Intelligent question and answer method and device, computing equipment, program product and storage medium
Through the importance evaluation and semantic clustering of the original question, combined with multiple evaluation factors and network search, the question of meaningless or repeated questioning in multi-person online intelligent question-and-answer scenarios is solved, and the answer efficiency and process stability are improved.
Patent Information
- Application Number
- CN202310833369.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-05-05
- Filing Date
- 2023-07-07
- Publication Date
- 2024-11-05
AI Technical Summary
In the multi-person online smart question and answer scenario, there are a large number of meaningless questions or repeated questions, which makes computing devices unable to handle higher priority issues, and may even lead to paralysis of the smart question and answer process.
By evaluating the importance of the original question, determining the importance of its answers, and determining the degree of detail and priorities based on the evaluation results, we can flexibly deal with the question and avoiding equal answers. This method combines multiple influencing factors such as questioner level, question category and frequency, uses semantic clustering to identify the same problem of different expressions, and generates answers dynamically through web searches.
It improves the efficiency of answering questions, avoids the processing of meaningless or redundant questions, ensures that computing devices can effectively handle higher priority questions, and enhances the robustness and personalized answers of the intelligent question-and-answer process.
Smart Images

Figure CN118897877A_ABST
Abstract
Description
[0001] This application claims priority to Chinese patent application No. 202310497803.1, filed on May 5, 2023, and entitled “A task scheduling method, device and system”, the entire contents of which are incorporated by reference into this application. Technical Field
[0002] The present application relates to the field of artificial intelligence, and in particular to an intelligent question-answering method, apparatus, computing device, program product, and storage medium. Background Art
[0003] Intelligent question-answering is used in the field of natural language processing to answer questions asked by users in natural language. Intelligent question-answering is usually divided into generation type and retrieval type. In the retrieval type intelligent question-answering scenario, the computing device receives the user's question and retrieves the question in the corpus. The corpus includes multiple question-answer corpus pairs, each of which includes a question and a corresponding answer. If the question is retrieved in the corpus, the corresponding answer is fed back to the user; if the question is not retrieved in the corpus, the question is directly ignored or a fixed answer is fed back to the user. However, in intelligent question-answering scenarios with multiple people online, such as live barrage, conference chat, webinars, online classes, etc., when there are a large number of meaningless questions or a large number of screen swipes for the same question, if these questions are answered continuously according to the above process, not only will the computing device be unable to process important questions, but the entire intelligent question-answering process may also be paralyzed due to insufficient processing power of the computing device. Summary of the invention
[0004] The present application provides an intelligent question-answering method, apparatus, computing device, program product, and storage medium, which can flexibly answer a large number of diverse questions and output answers of different levels of detail for questions of different importance. The technical solution is as follows:
[0005] In a first aspect, an intelligent question-answering method is provided, the method comprising: receiving an original question; determining an importance assessment result of the original question, the importance assessment result indicating the importance of the answer to the original question, the importance assessment result being determined by at least one assessment influencing factor; based on the importance assessment result, determining an answer corresponding to the original question from at least one candidate answer, at least one candidate answer having a different degree of detail; and outputting the answer corresponding to the original question.
[0006] It can be seen that this application evaluates the importance of the original questions and determines the level of detail in answering the original questions based on the importance evaluation results, thereby flexibly handling the original questions and improving the efficiency of answering questions. In the case of a large number of original questions, it is necessary to avoid answering all questions equally, which will cause the computing device to be unable to process questions with higher priority, and may even paralyze the entire intelligent question-answering process.
[0007] Optionally, the method further includes: receiving an evaluation rule configuration request, the evaluation rule configuration request being used to request configuration of one or more of a type, a priority and a weight of at least one evaluation influencing factor.
[0008] By configuring at least one evaluation influencing factor, the computing device can combine at least one evaluation influencing factor to perform importance evaluation on the original question, thereby improving the accuracy of the importance evaluation result of the original question.
[0009] Optionally, the type of at least one evaluation influencing factor includes one or more of the questioner level, question category, question frequency and answer indication information of the original question, the questioner level indicates the importance of the questioner, the question category indicates whether the original question is a meaningless question, the question frequency indicates the historical question frequency of the original question, and the answer indication information indicates whether the original question has been answered within a specified time period before the current time.
[0010] Optionally, determining the importance evaluation result of the original question includes: determining an evaluation result of the original question under at least one evaluation influencing factor; and determining the importance evaluation result of the original question based on the evaluation result of the original question under at least one evaluation influencing factor.
[0011] That is, when evaluating the importance of the original question, the original question can be evaluated based on multiple evaluation influencing factors, so as to determine the importance evaluation result of the original question based on the evaluation results of the original question under multiple evaluation influencing factors. Since the importance evaluation result takes into account multiple influencing factors, the evaluation result is more reliable and more conducive to accurately determining the answer to the original question.
[0012] Optionally, based on the importance assessment result, determining the answer corresponding to the original question from at least one candidate answer, including: based on the importance assessment result, determining the answer strategy for the original question from the correspondence between the importance assessment result and the answer strategy, the answer strategy indicating a fixed term answer, a simple answer or a comprehensive answer, the level of detail of the answer corresponding to the simple answer being lower than the level of detail of the answer corresponding to the comprehensive answer; based on the answer strategy, determining the answer corresponding to the original question from at least one candidate answer.
[0013] Based on the importance evaluation results of the original question, it is determined whether the original question needs to be answered, the priority of the answer, and the level of detail of the answer. The original question can be handled flexibly, which improves the efficiency of answering questions. In the case of a large number of original questions, it is avoided to answer all questions equally, and to avoid redundant answers to a large number of meaningless questions or to the same question, which will cause the computing device to be unable to process questions with higher priority, and may even cause the entire question-answering process to be paralyzed.
[0014] Optionally, the method also includes: based on the original question, determining a target fundamental question corresponding to the original question from a corpus by semantic clustering, the corpus including multiple fundamental questions and basic answers corresponding to each fundamental question; based on the basic answer corresponding to the target fundamental question, determining at least one candidate answer.
[0015] By using semantic clustering, the corresponding target fundamental question is determined from the corpus, and then the answer to the original question is determined based on the basic answer to the target fundamental question. In this way, semantic clustering can identify questions with different questions but the same semantics, and cluster different expressions of the same question into one fundamental question. Therefore, only some basic answers to the fundamental questions need to be configured in the corpus, and the corpus can be used to respond to different forms of questions from the questioner, which greatly saves the time of configuring the question-answer corpus in the corpus; at the same time, it makes the intelligent question-answering process more robust and able to respond to original questions with different expressions in various scenarios.
[0016] Optionally, based on the basic answer corresponding to the target root question, at least one candidate answer is determined, including: based on the original question and / or the target root question, obtaining at least one network answer through network search; based on the basic answer corresponding to the target root question and at least one network answer, generating at least one candidate answer.
[0017] Since the corpus is pre-constructed, the basic answers corresponding to the fundamental questions are all preset, but with the enrichment and diversification of knowledge, the answers corresponding to each fundamental question may have more accurate and richer expressions. Therefore, it is possible to use the information search advantages of the Internet to search for network answers to the original question and / or the target fundamental question, so as to dynamically generate candidate answers with different levels of detail corresponding to the original question. Secondly, when there are multiple candidate answers corresponding to the original question, it is convenient for the computing device to flexibly determine the answer to the original question from the candidate answers, that is, for the same original question, different candidate answers can also be selected to answer, making the answer more personalized.
[0018] Optionally, the method further includes: receiving a corpus configuration request, the corpus configuration request carrying multiple fundamental question-answer corpus pairs, each fundamental question-answer corpus pair including a fundamental question and a basic answer corresponding to the fundamental question; and storing the multiple fundamental question-answer corpus pairs in a corpus.
[0019] Optionally, the multiple fundamental question-answer corpus pairs are fundamental question-answer corpus pairs corresponding to multiple levels, each level corresponds to at least one fundamental question-answer corpus pair, the fundamental question-answer corpus pairs corresponding to different levels have different granularities, and the corpus configuration request also carries identifiers of multiple levels; storing the multiple fundamental question-answer corpus pairs in the corpus includes: based on the identifiers of the multiple levels, storing the fundamental question-answer corpus pairs corresponding to the multiple levels in the corpus.
[0020] That is, when the administrator needs to build a corpus, add a new level to the original corpus, or add a fundamental question-answer corpus pair to the original level of the corpus, a corpus configuration request can be sent to the computing device through the administrator terminal (or triggered in the computing device), and the corpus configuration request carries the identifiers of multiple levels to be configured and the fundamental question-answer corpus pairs of each level. The computing device responds to the corpus configuration request and stores the fundamental question-answer corpus pairs corresponding to the level in the corresponding level of the corpus according to the level identifier; or, the computing device responds to the corpus configuration request and adds at least one level to the corpus according to the level identifier, and stores the fundamental question-answer corpus pairs corresponding to the level in the newly added level.
[0021] Optionally, the administrator may also set a target level corresponding to the fundamental question-answer corpus pair for semantic clustering with the original question. In this way, the computing device may obtain multiple fundamental questions from the fundamental question-answer corpus pairs corresponding to the target level based on the target level preset by the administrator and perform semantic clustering with the original question. This makes the multiple fundamental questions selected for the original question more accurate. At the same time, it reduces the computational complexity of semantic clustering and greatly improves the efficiency of semantic clustering.
