Agent-based question-answering processing method, device, medium, equipment and product

Abstract

The present disclosure relates to the fields of large model technology, large language model technology, intelligent agent technology, and artificial intelligence technology, and specifically discloses an intelligent agent-based question and answer processing method, device, medium, equipment, and product, the method comprising: obtaining a question and answer data pair and a quality assessment rule for quality assessment of the question and answer data pair; obtaining a quality assessment result of the question and answer data pair according to a target question, a target answer, a question and answer data pair, and a quality assessment rule through a first deep learning model associated with an intelligent agent; when the quality assessment result of the question and answer data pair meets a preset quality condition, the question and answer data pair is stored, and the stored question and answer data pair is used to generate frequently asked questions and answers content. Thereby, the quality assessment efficiency of the question and answer data pair and the consistency of the quality assessment result are improved, the content quality of the frequently asked questions and answers content is guaranteed, and the user experience and satisfaction are improved.

Description

Technical Field

[0001] The present disclosure relates to the fields of large model technology, large language model technology, intelligent agent technology, and artificial intelligence technology, and specifically, to an intelligent agent-based question-answering processing method, device, medium, equipment, and product. Background Art

[0002] Frequently Asked Questions (FAQ) are used to list questions that users may frequently ask and their answers. They are usually used in websites, product documents, and service manuals to help users quickly find solutions to problems they may encounter.

[0003] The quality of FAQs will directly affect user experience and satisfaction. Related technologies usually rely on manual quality assessment of FAQs, which is a cumbersome and inefficient process and cannot guarantee the consistency of quality assessment results. Summary of the invention

[0004] This section is provided to introduce the concepts in a brief form, which will be described in detail in the detailed implementation section below. This section is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to be used to limit the scope of the technical solution claimed for protection.

[0005] In a first aspect, the present disclosure provides an agent-based question-answering processing method, the method comprising:

[0006] Obtaining a question-answer data pair and a quality assessment rule for performing quality assessment on the question-answer data pair, wherein the question-answer data pair is obtained based on a target question input by a user and a target answer corresponding to the target question;

[0007] Obtaining, by means of a first deep learning model associated with the agent, a quality assessment result of the question-answer data pair according to the target question, the target answer, the question-answer data pair, and the quality assessment rule, wherein the first deep learning model is used to perform quality assessment on the input question-answer data pair;

[0008] In the case where the quality assessment result of the question-answer data pair meets a preset quality condition, the question-answer data pair is stored, and the stored question-answer data pair is used to generate frequently asked questions and answer content.

[0009] In a second aspect, the present disclosure provides an agent-based question-answering processing device, the device comprising:

[0010] An acquisition module, used to acquire a question-answer data pair and a quality assessment rule for performing quality assessment on the question-answer data pair, wherein the question-answer data pair is obtained based on a target question input by a user and a target answer corresponding to the target question;

[0011] An evaluation module, configured to obtain a quality evaluation result of the question-answer data pair according to the target question, the target answer, the question-answer data pair and the quality evaluation rule through a first deep learning model associated with the agent, wherein the first deep learning model is used to perform quality evaluation on the input question-answer data pair;

[0012] A storage module is used to store the question and answer data pair when the quality assessment result of the question and answer data pair meets the preset quality condition, and the stored question and answer data pair is used to generate frequently asked questions and answer content.

[0013] In a third aspect, the present disclosure provides a computer-readable medium having a computer program stored thereon, which, when executed by a processing device, implements the steps of any one of the methods described in the first aspect.

[0014] In a fourth aspect, the present disclosure provides an electronic device, including:

[0015] a storage device having a computer program stored thereon;

[0016] A processing device is used to execute the computer program in the storage device to implement the steps of any method described in the first aspect above.

[0017] In a fifth aspect, the present disclosure provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of any one of the methods in the first aspect.

[0018] Through the above technical solution, firstly, the question-answer data pair and the quality assessment rules for quality assessment of the question-answer data pair are obtained, and then the quality assessment result of the question-answer data pair is obtained according to the target question, the target answer, the question-answer data pair and the quality assessment rules through the first deep learning model associated with the intelligent agent. When the quality assessment result of the question-answer data pair meets the preset quality conditions, the question-answer data pair can be used to generate the FAQ content. With this method, the first deep learning model associated with the intelligent agent can automatically perform the quality assessment on the question-answer data pair, improve the quality assessment efficiency, and can realize the standardized quality assessment based on the quality assessment rules, so as to ensure the consistency of the quality assessment results, so as to ensure the content quality of the FAQ content, and then improve the user experience and satisfaction.

