors and timeliness feature values ​​to obtain the answer score.

[0073] Specifically, in step S42 of this embodiment, the adoption rate factor, credibility factor, and timeliness feature value are weighted to obtain the answer score. The expression for the answer score is: W = w1 × S(j) + w2 × U(i,j) + w3 × T(i), where W represents the answer score, w1 represents the adoption rate weight, w2 represents the credibility weight, w3 represents the timeliness weight, and w1 + w2 + w3 = 1.

[0074] Specifically, in this embodiment, the adoption rate weight is set to 0.4, the credibility weight is set to 0.3, and the timeliness weight is set to 0.3. This embodiment does not impose specific limitations on the settings of the adoption rate weight, credibility weight, and timeliness weight, and those skilled in the art can set them freely.

[0075] Specifically, in step S4 of this embodiment, weights are dynamically allocated by integrating user credibility, answer acceptance rate, and question timeliness in order to achieve a quantitative score for answer quality.

[0076] Please continue reading Figure 1 As shown, the AI-based question-answering learning method further includes:

[0077] Step S5: Extract candidate answers based on answer scores and push them to the user.

[0078] Specifically, in step S5 of this embodiment, the answer scores are sorted in descending order, and the answer content corresponding to the top X answer scores is taken as candidate answers, where X represents the extraction parameter, and 3≤X≤5.

[0079] Please continue reading Figure 1 As shown, the AI-based question-answering learning method further includes:

[0080] Step S6: Collect user feedback data on candidate answers and update the quality assessment model based on the feedback data, including accepted and unaccepted answers.

[0081] Specifically, in step S6 of this embodiment, when the feedback data is "accepted," the analysis process of the acceptance rate factor for the next analysis of the current problem is updated, and the expression of the acceptance rate factor is updated to: S(j)=C(j)×100×e -λ(i)×30 +10×(1-e -λ(i)×30 ).

[0082] Specifically, in step S6 of this embodiment, model optimization rules are triggered based on the adoption rate and type consistency parameters. The model optimization rules include: when the user's adoption rate decreases by more than Q1 within a week, the statistical time range of the timeliness count is increased to 60 days, and the timeliness feature value is recalculated; when the user's type consistency parameter fluctuates by more than Q2 within a week, the consistency weight is reduced and the activity weight and timeliness weight are increased, so that the consistency weight decreases by Q2, and the increase in the activity weight and timeliness weight is: u3×Q2 / 2; where Q1 represents the first amplitude threshold, 10%≤Q1≤20%, and Q2 represents the second amplitude threshold, 15≤Q2≤30%.

[0083] It is understood that this embodiment does not impose specific limitations on the value of the amplitude threshold. Those skilled in the art can set it freely, as long as it meets the requirements for formulating the model optimization rules. The optimal values ​​for the amplitude threshold are: Q1 = 20%, Q2 = 25%.

[0084] Specifically, in step S6 of this embodiment, by collecting user adoption and non-adoption behaviors, the adoption rate calculation strategy and weight allocation rules are adjusted in real time to form a closed-loop optimization mechanism.

[0085] Please see Figure 4 As shown, this is an artificial intelligence-based question-answering learning device in this embodiment, including:

[0086] The data collection module is used to collect question-and-answer data, user data, and question tracking data, and to build a question-and-answer database.

[0087] The feature analysis module is used to extract features from issue tracking data and user data to obtain a set of issue keywords and timeliness feature values;

[0088] The similarity matching module is used to analyze semantic similarity and keyword matching based on the question keyword set and question-answer data in order to extract effective question-answer content;

[0089] The quality assessment module is used to analyze user discrimination factors based on effective question and answer content, and to build a quality assessment model based on user discrimination factors and timeliness feature values ​​to obtain answer scores;

[0090] The candidate push module is used to extract candidate answers based on answer scores and push them to users.

[0091] The feedback analysis module is used to collect user feedback data on candidate answers and update the quality assessment model based on the feedback data.

[0092] This application also provides a computer-readable storage medium storing instructions that, when run on a computer, cause the computer to execute the artificial intelligence-based question-answering learning method as described in the above method embodiments.

[0093] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable programs, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable programs, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0094] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is impossible to exhaustively list all the implementation methods here. All obvious variations or modifications derived from the technical solutions of the present invention are still within the protection scope of the present invention.

Claims

1. A question-answering learning method based on artificial intelligence, characterized in that, include: Collect question-and-answer data, user data, and question tracking data to build a question-and-answer database; Feature extraction is performed on issue tracking data and user data to obtain a set of issue keywords and timeliness feature values; Based on the set of question keywords and question-answer data, semantic similarity and keyword matching are analyzed to extract effective question-answer content; Based on the analysis of effective question and answer content, user discrimination factors are analyzed, and a quality assessment model is constructed based on user discrimination factors and timeliness feature values ​​to obtain answer scores; Based on the answer rating, candidate answers are extracted and pushed to users; Collect user feedback data on candidate answers and update the quality assessment model based on the feedback data.