[0022] It should be noted that the granularity indicates the level of detail of the description of the fundamental question-answer corpus at that level. The higher the granularity, the more macroscopic the description of the fundamental question; the lower the granularity, the more detailed the description of the fundamental question.
[0023] Optionally, the original question includes a first question and a second question; outputting the answer corresponding to the original question includes: if, based on the importance assessment result of the first question and the importance assessment result of the second question, it is determined that the priority of the first question is higher than the priority of the second question, then outputting the answer corresponding to the first question first.
[0024] In this way, when the original question includes multiple questions, after determining the answer corresponding to each question, the order of answering the questions can be flexibly adjusted based on the importance evaluation result of each question.
[0025] In a second aspect, an intelligent question-answering device is provided, which has the function of implementing the intelligent question-answering method in the first aspect. The intelligent question-answering device includes at least one module, which is used to implement the steps of the intelligent question-answering method provided in the first aspect.
[0026] In a third aspect, a computing device is provided, which includes a processor and a memory, the memory being used to store a computer program for executing the intelligent question and answer method provided in the first aspect above; the processor being configured to execute the computer program stored in the memory to implement the steps of the intelligent question and answer method provided in the first aspect above.
[0027] Optionally, the computing device may further include a communication bus, which is used to establish a connection between the processor and the memory.
[0028] In a fourth aspect, a computer-readable storage medium is provided, which stores instructions. When the instructions are executed on a computing device, the computing device executes the steps of the intelligent question-answering method provided in the first aspect.
[0029] In a fifth aspect, a computer program product including instructions is provided, which, when executed on a computing device, enables the computing device to execute the steps of the intelligent question-answering method provided in the first aspect. In other words, a computer program is provided, which, when executed on a computing device, enables the computing device to execute the steps of the intelligent question-answering method provided in the first aspect.
[0030] The technical effects obtained by the above-mentioned second, third, fourth and fifth aspects are similar to the technical effects obtained by the corresponding technical means in the first aspect, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 It is a flowchart of an intelligent question-answering method provided in an embodiment of the present application;
[0032] Figure 2 This is a question-and-answer process diagram provided in an embodiment of the present application;
[0033] Figure 3 It is a flowchart of another intelligent question-answering method provided in an embodiment of the present application;
[0034] Figure 4 is a structural diagram of an intelligent question-answering device provided in an embodiment of the present application;
[0035] Figure 5 is a schematic diagram of the structure of a computing device provided in an embodiment of the present application;
[0036] Figure 6 is a schematic diagram of the structure of a computing device cluster provided in an embodiment of the present application;
[0037] Figure 7 It is a schematic diagram of the network connection relationship of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0038] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.
[0039] In some scenarios of intelligent question-and-answer (Q&A) with multiple people online, such as live broadcasts, conference chats, webinars, and online classes, users often ask a large number of questions, and for each question, the computing device uses a pre-built corpus containing multiple question-and-answer pairs to implement intelligent Q&A. When there are a large number of meaningless questions or a large number of repeated questions asking the same question, the computing device will still answer these redundant questions in the same way, causing the computing device to be unable to handle questions with higher priority, thus paralyzing the entire intelligent Q&A process.
[0040] Based on this, the present application provides an intelligent question-answering method that can evaluate the importance of the original question to determine the importance of the answer to the original question, so as to flexibly answer the original question based on the importance evaluation result of the original question, and for important original questions, the output answer is more detailed.
[0041] Please refer to Figure 1 , Figure 1 This is a flowchart of an intelligent question-answering method provided in an embodiment of the present application. The method can be applied to any computing device or a computing device cluster including multiple computing devices. The method includes the following steps.
[0042] Step 101: Receive an original question.
[0043] The original question is the original question sent by the questioner through the questioner terminal, or the original question directly input by the questioner in the computing device. The original question can be presented in any form such as text, picture, voice, etc. The embodiment of the present application takes the original question in text form as an example to explain the intelligent question-answering process, but does not constitute a limitation on the question form of the original question.
[0044] In some embodiments, if the original question is a question in voice form, the computing device may perform a voice-to-text operation on the original question after receiving the original question, and then perform the following steps 102 and 103 based on the question text of the original question. That is, according to the original question in text form, the importance of the original question is evaluated, and based on the importance evaluation result of the original question, the answer corresponding to the original question is determined.
[0045] Step 102: Determine an importance evaluation result of the original question, where the importance evaluation result indicates the importance of the answer to the original question, and the importance evaluation result is determined by at least one evaluation influencing factor.
[0046] In some embodiments, the computing device may perform an importance assessment on the original question in combination with at least one assessment influencing factor to obtain an importance assessment result.
[0047] It should be noted that the at least one evaluation factor for evaluating the importance of the original question may be an evaluation factor that the administrator initially configures for the computing device; it may also be an evaluation factor that is adjusted based on actual business needs; it may also be an evaluation factor that the administrator newly configures in the computing device before receiving the original question. The embodiment of the present application does not limit the number and content of the configured evaluation factors.
[0048] It should be noted that the "administrator" here can be a designated user who manages the intelligent question and answer process, such as an anchor who uses the intelligent question and answer process to answer questions asked in live broadcast barrages (in this case, the computing device is a anchor-side device, such as a terminal or server), a teaching assistant who uses the intelligent question and answer process to answer questions asked by multiple students (in this case, the computing device is a teaching assistant-side device, such as a terminal or server), a robot that uses the intelligent question and answer process to interact intelligently with tourists, and can also be an administrator of other computing devices that need to interact with multiple users for intelligent question and answer. The embodiments of the present application do not impose any restrictions on this.
[0049] In some embodiments, the configuration process of the evaluation rule is as follows: the computing device receives an evaluation rule configuration request, the evaluation rule configuration request is used to request configuration of one or more of the type, priority, and weight of at least one evaluation influencing factor. Based on this, the computing device stores at least one evaluation influencing factor, and the priority and weight of each evaluation influencing factor according to the information carried in the evaluation rule configuration request.
[0050] Optionally, when there are multiple configured evaluation influencing factors, the computing device may also, in response to the administrator's configuration, select a preset number of evaluation influencing factors with higher priorities to perform importance evaluation on the original question. The embodiment of the present application does not limit the number of evaluation influencing factors used when performing importance evaluation on the original question.
[0051] Taking the questions in the live broadcast barrage as an example, the importance of different questions is different. For meaningless barrages, you can selectively block them during the Q&A process, that is, do not answer them. The processing priority of questions asked by users with different labels also needs to be different. For example, for the "number one" user who can bring a lot of economic income, it is necessary to give priority to the user's questions, while for users with low interaction frequency and simply watching the live broadcast, their questions can be processed with a lower priority. For the same question that is swiped on the screen in large quantities, it can be processed with a low priority, and it can be processed selectively when processing to avoid repeatedly replying to redundant questions. In addition, for the same question that has been answered in a short period of time, it can also be processed with a low priority to avoid the computing device always perceiving the same question and missing the information of other questions.
[0052] Based on this, the type of at least one evaluation influencing factor used when evaluating the importance of the original question may include: one or more of the questioner level, question category, question frequency, and answer indication information of the original question. The questioner level indicates the importance of the questioner, the question category indicates whether the original question is a meaningless question, the question frequency indicates the historical question frequency of the original question, and the answer indication information indicates whether the original question has been answered within a specified time period before the current time.
[0053] The specified duration may be any pre-set value, such as 1 minute, 5 minutes, 10 minutes, etc.
[0054] For the above live broadcast scenario, more evaluation factors can be added according to actual business needs, such as giving priority to replying to female users, giving priority to replying to new followers, etc.
[0055] In some embodiments, the implementation process of the above step 102 may be: determining an evaluation result of the original question under at least one evaluation influencing factor; and determining an importance evaluation result of the original question based on the evaluation result of the original question under at least one evaluation influencing factor.
[0056] The importance evaluation result may be the importance level of the original question, or the importance score of the original question, or may be in other forms. The embodiment of the present application does not limit the form of expression of the importance evaluation result. Original questions with high importance levels or importance scores are more important, and answers should be generated and answered first; original questions with low importance levels or importance scores are less important, and can be answered later, or simply answered or not answered.
[0057] Taking the use of multiple evaluation influencing factors to conduct importance evaluation on the original question as an example, if the importance evaluation result is an importance level, the importance evaluation process of the original question can be: determine the evaluation level of the original question under each evaluation influencing factor respectively, obtain multiple evaluation levels, and then sum up the multiple evaluation levels to determine the importance level of the original question.
[0058] Similarly, if the importance assessment result is an importance score, the importance assessment process of the original question can be: determine the evaluation score of the original question under each evaluation influencing factor respectively, obtain multiple evaluation scores, and then sum up the multiple evaluation scores to determine the importance score of the original question.
[0059] It should be understood that if the evaluation criteria for importance assessment are different from the evaluation criteria for each evaluation influencing factor, after summing up the evaluation results of the original question under multiple evaluation influencing factors, the summation result needs to be converted according to the evaluation criteria for importance assessment to obtain the importance assessment result of the original question.
[0060] As an example, the evaluation standard for evaluation influencing factors is a ten-level system, and the evaluation level of the original question under each evaluation influencing factor can be 1-10, while the evaluation standard for importance evaluation is a five-level system. If the sum of the evaluation levels of the original question under at least one evaluation influencing factor is 8, then the importance evaluation result of the original question is determined to be 4.
[0061] As another example, the evaluation standard for evaluation influencing factors is based on a percentage system, and the evaluation score of the original question under each evaluation influencing factor can be 0-100 points, while the evaluation standard for importance evaluation is based on a ten-point system. If the sum of the evaluation scores of the original question under at least one evaluation influencing factor is 80 points, then the importance evaluation result of the original question is determined to be 8 points.
[0062] In some embodiments, since different types of evaluation influencing factors have different degrees of influence on the importance evaluation results of the original question, a priority or weight can be set for each evaluation influencing factor so that when evaluating the importance of the original question, the important influencing factors that affect the importance evaluation results can be focused on.
[0063] In one possible implementation, taking the configuration of multiple evaluation influencing factors and the priority of each evaluation influencing factor as an example, the implementation process of determining the importance evaluation result of the original question can be: determining the evaluation level of the original question under each evaluation influencing factor; based on the evaluation level of each evaluation influencing factor and the priority of each evaluation influencing factor, determining the importance evaluation result of the original question.
[0064] That is, after determining the evaluation level of the original question under each evaluation influencing factor, weighted sum of each evaluation level and the priority of each evaluation influencing factor is performed. Then, based on the evaluation criteria of the importance evaluation result, the weighted sum result is transformed to obtain the importance evaluation result of the original question.
[0065] As an example, if there are three evaluation factors with a ten-level evaluation standard: evaluation factor a, evaluation factor b, and evaluation factor c, and the priority of evaluation factor a is 1, the priority of evaluation factor b is 2, and the priority of evaluation factor c is 5. Assuming that the evaluation level of the original question under evaluation factor a is 8, the evaluation level under evaluation factor b is 2, and the evaluation level under evaluation factor c is 5, then the evaluation level of the original question under evaluation factor a and the priority of evaluation factor a, the evaluation level of the original question under evaluation factor b and the priority of evaluation factor b, and the evaluation level of the original question under evaluation factor c and the priority of evaluation factor c are weighted and summed (8*1+2*2+5*5=37). Assuming that the evaluation standard of the importance evaluation result is a five-level system, after the weighted sum result is converted, it is determined that the importance evaluation result of the original question is 1.85.
[0066] In one possible implementation, taking the configuration of multiple evaluation influencing factors and the weight of each evaluation influencing factor as an example, the implementation process of determining the importance evaluation result of the original question can be: determining the evaluation score of the original question under each evaluation influencing factor; based on the evaluation score of each evaluation influencing factor and the weight of each evaluation influencing factor, determining the importance evaluation result of the original question.
[0067] That is, after determining the evaluation score of the original question under each evaluation factor, the weighted sum of each evaluation score and the weight of each evaluation factor is performed. Then, based on the evaluation criteria of the importance evaluation result, the weighted sum result is transformed to obtain the importance evaluation result of the original question.
[0068] As an example, if there are three evaluation factors with a percentage evaluation standard: evaluation factor d, evaluation factor e, and evaluation factor f, and the weight of evaluation factor d is 0.8, the weight of evaluation factor e is 0.2, and the weight of evaluation factor f is 0.6. Assuming that the evaluation score of the original question under evaluation factor d is 70 points, the evaluation score under evaluation factor e is 80 points, and the evaluation score under evaluation factor f is 30 points, then the evaluation score of the original question under evaluation factor d and the weight of evaluation factor d, the evaluation score of the original question under evaluation factor e and the weight of evaluation factor e, and the evaluation score of the original question under evaluation factor f and the weight of evaluation factor f are weighted and summed (70*0.8+80*0.2+30*0.6=90). Assuming that the evaluation standard of the importance evaluation result is a ten-point system, after the weighted sum result is converted, it is determined that the importance evaluation result of the original question is 9 points.
[0069] Optionally, for some meaningless original questions, the importance assessment result of the original question can be directly determined as level 0 or 0 points. Alternatively, when evaluating the importance of such meaningless original questions, the assessment result corresponding to the "question category" in the above-mentioned evaluation influencing factors can be determined as level 0 or 0 points. The embodiments of the present application do not impose any restrictions on this.
[0070] Step 103: Based on the importance evaluation result, determine an answer corresponding to the original question from at least one candidate answer, and at least one candidate answer has a different level of detail.
[0071] Before explaining the above step 103, the implementation process of the computing device determining at least one candidate answer is explained first.
[0072] In intelligent question-and-answer scenarios with multiple people online, such as live broadcast barrage, conference chat, webinars, and online classes, there are often multiple questions. Considering that different people have different expression habits, there may be multiple different ways of asking questions for the same question. If intelligent question-and-answer is achieved by pre-building a corpus containing multiple question-and-answer corpus pairs, the administrator of the computing device needs to configure corresponding answers for different expressions of the same question, which will cause a lot of repetitive work; and it is difficult for the administrator to exhaust all expressions of a question. Therefore, for some questions that the administrator has not configured answers, the computing device may ignore these questions because there is no preset answer, which reduces the user experience. Moreover, when answering questions, if the computing device always uses the question-and-answer corpus pairs in the preset corpus to answer the questions, when the user repeatedly asks a question, the computing device cannot generate answers with more information, and will still repeatedly answer the same answer based on the question-and-answer corpus pairs in the corpus, resulting in a poor user experience.
[0073] Based on this, the embodiment of the present application can determine the target fundamental question corresponding to the original question from the corpus according to the received original question, and then determine at least one candidate answer corresponding to the original question based on the basic answer corresponding to the target fundamental question in the corpus. That is, by identifying questions with different asking methods but the same semantics, multiple questions with the same semantics are clustered into one fundamental question, so that only the basic answers to the fundamental questions need to be configured in the corpus, and the corpus can be used to respond to questions of different forms from the questioner. Moreover, for questions of different forms from the questioner, the embodiment of the present application selects the answer corresponding to the original question from the candidate answers in combination with the importance evaluation results of the above-mentioned original question, so as to flexibly answer questions from different questioners, in different forms, and with different contents.
[0074] In some embodiments, the process of determining at least one candidate answer to the original question may include the following steps:
[0075] (1) Based on the original question, the target fundamental question corresponding to the original question is determined from the corpus through semantic clustering. The corpus includes multiple fundamental questions and the basic answers corresponding to each fundamental question.
[0076] Among them, semantic clustering can cluster multiple original questions with the same semantics but different expressions into the same root question, and each root question corresponds to a semantic feature.
[0077] In other words, semantic clustering can determine the target fundamental questions that reflect the semantic features of the original questions from the multiple fundamental questions in the corpus. When the number of original questions is one, the number of target fundamental questions determined is also one; when the number of original questions is multiple, the target fundamental questions determined may be one or multiple, depending on whether the semantic features of the multiple original questions are the same or similar. When the deviation between the semantic features of the multiple original questions is within the semantic deviation range, the multiple original questions may be different expressions of the same semantics, and correspond to the same target fundamental question.
[0078] In some embodiments, the configuration process of the corpus may be: a computing device receives a corpus configuration request, the corpus configuration request carries multiple fundamental question-answer corpus pairs, each fundamental question-answer corpus pair includes a fundamental question and a basic answer corresponding to the fundamental question; the computing device stores the multiple fundamental question-answer corpus pairs in the corpus.
[0079] In a possible implementation, the multiple fundamental question-answer corpus pairs are fundamental question-answer corpus pairs corresponding to multiple levels, each level corresponds to at least one fundamental question-answer corpus pair, and the fundamental question-answer corpus pairs corresponding to different levels have different granularities, and the above-mentioned corpus configuration request also carries identifiers of multiple levels. Based on this, after receiving the corpus configuration request, the computing device stores the fundamental question-answer corpus pairs corresponding to the multiple levels in the corpus based on the identifiers of the multiple levels.
[0080] That is, when the administrator needs to build a corpus, add a new level to the original corpus, or add a fundamental question-answer corpus pair to the original level of the corpus, a corpus configuration request can be sent to the computing device through the administrator terminal (or triggered in the computing device), and the corpus configuration request carries the identifiers of multiple levels to be configured and the fundamental question-answer corpus pairs of each level. The computing device responds to the corpus configuration request and stores the fundamental question-answer corpus pairs corresponding to the level in the corresponding level of the corpus according to the level identifier; or, the computing device responds to the corpus configuration request and adds at least one level to the corpus according to the level identifier, and stores the fundamental question-answer corpus pairs corresponding to the level in the newly added level.
[0081] In some embodiments, the corpus includes multiple corpus scenarios, such as live broadcast, conference chat, webinar, online class, etc.; each corpus scenario corresponds to a category tree, which includes multiple levels, each level has a different granularity, each level includes at least one category node, and each category node corresponds to at least one fundamental question.
[0082] It should be noted that granularity indicates the level of detail of the description of the fundamental question-answer corpus in this level. The higher the level, the greater the granularity, and the more macroscopic the description of the fundamental questions corresponding to its category nodes; the lower the level, the smaller the granularity, and the more detailed the description of the fundamental questions corresponding to its category nodes.