2. The question-answering learning method based on artificial intelligence according to claim 1, characterized in that, The time interval between the current time and the last modification time corresponding to the question ID is used as the decay duration, and the question-answer decay factor is analyzed based on the decay duration.

3. The question-answering learning method based on artificial intelligence according to claim 2, characterized in that, The question and answer content are segmented into words to obtain multiple sets of question and answer terms. The question and answer terms are then matched with preset secondary keywords. Words that are the same as the preset secondary keywords in the question and answer terms are removed. The removed question and answer terms are used as the question keyword set. The number of times the modification time corresponding to the statistical problem ID is within 30 days before the current time is used as the timeliness count. The timeliness characteristic value is analyzed based on the timeliness count and the number of modifications. The expression of the timeliness characteristic value is: T(i)=α(i)×N1(i) / N2(i), where T(i) represents the timeliness characteristic value, N1(i) represents the timeliness count, and N2(i) represents the number of modifications.

4. The question-answering learning method based on artificial intelligence according to claim 3, characterized in that, Analyze the semantic similarity between the answer content and the standard answer, match the question keywords with related keywords, count the number of keywords that are the same in the question keywords and related keywords as the matching number, and analyze the keyword matching degree based on the matching number and the question keywords and related keywords; The answer content is extracted based on semantic similarity and keyword matching degree. Answer content that meets the requirements of semantic similarity greater than or equal to the similarity threshold and keyword matching degree greater than or equal to the matching degree threshold is selected as valid answer content.

5. The question-answering learning method based on artificial intelligence according to claim 4, characterized in that, The adoption rate factor is analyzed based on the adoption rate and question-answer decay factor. The number of times a user logs in within a month is counted as the recent login count. The time interval between the current time and the user's last active time is counted as the user's validity period. The number of times a user answers different question types is counted as the user's answer type count. The number of answers in the user's historical answer records is counted as the user's total answer count. The proportion of the user's answer type count for the question type corresponding to the current analysis question ID in the user's total answer count is counted as the type consistency parameter. The credibility factor is analyzed based on the recent login count, user validity period, and type consistency parameter.

6. The question-answering learning method based on artificial intelligence according to claim 5, characterized in that, The answer score is obtained by weighting the adoption rate factor, credibility factor, and timeliness feature value. The expression of the answer score is: W=w1×S(j)+w2×U(i,j)+w3×T(i), where W represents the answer score, w1 represents the adoption rate weight, w2 represents the credibility weight, w3 represents the timeliness weight, S(j) represents the adoption rate factor, and U(i,j) represents the credibility factor.

7. The question-answering learning method based on artificial intelligence according to claim 6, characterized in that, Sort the answers in descending order of their scores, and select the answers corresponding to the top X scores as candidate answers, where X represents the extracted parameters.

8. The question-answering learning method based on artificial intelligence according to claim 7, characterized in that, When the feedback data indicates acceptance, update the analysis process of the acceptance rate factor for the next analysis of the current issue; The model optimization rules are triggered based on the adoption rate and type consistency parameters. These rules include: when a user's adoption rate decreases by more than Q1 within a week, the statistical time range for the number of timeliness events is increased to 60 days, and the timeliness feature value is recalculated; when a user's type consistency parameter fluctuates by more than Q2 within a week, the consistency weight is reduced, and the activity weight and timeliness weight are increased; where Q1 represents the first amplitude threshold and Q2 represents the second amplitude threshold.

9. An artificial intelligence-based question-answering learning device, applied to the artificial intelligence-based question-answering learning method as described in any one of claims 1-8, characterized in that, include: The data collection module is used to collect question-and-answer data, user data, and question tracking data, and to build a question-and-answer database. The feature analysis module is used to extract features from issue tracking data and user data to obtain a set of issue keywords and timeliness feature values; The similarity matching module is used to analyze semantic similarity and keyword matching based on the question keyword set and question-answer data in order to extract effective question-answer content; The quality assessment module is used to analyze user discrimination factors based on effective question and answer content, and to build a quality assessment model based on user discrimination factors and timeliness feature values ​​to obtain answer scores; The candidate push module is used to extract candidate answers based on answer scores and push them to users. The feedback analysis module is used to collect user feedback data on candidate answers and update the quality assessment model based on the feedback data.

10. A storage medium, characterized in that, It stores instructions that, when run on a computer, cause the computer to perform the artificial intelligence-based question-answering learning method as described in any one of claims 1-8.

Images