[0083] Taking the live broadcast scene as an example, its corresponding category tree may include the first level, the second level, the third level and the fourth level, wherein the first level includes category node 1, and the category description information of category node 1 is live broadcast. The second level includes category node 2, category node 3 and category node 4, the category description information of category node 2 is shopping, the category description information of category node 3 is games, and the category description information of category node 4 is entertainment. The third level includes category nodes 5 to category nodes 30, taking category node 2 in the second level as an example, it is subdivided into category nodes 5, category nodes 6, category nodes 7 and category nodes 8 in the third level, the category description information of category node 5 is clothing, the category description information of category node 6 is home appliances, the category description information of category node 7 is food, and the category description information of category node 8 is travel services. The fourth level includes category nodes 31 to 60. Taking category node 5 in the third level as an example, it is subdivided into category nodes 31, 32, 33, 34, 35 and 36 in the fourth level. The category description information of category node 31 is tops, the category description information of category node 32 is pants, the category description information of category node 33 is shoes, the category description information of category node 34 is hats and headwear, the category description information of category node 35 is bags, and the category description information of category node 36 is underwear. By analogy, as the level increases, the category description information set by its category nodes is more detailed, so that more specific and detailed categories can be defined.
[0084] Taking category node 31 as an example, the corresponding fundamental questions may include how much the clothes cost, what the clothes are made of, how to choose the size of the clothes, etc. The embodiment of the present application does not limit the number and content of the fundamental questions that can be set for a category node.
[0085] Optionally, based on the category tree under each corpus scenario in the above corpus, the administrator can also set the target level, that is, the level to which multiple fundamental questions selected from the corpus belong when the target fundamental question corresponding to the original question is determined by semantic clustering. If the administrator does not set the target level, the fundamental questions corresponding to each category node in the lowest level can be selected by default, or the fundamental questions corresponding to the category nodes in all levels can be selected, and this embodiment of the application does not limit this.
[0086] For example, if the original question is a question obtained in a live broadcast scenario, and the target level set by the administrator is the third level, then multiple fundamental questions corresponding to category nodes 5 to category nodes 30 in the third level are selected from the corpus, and then based on the selected multiple fundamental questions, the target fundamental question corresponding to the original question is determined by semantic clustering.
[0087] In this way, based on the collection scenario of the original question and the target level preset by the administrator, multiple fundamental questions can be obtained from the fundamental question and answer corpus pairs corresponding to the target level, making the fundamental questions selected for the original question more accurate. At the same time, the computational complexity of semantic clustering based on the corpus is reduced, greatly improving the efficiency of semantic clustering.
[0088] In some embodiments, based on the above-mentioned corpus, the implementation process of determining the target fundamental question corresponding to the original question by means of semantic clustering can be: determining the semantic feature vector of the original question; based on the semantic feature vector of the original question and the semantic feature vectors of multiple fundamental questions, determining the semantic similarity between the original question and the multiple fundamental questions respectively; based on the semantic similarity, determining the target fundamental question corresponding to the original question from the multiple fundamental questions.
[0089] Optionally, before determining the semantic feature vector of the original question, the original question may be subjected to noise filtering to filter out information that interferes with or is useless for extracting semantic features, such as noise information such as Emoji expressions, symbols, URLs, and conversation references, and then the corresponding semantic feature vector may be obtained based on the filtered original question.
[0090] In a possible implementation, for the original question and multiple fundamental questions, a convolution operation can be performed to determine the semantic feature vector corresponding to each question. The implementation method for determining the semantic feature vector can also refer to other related technologies, which is not limited in the present embodiment.
[0091] After determining the semantic feature vector of the original question and the semantic feature vectors of the multiple root questions, the semantic similarity between each root question and the original question can be determined based on the distance between the semantic feature vector of the original question and the semantic feature vector of each root question. The closer the distance between the semantic feature vectors, the higher the semantic similarity; the farther the distance between the semantic feature vectors, the lower the semantic similarity.
[0092] Among them, the distance between semantic feature vectors can be determined based on Euclidean distance, Manhattan distance, cosine distance, etc., and the embodiments of the present application do not limit this.
[0093] In a possible implementation, based on the semantic similarities between the original question and multiple root questions, and a semantic threshold, a target root question corresponding to the original question can be determined from at least one root question whose semantic similarity with the original question is greater than or equal to the semantic threshold.
[0094] The semantic threshold may be a set semantic feature deviation value, which is used to define whether two semantic feature vectors have semantic relevance. When the semantic similarity of two semantic feature vectors is greater than or equal to the semantic threshold, it is considered that the two semantic feature vectors have semantic relevance and are very likely to be different expressions of the same semantics; and when the semantic similarity of two semantic feature vectors is less than the semantic threshold, it is considered that the two semantic feature vectors may not have semantic relevance and are not different expressions of the same semantics.
[0095] It should be noted that when there is one fundamental question with a semantic similarity greater than or equal to the semantic threshold, the fundamental question is directly determined as the target fundamental question corresponding to the original question. When there are multiple fundamental questions with a semantic similarity greater than or equal to the semantic threshold, a target fundamental question can be randomly determined from the multiple fundamental questions; or the target fundamental question can be determined from the multiple fundamental questions based on a certain selection strategy, which is not limited in the embodiments of the present application.
[0096] In some embodiments, based on the above-mentioned corpus, the implementation process of determining the target fundamental question corresponding to the original question by means of semantic clustering can be: inputting the original question and multiple fundamental questions into the question clustering model, determining the target fundamental question corresponding to the original question among the multiple fundamental questions through the question clustering model, and then outputting the target fundamental question corresponding to the original question.
[0097] The question clustering model is a neural network model pre-trained using a training data set. The training data set includes multiple sets of training data, each set of training data includes multiple questions and a fundamental question, and the fundamental question is a question set manually based on the semantic features of the multiple questions, and the multiple questions have the same semantics as the fundamental question. In other words, the multiple questions in each set of training data semantically describe the fundamental question, but the expression is different.
[0098] In one possible implementation, the training process of the question clustering model is as follows: based on the constructed training data set, multiple questions in each group of training data are input into the initial question clustering model to be trained, and the predicted questions output by the initial question clustering model based on the multiple questions in each group of training data are obtained; according to the predicted questions corresponding to each group of training data and the fundamental questions in each group of training data, the initial question clustering model is iteratively trained, and a trained question clustering model is obtained after the training is completed.
[0099] As an example, the training end conditions include: the number of iterations reaches a preset total number of training times, or the semantic deviation between the predicted problem output by the initial problem clustering model and the corresponding fundamental problem is within a preset semantic deviation range, or the semantic deviation between the predicted problem output by the initial problem clustering model and the fundamental problem no longer changes within a preset number of iterations. The embodiments of the present application do not impose any restrictions on this.
[0100] In some embodiments, based on the above corpus, the implementation process of determining the target fundamental question corresponding to the original question by semantic clustering can be: comparing the characters of the original question with multiple fundamental questions, determining the number of identical characters between each fundamental question and the original question, and then determining the target fundamental question corresponding to the original question from the multiple fundamental questions based on the number of identical characters.
[0101] As an example, a fundamental question with the highest number of identical characters can be determined as the target fundamental question corresponding to the original question. Alternatively, multiple fundamental questions can be sorted from high to low according to the number of identical characters, and then, based on the sorting results of the same number of characters, at least two fundamental questions with the highest number of identical characters can be selected as candidates, and then, based on a certain selection strategy, the target fundamental question corresponding to the original question can be determined from the at least two candidate fundamental questions. The embodiments of the present application are not limited to this.
[0102] It should be noted that the multiple fundamental questions that are semantically clustered with the original question in the above embodiments may be all the fundamental questions in the corpus, or may be part of the fundamental questions in the corpus, or may be multiple fundamental questions corresponding to the target level set by the administrator in the corpus, and the embodiments of the present application do not impose any restrictions on this.
[0103] In addition, in order to ensure that each original question can be aggregated into a corresponding target fundamental question, a fundamental question for identifying meaningless questions can also be set. When the semantic similarities between the semantic feature vectors of the original question and multiple fundamental questions are lower than the semantic threshold, the target fundamental question corresponding to the original question is determined to be the fundamental question for identifying meaningless questions.
[0104] As an example, for the original questions collected through the bullet screen in the live broadcast scene, if the content of the original question is a statement with a fixed format and expression tone such as "666", "1", "xx sign-in", etc., it can be determined that the target fundamental question corresponding to the original question is the fundamental question for identifying meaningless questions. In this way, the original question can be answered based on the subsequent steps 103 and 104, or these meaningless questions can be directly blocked, that is, this type of original question is not answered.
[0105] As an example, in response to the situation where a large number of malicious barrages appear during a live broadcast, a fundamental question that identifies malicious questions can also be set. At this time, when it is determined that the target fundamental question corresponding to the original question is the fundamental question that identifies the malicious question, a reply answer with a fixed rhetoric can be generated. For example, when the question content of the original question contains uncivilized language, the answer corresponding to the original question is determined to be a call for the audience to use civilized language; when the question content of the original question is confrontational and provocative, the answer corresponding to the original question is determined to be a call for the audience to speak rationally and not to lead the rhythm. The embodiments of the present application do not limit this.
[0106] In this way, it is ensured that each original question can be determined through semantic clustering to its corresponding target fundamental problem, and there will be no omission or neglect of problems, so that each original question can be processed and the problem clustering effect is better.
[0107] In the above embodiment, by semantically clustering the original questions, questions with different asking methods but the same semantics can be identified, and different expressions of the same question can be clustered into a fundamental question. Therefore, only some basic answers to the fundamental questions need to be configured in the corpus, and the corpus can be used to respond to different forms of questions asked by the questioner, which greatly saves the time for configuring the fundamental question-answer corpus pairs in the corpus; at the same time, the intelligent question-answering process is made more robust and can respond to original questions with different expressions in various scenarios.
[0108] (2) Based on the basic answer corresponding to the target fundamental question, determine at least one candidate answer.
[0109] In a possible implementation, since the target fundamental question can accurately reflect the semantic features of the original question, the basic answer to the target fundamental question can also answer the original question. Therefore, when there is a basic answer corresponding to the target fundamental question in the corpus, the basic answer corresponding to the target fundamental question can be directly determined as the answer corresponding to the original question.
[0110] Optionally, at least one candidate answer may be generated based on some fixed terms that can be used to answer the question, so as to enrich the number of candidate answers and facilitate the selection of different candidate answers for answering according to the importance evaluation result of the original question.
[0111] As an example, the candidate answers generated based on the fixed terms may be: I will answer you later, please wait patiently; this question has been answered before, please check the historical question records; please describe the question accurately. The embodiments of the present application do not limit this.
[0112] Since the corpus is pre-built, the basic answers to the fundamental questions are all preset, but as knowledge becomes richer and more diverse, the answers to each fundamental question may be more accurate and richer. Therefore, we can use the information search advantages of the Internet to search for online answers to the original questions and / or the target fundamental questions, so as to dynamically generate the answers to the original questions.
[0113] In another possible implementation, the process of determining at least one candidate answer may be: based on the original question and / or the target fundamental question, obtaining at least one network answer through network search; and generating at least one candidate answer based on the basic answer corresponding to the target fundamental question and at least one network answer.
[0114] That is, based on the network answers of the network search, generating at least one candidate answer includes the following three ways:
[0115] Method 1: Based on the original question, at least one first network answer is obtained through network search, and then the basic answer corresponding to the target root question is combined with the at least one first network answer to generate at least one candidate answer.
[0116] As an example, if the first network answer obtained based on the search of the original question includes network answer A and network answer B, then network answer A and the basic answer corresponding to the target fundamental question can be combined to obtain a first candidate answer; network answer B and the basic answer corresponding to the target fundamental question can be combined to obtain a second candidate answer; network answer A, network answer B and the basic answer corresponding to the target fundamental question can be combined to obtain a third candidate answer.
[0117] Method 2: Based on the target fundamental question, at least one second network answer is obtained through network search, and then the basic answer corresponding to the target fundamental question is combined with the at least one second network answer to generate at least one candidate answer.
[0118] As an example, if a network answer C is obtained based on the search for the target fundamental question, the network answer C and the basic answer corresponding to the target fundamental question can be combined to obtain a fourth candidate answer.
[0119] It should be noted that even if only one online answer is searched, the online answer may include more or less information. When it is combined with the basic answer corresponding to the target root question, one candidate answer or multiple candidate answers can be generated. The embodiments of the present application do not limit this.
[0120] Method three: Based on the original question, obtain at least one first network answer through network search; based on the target fundamental question, obtain at least one second network answer through network search; then combine the basic answer corresponding to the target fundamental question with the at least one first network answer and the at least one second network answer to generate at least one candidate answer.
[0121] As an example, if the first network answer includes network answer A and network answer B, and the second network answer includes network answer C, then network answer A, network answer C and the basic answer corresponding to the target fundamental question can be combined to obtain the fifth candidate answer; network answer B, network answer C and the basic answer corresponding to the target fundamental question can be combined to obtain the sixth candidate answer; network answer A, network answer B, network answer C and the basic answer corresponding to the target fundamental question can also be combined to obtain the seventh candidate answer.
[0122] It should be noted that when generating at least one candidate answer based on the basic answer corresponding to the target fundamental question and at least one network answer, a pre-trained neural network model can also be used to combine or splice the network answer and the basic answer to make the expression of the obtained candidate answer smoother, clearer, and more in line with actual natural language expression habits.
[0123] In this way, based on the basic answer corresponding to the target fundamental question, the computing device can use the network search capability to search for the network answer to the original question and / or the target fundamental question, thereby integrating the basic answer and the network answer to generate at least one candidate answer corresponding to the original question. In this way, the diversity of the answer content is enriched, so that the computing device can flexibly answer different questioners and different types of questions based on different answer strategies, making the answer more personalized.
[0124] In some embodiments, after performing a network search based on the original question and / or the target root question, the basic answer corresponding to the target root question in the corpus can also be improved based on the searched network answers to improve the accuracy of the basic answers in the question-answer corpus pairs in the corpus, thereby improving the accuracy of the computing device in answering the original question.
[0125] In some embodiments, the computing device may also store and integrate multiple original questions aggregated into each root question, and modify and improve the root questions in the corpus and the basic answers corresponding to each root question based on the information carried by the multiple original questions themselves, so as to improve the accuracy of clustering the original questions based on the root questions in the corpus and the accuracy of the basic answers corresponding to each root question, thereby improving the accuracy of the computing device in answering the original questions.
[0126] After generating at least one candidate answer with different levels of detail for the original question, the above step 103 can be used to flexibly determine the level of detail of the answer corresponding to the original question; for the same original question, different levels of detail can be used to answer according to different questioners, making the answer more personalized.
[0127] In some embodiments, the implementation process of the above step 103 may include the following steps:
[0128] (1) based on the importance evaluation result, determining the answer strategy for the original question from the correspondence between the importance evaluation result and the answer strategy, the answer strategy indicating a fixed term answer, a simple answer, or a comprehensive answer, and the detail level of the answer corresponding to the simple answer is lower than the detail level of the answer corresponding to the comprehensive answer;
[0129] In a possible implementation, the correspondence between the importance evaluation result and the answer strategy is pre-configured by an administrator. The computing device receives a policy configuration request, and the corresponding relationship between the importance evaluation result and the answer strategy is carried in the policy configuration request. Then, the computing device stores the corresponding relationship between the importance evaluation result and the answer strategy, so as to find the corresponding answer strategy in the corresponding relationship between the importance evaluation result and the answer strategy according to the importance evaluation result of the original question.
[0130] As an example, the answer strategy of comprehensive answer corresponds to a high level range or a high score range, and the fixed term answer corresponds to a low level range or a low score range. When the importance level of the original question is in the high level range, or the importance score of the original question is in the high score range, the answer strategy corresponding to the original question is comprehensive answer. When the importance level of the original question is in the low level range, or the importance score of the original question is in the low score range, the answer strategy corresponding to the original question is fixed term answer. When the importance level of the original question is neither in the high level range nor in the low level range, or the importance score of the original question is neither in the high score range nor in the low score range, the answer strategy corresponding to the original question is simple answer.
[0131] In another possible implementation, the correspondence between the importance evaluation result and the answer strategy can also be determined in real time, that is, the computing device determines the correspondence between the original question and the answer strategy according to the number of original questions received at the current moment.
[0132] The upper limit value and the lower limit value of the number of questions are values configured by the administrator. That is, the computing device receives a policy configuration request, which carries the upper limit value and the lower limit value of the number of questions and the answer strategy. Then, the computing device stores the upper limit value and the lower limit value of the number of questions to determine the answer strategy corresponding to the original question based on the importance evaluation result of the original question.
[0133] As an example, if the number of original questions received at the current moment is greater than the upper limit of the number of questions, the answer strategy for the original questions is determined to be a fixed-word answer; if the number of original questions received at the current moment is less than the lower limit of the number of questions, the answer strategy for the original questions is determined to be a comprehensive answer; if the number of original questions received at the current moment is greater than the lower limit of the number of questions and the upper limit of the number of questions, the answer strategy for the original questions is determined to be a simple answer.
[0134] In some embodiments, it can also be determined based on an evaluation threshold whether to use a candidate answer to answer the original question simply or in detail, or to use fixed terms to answer the original question. The evaluation threshold is any pre-set value that is not 0.
[0135] When the importance level of the original question is less than or equal to the evaluation threshold, or the importance score of the original question is less than or equal to the evaluation threshold, the answer strategy for the original question is determined to be a fixed term answer; when the importance level of the original question is less than or equal to the evaluation threshold, or the importance score of the original question is greater than the evaluation threshold, the answer strategy for the original question is determined to be a simple answer or a comprehensive answer, that is, selecting an answer to the original question from at least one candidate answer.
[0136] In some embodiments, for an original question whose importance level is less than the evaluation threshold, or whose importance score is less than the evaluation threshold, the corresponding answer may be directly determined to be empty. In other words, the answer to the original question is not output, and the original question is not answered.
[0137] (2) Based on the answer strategy, determine the answer corresponding to the original question from at least one candidate answer.
[0138] When selecting an answer corresponding to the original question from at least one candidate answer based on the answer strategy, the candidate answers may be first sorted in descending order of the number of characters contained in the candidate answers, and the character number sorting result is used as the detail level sorting result of the candidate answers, where the detail level of the candidate answers with more characters is higher than the detail level of the candidate answers with less characters. Then, based on the detail level sorting result of the candidate answers, the answer corresponding to the original question is determined from at least one candidate answer.
[0139] In one possible implementation, when the importance level or importance score of the original question is high, the results can be sorted according to the detail level of the candidate answers, and the answer corresponding to the original question can be determined from the candidate answers with higher order; when the importance level or importance score of the original question is low, the results can be sorted according to the detail level of the candidate answers, and the answer corresponding to the original question can be determined from the candidate answers with lower order.
[0140] As an example, taking the importance evaluation result of the original question as the importance score, if there are three candidate answers, and the detail ranking result of the candidate answers is: candidate answer x>candidate answer y>candidate answer z, when the importance score of the original question is 80 points, select candidate answer x or candidate answer y with a lower level of detail as the answer corresponding to the original question; when the importance score of the original question is 35 points, select candidate answer y or candidate answer z with a lower level of detail as the answer corresponding to the original question. The embodiments of the present application do not limit this.
[0141] In a possible implementation, the evaluation criteria of importance evaluation and the number of candidate answers are used to sort the results according to the degree of detail of the candidate answers, and the evaluation value range corresponding to each candidate answer is determined; based on the importance score of the original question and the evaluation value range corresponding to each candidate answer, the candidate answer corresponding to the original question is determined, and the candidate answer is used as the answer corresponding to the original question. In other words, based on the number of candidate answers, the evaluation criteria of importance evaluation are divided to obtain the evaluation value range corresponding to each candidate answer, and then the candidate answer corresponding to the evaluation value range where the importance evaluation result of the original question is located is determined as the answer corresponding to the original question.
[0142] As an example, taking the importance evaluation result of the original question as the importance score, if there are four candidate answers, and the detail level of the four candidate answers is ranked as follows: candidate answer m>candidate answer n>candidate answer p>candidate answer q, and the importance score is a percentage score, then the score range of candidate answer m is determined to be (75, 100], the score range of candidate answer n is (50, 75], the score range of candidate answer p is (25, 50], and the score range of candidate answer q is (0, 25]. Based on this, when the importance score of the original question is 50 points, candidate answer p is selected as the answer corresponding to the original question; when the importance score of the original question is 85 points, candidate answer m is selected as the answer corresponding to the original question.
[0143] Step 104: Output the answer corresponding to the original question.
[0144] In some embodiments, when the presentation format of the answer corresponding to the original question is different from the question format of the original question, the answer corresponding to the original question can be output according to the question format of the original question to enhance the question-and-answer experience.
[0145] As an example, when the original question is asked by the questioner in voice, the computing device may output the answer to the original question in the form of audio after determining the answer to the original question.
[0146] In some embodiments, when the original question includes multiple questions, after the answer corresponding to each question is determined, the order of answering the questions can be flexibly adjusted based on the importance evaluation result of each question.
[0147] The original questions include a first question and a second question. In a possible implementation, the implementation process of the above step 104 can be: if it is determined that the priority of the first question is higher than the priority of the second question based on the importance evaluation result of the first question and the importance evaluation result of the second question, then the answer corresponding to the first question is outputted preferentially.
[0148] As an example, after evaluating the importance of the first question and the second question, if the importance level of the first question is higher than the importance level of the second question, it is determined that the priority of the first question is higher than the priority of the second question.
[0149] As another example, after evaluating the importance of the first question and the second question, if the importance score of the first question is higher than the importance score of the second question, it is determined that the priority of the first question is higher than the priority of the second question.
[0150] In summary, in an embodiment of the present application, the computing device first performs an importance assessment on the original question received according to at least one evaluation influencing factor to obtain the importance assessment result of the original question, and the importance assessment result indicates the importance of the answer to the original question. Then, based on the importance assessment result, the answer corresponding to the original question is determined from at least one candidate answer with different levels of detail, and the answer corresponding to the original question is output. In this way, by performing an importance assessment on the original question, and based on the importance assessment result of the original question, the priority of answering the original question and the level of detail of the answer are determined, so as to flexibly handle the original question and improve the efficiency of answering questions. In the case of a large amount of original questions, it is avoided to answer all questions equally, and to answer a large number of meaningless questions or redundant answers to the same question, which causes the computing device to be unable to handle questions with higher priority, and may even cause the entire intelligent question-answering process to be paralyzed.
[0151] Based on the above embodiments, Figure 2 and Figure 3 The specific implementation process of a computing device from receiving an original question to outputting an answer corresponding to the original question is explained by way of example.
[0152] Figure 2 This is a question-and-answer flow diagram provided in an embodiment of the present application. Figure 2 The computing device collects the original question in the target scenario, and performs question clustering on the original question to determine the target fundamental question corresponding to the original question, and then dynamically generates at least one candidate answer corresponding to the original question in combination with the basic answer corresponding to the target fundamental question in the corpus. For the at least one dynamically generated candidate answer, based on the importance evaluation result of the original question, an answer corresponding to the original question is selected from the at least one candidate answer to answer the original question.
[0153] The target scenario may be a live broadcast, intelligent customer service, intelligent interaction, etc., which is not limited in the present embodiment. Taking the live broadcast scenario as an example, the computing device may collect original questions from the live broadcast bullet screen.
[0154] Specifically, see Figure 3The flowchart of the intelligent question-answering method shown in the figure, for the original question received, a semantic clustering operation is performed on the original question to determine the target fundamental question corresponding to the original question. On the one hand, after the target fundamental question is determined through the semantic clustering operation, the basic answer corresponding to the target fundamental question can be determined from the fundamental question-answer corpus pairs in the corpus based on the target fundamental question, and a search is performed on the network based on the target fundamental question and / or the original question to obtain the network answer, and then the basic answer corresponding to the target fundamental question and the network answer obtained by the search are integrated to dynamically generate at least one candidate answer that can be used to answer the original question and has different levels of detail. On the other hand, after the target fundamental question is determined through the semantic clustering operation, or when the original question is received (corresponding to Figure 3 The importance of the answer to the original question is evaluated to obtain the importance evaluation result of the original question.
[0155] For at least one candidate answer with different generated levels of detail, when answering the question, based on the importance evaluation result of the original question, one candidate answer is selected as the answer corresponding to the original question, thereby outputting the answer corresponding to the original question.
[0156] See also Figure 4 , Figure 4 : is a schematic diagram of the structure of an intelligent question-answering device provided in an embodiment of the present application. The intelligent question-answering device 400 includes:
[0157] The question receiving module 401 is used to receive the original question. For the detailed implementation process, please refer to the relevant content in step 101 above, which will not be repeated here.
[0158] The question evaluation module 402 is used to determine the importance evaluation result of the original question. The importance evaluation result indicates the importance of the answer to the original question. The importance evaluation result is determined by at least one evaluation influencing factor. For the detailed implementation process, please refer to the relevant content in step 102 above, which will not be repeated here.
[0159] The answer selection module 403 is used to determine the answer corresponding to the original question from at least one candidate answer based on the importance evaluation result, and at least one candidate answer has a different level of detail; for the detailed implementation process, please refer to the relevant content in step 103 above, which will not be repeated here.
[0160] The answer output module 404 is used to output the answer corresponding to the original question. Please refer to the relevant content in the above step 104 for the detailed implementation process, which will not be repeated here.
[0161] Optionally, the intelligent question-answering device 400 further includes:
[0162] The first configuration module is used to receive an evaluation rule configuration request, where the evaluation rule configuration request is used to request configuration of one or more of a type, a priority, and a weight of at least one evaluation influencing factor.
[0163] Optionally, the type of at least one evaluation influencing factor includes one or more of the questioner level, question category, question frequency and answer indication information of the original question, the questioner level indicates the importance of the questioner, the question category indicates whether the original question is a meaningless question, the question frequency indicates the historical question frequency of the original question, and the answer indication information indicates whether the original question has been answered within a specified time period before the current time.
[0164] Optionally, the problem evaluation module 402 is specifically configured to:
[0165] Determine the evaluation result of the original question under at least one evaluation influencing factor;
[0166] Based on the evaluation result of the original question under at least one evaluation influencing factor, an importance evaluation result of the original question is determined.
[0167] Optionally, the answer selection module 403 includes:
[0168] a strategy determination unit, configured to determine, based on the importance evaluation result, an answer strategy for the original question from a correspondence between the importance evaluation result and the answer strategy, the answer strategy indicating a fixed term answer, a simple answer, or a comprehensive answer, the detail level of the answer corresponding to the simple answer being lower than the detail level of the answer corresponding to the comprehensive answer;
[0169] The answer selection unit is used to determine the answer corresponding to the original question from at least one candidate answer based on the answer strategy.
[0170] Optionally, the intelligent question-answering device 400 further includes:
[0171] A question clustering module is used to determine the target fundamental question corresponding to the original question from the corpus by semantic clustering based on the original question. The corpus includes multiple fundamental questions and the basic answer corresponding to each fundamental question.
[0172] The answer determination module is used to determine at least one candidate answer based on the basic answer corresponding to the target root question.
[0173] Optionally, the answer determination module is specifically configured to:
[0174] Based on the original question and / or the target root question, obtain at least one online answer through online search;
[0175] At least one candidate answer is generated based on a basic answer corresponding to the target root question and at least one network answer.
[0176] Optionally, the intelligent question-answering device 400 further includes:
[0177] A second configuration module is configured to receive a corpus configuration request, the corpus configuration request carrying a plurality of fundamental question-answer corpus pairs, each fundamental question-answer corpus pair including a fundamental question and a basic answer corresponding to the fundamental question;
[0178] The corpus storage module is used to store multiple root question-answer corpus pairs in the corpus.
[0179] Optionally, the multiple fundamental question-answer corpus pairs are fundamental question-answer corpus pairs corresponding to multiple levels, each level corresponds to at least one fundamental question-answer corpus pair, the fundamental question-answer corpus pairs corresponding to different levels have different granularities, and the corpus configuration request further carries identifiers of multiple levels;
[0180] Corpus storage module, specifically used for:
[0181] Based on multiple levels of identification, fundamental question-answer corpus pairs corresponding to multiple levels are stored in the corpus.
[0182] Optionally, the original question includes a first question and a second question;
[0183] The answer output module 404 is specifically used for:
[0184] If it is determined that the priority of the first question is higher than the priority of the second question based on the importance evaluation results of the first question and the second question, the answer corresponding to the first question is outputted preferentially.
[0185] In an embodiment of the present application, the intelligent question-answering device first evaluates the importance of the original question received according to at least one evaluation influencing factor to obtain the importance evaluation result of the original question, and the importance evaluation result indicates the importance of the answer to the original question. Then, based on the importance evaluation result, the answer corresponding to the original question is determined from at least one candidate answer with different levels of detail, and the answer corresponding to the original question is output. In this way, by evaluating the importance of the original question, and based on the importance evaluation result of the original question, the priority of answering the original question and the level of detail of the answer are determined, so as to flexibly handle the original question and improve the efficiency of answering questions. In the case of a large number of original questions, it is avoided to answer all questions equally, and to answer a large number of meaningless questions or redundant answers to the same question, which causes the computing device to be unable to process questions with higher priority, and may even cause the entire intelligent question-answering process to be paralyzed.
[0186] It should be noted that the intelligent question-answering device provided in the above embodiment only uses the division of the above functional modules as an example when performing intelligent question-answering. In actual applications, the above functional distribution can be completed by different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the intelligent question-answering device provided in the above embodiment and the intelligent question-answering method embodiment belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.
[0187] Among them, the question receiving module 401, the question evaluating module 402, the answer selecting module 403 and the answer outputting module 404 can all be implemented by software or by hardware.
[0188] For example, the question receiving module 401 is taken as an example to introduce the implementation of the question receiving module 401. Similarly, the implementation of the question evaluation module 402, the answer selection module 403 and the answer output module 404 can be the same as the implementation of the question receiving module 401.
[0189] As an example of a software functional unit, the problem receiving module 401 may include code running on a computing instance. Among them, the computing instance may include at least one of a physical host (computing device), a virtual machine, and a container. Further, the above-mentioned computing instance may be one or more. For example, the problem receiving module 401 may include code running on multiple hosts / virtual machines / containers. It should be noted that the multiple hosts / virtual machines / containers used to run the code may be distributed in the same region (region) or in different regions. Furthermore, the multiple hosts / virtual machines / containers used to run the code may be distributed in the same availability zone (AZ) or in different AZs, each AZ including one data center or multiple data centers with close geographical locations. Among them, usually a region may include multiple AZs.
[0190] Similarly, multiple hosts / virtual machines / containers used to run the code can be distributed in the same virtual private cloud (VPC) or in multiple VPCs. Usually, a VPC is set up in a region. For cross-region communication between two VPCs in the same region and between VPCs in different regions, a communication gateway needs to be set up in each VPC to achieve interconnection between VPCs through the communication gateway.
[0191] As an example of a hardware functional unit, the question receiving module 401 may include at least one computing device, such as a server, etc. Alternatively, the question receiving module 401 may also be a device implemented by an application-specific integrated circuit (ASIC) or a programmable logic device (PLD). The PLD may be a complex programmable logical device (CPLD), a field programmable gate array (FPGA), a generic array logic (GAL) or any combination thereof.
[0192] The multiple computing devices included in the question receiving module 401 can be distributed in the same region or in different regions. The multiple computing devices included in the question receiving module 401 can be distributed in the same AZ or in different AZs. Similarly, the multiple computing devices included in the question receiving module 401 can be distributed in the same VPC or in multiple VPCs. The multiple computing devices can be any combination of computing devices such as servers, ASICs, PLDs, CPLDs, FPGAs, and GALs.
[0193] It should be noted that, in other embodiments, the question receiving module 401 can be used to execute any step in the intelligent question-answering method, the question evaluation module 402 can be used to execute any step in the intelligent question-answering method, the answer selection module 403 can be used to execute any step in the intelligent question-answering method, and the answer output module can be used to execute any step in the intelligent question-answering method. That is, the steps that the question receiving module 401, the question evaluation module 402, the answer selection module 403, and the answer output module 404 are responsible for implementing can be specified as needed, and the full functions of the intelligent question-answering device are realized by respectively implementing different steps in the intelligent question-answering method through the question receiving module 401, the question evaluation module 402, the answer selection module 403, and the answer output module 404.
[0194] See also Figure 5 , Figure 5 5 is a schematic diagram of a computing device provided in an embodiment of the present application, wherein the computing device 500 includes: a bus 502, a processor 504, a memory 506, and a communication interface 508. The processor 504, the memory 506, and the communication interface 508 communicate with each other via the bus 502. The computing device 500 may be a server or a terminal device. It should be understood that the present application does not limit the number of processors and memories in the computing device 500.
[0195] The bus 502 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 The bus 502 is represented by only one line, but it does not mean that there is only one bus or one type of bus. The bus 502 may include a path for transmitting information between various components of the computing device 500 (eg, the memory 506, the processor 504, and the communication interface 508).
[0196] The processor 504 may include any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).
[0197] The memory 506 may include a volatile memory, such as a random access memory (RAM). The processor 504 may also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid state drive (SSD).
[0198] The memory 506 stores executable program codes, and the processor 504 executes the executable program codes to respectively implement the functions of the aforementioned question receiving module 401, question evaluation module 402, answer selection module 403, and answer output module 404, thereby implementing the intelligent question-answering method. That is, the memory 506 stores instructions for executing the intelligent question-answering method.
[0199] The communication interface 508 uses a transceiver module such as, but not limited to, a network interface card or a transceiver to implement communication between the computing device 500 and other devices or a communication network.
[0200] The embodiment of the present application also provides a computing device cluster. The computing device cluster includes at least one computing device. The computing device can be a server, such as a central server, an edge server, or a local server in a local data center. In some embodiments, the computing device can also be a terminal device such as a desktop computer, a laptop computer, or a smart phone.
[0201] like Figure 6 As shown, the computing device cluster includes at least one computing device 500. The memory 506 in one or more computing devices 500 in the computing device cluster may store the same instructions for executing the intelligent question-answering method.
[0202] In some possible implementations, the memory 506 of one or more computing devices 500 in the computing device cluster may also store some instructions for executing the intelligent question-answering method. In other words, the combination of one or more computing devices 500 may jointly execute the instructions for executing the intelligent question-answering method.
[0203] It should be noted that the memory 506 in different computing devices 500 in the computing device cluster can store different instructions, which are respectively used to execute part of the functions of the intelligent question-answering device. That is, the instructions stored in the memory 506 in different computing devices 500 can implement the functions of one or more modules among the question receiving module 401, the question evaluation module 402, the answer selection module 403 and the answer output module 404.
[0204] In some possible implementations, one or more computing devices in the computing device cluster may be connected via a network, which may be a wide area network or a local area network, etc. Figure 7 A possible implementation is shown, such as Figure 7 As shown, two computing devices 500A and 500B are connected via a network. Specifically, the network is connected via a communication interface in each computing device. In this type of possible implementation, the memory 506 in the computing device 500A stores instructions for executing the functions of the question receiving module 401 and the answer output module 404. At the same time, the memory 506 in the computing device 500B stores instructions for executing the functions of the question evaluation module 402 and the answer selection module 403.
[0205] Figure 7 The connection method between the computing device clusters shown can be based on the consideration that the intelligent question-answering method provided in the present application needs to cluster a large number of collected original questions and generate answers of varying degrees of detail to answer the original questions. Therefore, it is considered that the functions implemented by the question evaluation module 402 and the answer selection module 403 are handed over to the computing device 500B for execution.
[0206] It should be understood that Figure 7 The functions of the computing device 500A shown in FIG. 5A may also be completed by multiple computing devices 500. Similarly, the functions of the computing device 500B may also be completed by multiple computing devices 500.
[0207] The present application embodiment also provides another computing device cluster. The connection relationship between the computing devices in the computing device cluster can be similar to that of Figure 6 and Figure 7 The connection mode of the computing device cluster is different in that the memory 506 in one or more computing devices 500 in the computing device cluster may store the same instructions for executing the intelligent question-answering method.
[0208] In some possible implementations, the memory 506 of one or more computing devices 500 in the computing device cluster may also store some instructions for executing the intelligent question-answering method. In other words, the combination of one or more computing devices 500 may jointly execute the instructions for executing the intelligent question-answering method.
[0209] The embodiment of the present application also provides a computer program product including instructions. The computer program product may be software or a program product including instructions that can be run on a computing device or stored in any available medium. When the computer program product is run on at least one computing device, the at least one computing device executes the intelligent question-answering method.
[0210] The embodiment of the present application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that can be stored by the computing device or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk). The computer-readable storage medium includes instructions that instruct the computing device to execute the intelligent question-answering method.
[0211] It should be understood that the "multiple" mentioned herein refers to two or more. In the description of the embodiments of the present application, unless otherwise specified, " / " means or, for example, A / B can mean A or B; "and / or" in this article is only a description of the association relationship of the associated objects, indicating that there can be three relationships, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in order to facilitate the clear description of the technical solution of the embodiments of the present application, in the embodiments of the present application, the words "first", "second" and the like are used to distinguish between the same items or similar items with basically the same functions and effects. Those skilled in the art can understand that the words "first", "second" and the like do not limit the quantity and execution order, and the words "first", "second" and the like do not limit them to be necessarily different.
[0212] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.) and signals involved in the embodiments of the present application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.
[0213] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present application.
Claims
1. An intelligent question-answering method, characterized in that: The method is applied to a multi-person online service, and the method comprises: Receive original questions; Determine an importance evaluation result of the original question, wherein the importance evaluation result indicates the importance of an answer to the original question, and the importance evaluation result is determined by at least one evaluation influencing factor; Based on the importance evaluation result, determining an answer corresponding to the original question from at least one candidate answer, wherein the at least one candidate answer has a different level of detail; Output the answer corresponding to the original question.
2. The method according to claim 1, characterized in that The method further comprises: An evaluation rule configuration request is received, where the evaluation rule configuration request is used to request configuration of one or more of a type, a priority, and a weight of the at least one evaluation influencing factor.
3. The method according to claim 2, characterized in that The type of at least one evaluation influencing factor includes one or more of the questioner level, question category, question frequency and answer indication information of the original question, wherein the questioner level indicates the importance of the questioner, the question category indicates whether the original question is a meaningless question, the question frequency indicates the historical question frequency of the original question, and the answer indication information indicates whether the original question has been answered within a specified time period before the current time.
4. The method according to claim 2 or 3, characterized in that Determining the importance evaluation result of the original question includes: Determine an evaluation result of the original question under the at least one evaluation influencing factor; Based on the evaluation result of the original question under the at least one evaluation influencing factor, an importance evaluation result of the original question is determined.
5. The method according to any one of claims 1 to 3, characterized in that The step of determining an answer corresponding to the original question from at least one candidate answer based on the importance evaluation result includes: Based on the importance evaluation result, determining an answer strategy for the original question from a correspondence between the importance evaluation result and the answer strategy, the answer strategy indicating a fixed term answer, a simple answer, or a comprehensive answer, the simple answer corresponding to an answer having a lower level of detail than the comprehensive answer; Based on the answer strategy, an answer corresponding to the original question is determined from the at least one candidate answer.
6. The method according to claim 5, characterized in that The method further comprises: Based on the original question, determine the target fundamental question corresponding to the original question from a corpus by semantic clustering, wherein the corpus includes multiple fundamental questions and basic answers corresponding to each fundamental question; The at least one candidate answer is determined based on a basic answer corresponding to the target root question.
7. The method according to claim 6, characterized in that The determining of the at least one candidate answer based on the basic answer corresponding to the target fundamental question includes: Based on the original question and / or the target fundamental question, obtaining at least one network answer through network search; The at least one candidate answer is generated based on a basic answer corresponding to the target root question and the at least one network answer.
8. The method according to claim 6, characterized in that The method further comprises: Receive a corpus configuration request, the corpus configuration request carrying a plurality of fundamental question-answer corpus pairs, each fundamental question-answer corpus pair including a fundamental question and a basic answer corresponding to the fundamental question; The plurality of root question-answer corpus pairs are stored in the corpus.
9. The method according to claim 8, characterized in that The multiple fundamental question-answer corpus pairs are fundamental question-answer corpus pairs corresponding to multiple levels, each level corresponds to at least one fundamental question-answer corpus pair, and the fundamental question-answer corpus pairs corresponding to different levels have different granularities, and the corpus configuration request further carries identifiers of the multiple levels; The storing of the plurality of fundamental question-answer corpus pairs in the corpus comprises: Based on the identifiers of the multiple levels, the fundamental question-answer corpus pairs corresponding to the multiple levels are stored in the corpus.
10. The method according to claim 1, characterized in that The original question includes a first question and a second question; The outputting the answer corresponding to the original question includes: If it is determined that the priority of the first question is higher than the priority of the second question based on the importance evaluation result of the first question and the importance evaluation result of the second question, the answer corresponding to the first question is outputted preferentially.
11. An intelligent question-answering device, characterized in that: The device comprises: A question receiving module is used to receive the original question; A question evaluation module, used to determine an importance evaluation result of the original question, wherein the importance evaluation result indicates the importance of an answer to the original question, and the importance evaluation result is determined by at least one evaluation influencing factor; an answer selection module, configured to determine, based on the importance evaluation result, an answer corresponding to the original question from at least one candidate answer, wherein the at least one candidate answer has a different level of detail; The answer output module is used to output the answer corresponding to the original question.
12. The device according to claim 11, characterized in that The device also includes: The first configuration module is used to receive an evaluation rule configuration request, where the evaluation rule configuration request is used to request configuration of one or more of the type, priority and weight of the at least one evaluation influencing factor.
13. The device according to claim 12, characterized in that The type of at least one evaluation influencing factor includes one or more of the questioner level, question category, question frequency and answer indication information of the original question, wherein the questioner level indicates the importance of the questioner, the question category indicates whether the original question is a meaningless question, the question frequency indicates the historical question frequency of the original question, and the answer indication information indicates whether the original question has been answered within a specified time period before the current time.
14. The device according to claim 12 or 13, characterized in that The problem assessment module is specifically used for: Determine an evaluation result of the original question under the at least one evaluation influencing factor; Based on the evaluation result of the original question under the at least one evaluation influencing factor, an importance evaluation result of the original question is determined.
15. The device according to any one of claims 11 to 13, characterized in that The answer selection module comprises: a strategy determination unit, configured to determine, based on the importance evaluation result, an answer strategy for the original question from a correspondence between the importance evaluation result and the answer strategy, the answer strategy indicating a fixed term answer, a simple answer, or a comprehensive answer, the simple answer corresponding to an answer having a lower level of detail than the comprehensive answer; An answer selection unit is used to determine an answer corresponding to the original question from the at least one candidate answer based on the answer strategy.
16. The device according to claim 15, characterized in that The device also includes: A question clustering module, configured to determine, based on the original question, a target fundamental question corresponding to the original question from a corpus by means of semantic clustering, wherein the corpus includes a plurality of fundamental questions and a basic answer corresponding to each fundamental question; The answer determination module is used to determine the at least one candidate answer based on the basic answer corresponding to the target root question.
17. The device according to claim 16, characterized in that The answer determination module is specifically used for: Based on the original question and / or the target fundamental question, obtaining at least one network answer through network search; The at least one candidate answer is generated based on a basic answer corresponding to the target root question and the at least one network answer.
18. The device according to claim 16, characterized in that The device also includes: A second configuration module is configured to receive a corpus configuration request, wherein the corpus configuration request carries a plurality of fundamental question-answer corpus pairs, each fundamental question-answer corpus pair comprising a fundamental question and a basic answer corresponding to the fundamental question; A corpus storage module is used to store the multiple root question-answer corpus pairs in the corpus.
19. The device according to claim 18, characterized in that The multiple fundamental question-answer corpus pairs are fundamental question-answer corpus pairs corresponding to multiple levels, each level corresponds to at least one fundamental question-answer corpus pair, and the fundamental question-answer corpus pairs corresponding to different levels have different granularities, and the corpus configuration request further carries identifiers of the multiple levels; The corpus storage module is specifically used for: Based on the identifiers of the multiple levels, the fundamental question-answer corpus pairs corresponding to the multiple levels are stored in the corpus.
20. The device according to claim 11, characterized in that The original question includes a first question and a second question; The answer output module is specifically used for: If it is determined that the priority of the first question is higher than the priority of the second question based on the importance evaluation result of the first question and the importance evaluation result of the second question, the answer corresponding to the first question is outputted preferentially.
21. A computing device cluster, characterized in that: comprising at least one computing device, each computing device comprising a processor and a memory; The processor of the at least one computing device is configured to execute instructions stored in the memory of the at least one computing device, so that the computing device cluster executes the method according to any one of claims 1 to 10.
22. A computer program product comprising instructions, characterized in that When the instructions are executed by a computing device cluster, the computing device cluster executes the method according to any one of claims 1 to 10.
23. A computer-readable storage medium, characterized in that: The method comprises computer program instructions. When the computer program instructions are executed by a computing device cluster, the computing device cluster performs the method according to any one of claims 1 to 10.
Citation Information
Cited By
Video live broadcast interaction method and system for investment advisor teaching
CN120568094A
A video live-streaming interactive method and system for investment advisory education
CN120568094B