[0019] Other features and advantages of the present disclosure will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale. In the drawings:

[0021] Figure 1 is a flow chart of an agent-based question-answering processing method according to an exemplary embodiment of the present disclosure;

[0022] Figure 2 is a schematic diagram of a process of generating FAQ according to an exemplary embodiment of the present disclosure;

[0023] Figure 3 is a schematic diagram showing a question-answer generation prompt word according to an exemplary embodiment of the present disclosure;

[0024] Figure 4 is a schematic diagram showing a quality assessment rule according to an exemplary embodiment of the present disclosure;

[0025] Figure 5 is a schematic diagram of a FAQ quality assessment process according to an exemplary embodiment of the present disclosure;

[0026] Figure 6 is a schematic diagram showing a quality assessment prompt word according to an exemplary embodiment of the present disclosure;

[0027] Figure 7 is a schematic diagram of a FAQ fusion process according to an exemplary embodiment of the present disclosure;

[0028] Figure 8 is a schematic diagram showing a fusion-regenerated prompt word according to an exemplary embodiment of the present disclosure;

[0029] Fig. 9 is a block diagram of an agent-based question-answering processing device according to an exemplary embodiment of the present disclosure;

[0030] Fig.10 It is a schematic structural diagram of an electronic device according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0031] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0032] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0033] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0034] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0035] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0036] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0037] All actions of acquiring signals, information or data in the present disclosure are carried out in compliance with the relevant data protection laws and policies of the country where they are located, and with the authorization given by the owner of the corresponding device.

[0038] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, scope of use, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0039] For example, in response to receiving an active request from a user, a prompt message is sent to the user to clearly prompt the user that the operation requested to be performed will require obtaining and using the user's personal information. Thus, the user can autonomously choose whether to provide personal information to software or hardware such as an electronic device, application, server, or storage medium that performs the operation of the technical solution of the present disclosure according to the prompt message.

[0040] As an optional but non-limiting implementation, in response to receiving an active request from the user, the prompt information may be sent to the user in the form of a pop-up window, in which the prompt information may be presented in text form. In addition, the pop-up window may also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0041] It is understandable that the above notification and the process of obtaining user authorization are merely illustrative and do not constitute a limitation on the implementation of the present disclosure. Other methods that meet the relevant laws and regulations may also be applied to the implementation of the present disclosure.

[0042] At the same time, it is understandable that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and relevant provisions.

[0043] In the related art, FAQ systems mainly rely on manual editing and maintenance, and it is difficult to respond to new questions and changes in user questions in a timely manner. And when conducting quality assessment on FAQs, although supervised text evaluation methods can be relied upon to conduct quality assessment on FAQs, it is necessary to first manually summarize a set of FAQ items based on a large number of user questions and corresponding answers, and then conduct quality assessment together with the FAQ items to be evaluated. The process is cumbersome and inefficient. In addition, the quality of the content of the manually summarized FAQ items cannot be guaranteed. Therefore, using the manually summarized FAQ items as a reference standard for quality assessment cannot guarantee the consistency of the quality assessment results.

[0044] In view of this, the present disclosure provides an agent-based question-answering processing method, device, medium, equipment and product to solve the above-mentioned technical problems.

[0045] The embodiments of the present disclosure are further explained below with reference to the accompanying drawings.

[0046] Figure 1 is a question-answering processing method based on an intelligent agent according to an exemplary embodiment of the present disclosure, referring to Figure 1 , the question-answering processing method includes:

[0047] S101: Obtain a question-answer data pair and a quality assessment rule for performing quality assessment on the question-answer data pair, wherein the question-answer data pair is obtained based on a target question input by a user and a target answer corresponding to the target question.

[0048] It should be noted that the question-answer data pair can be a FAQ manually summarized or generated by a model. That is to say, the embodiments of the present disclosure can perform quality assessment on FAQ items manually summarized or on FAQ items generated based on a model. The present disclosure does not impose any restrictions on this.

[0049] In a possible manner, the question-answer data pair is obtained in the following manner: obtaining a target question input by a user and a target answer corresponding to the target question; constructing a question-answer generation prompt word based on the target question and the target answer, the question-answer generation prompt word being used to instruct a fourth deep learning model associated with the intelligent agent to generate a question-answer data pair based on the target question and the target answer, the fourth deep learning model being able to generate a question-answer data pair based on at least the input question and answer; inputting the question-answer generation prompt word into the fourth deep learning model to obtain a question-answer data pair.

[0050] It is worth noting that the fourth deep learning model can be a pre-trained large language model (LLM) that can at least generate question-answer data pairs based on input questions and answers. In order to avoid confusion, it is collectively referred to as generating a large language model below, such as Figure 2 As shown, the generated large language model can also be deployed in a multi-agent collaborative (Muti-Agent) intelligent agent, for example, deployed in the generating agent (agent server) of the intelligent agent. The structure and deployment method of the model can be selected according to the needs, and the present disclosure does not limit this.

[0051] It should be understood that in actual application scenarios, users can enter the chat interface by triggering the interactive control in the application page and have a conversation with the customer service through the chat interface. Therefore, with the authorization of the user, the conversation content including the question input by the user and the customer service's answer to the question can be obtained. Among them, the customer service can be a manual customer service or a machine customer service, and this disclosure does not limit this.

[0052] For example, Figure 2 As shown in the figure, after the user ends the chat, the conversation content can be obtained with the user's authorization, and then the following Figure 3 The question and answer generation prompt words shown are input into the generated large language model to obtain the question and answer data pairs output by the model.

[0053] in, Figure 3The question and answer generation prompt words shown define the role and skills of generating a large language model to instruct the generated large language model to perform corresponding data processing and obtain question and answer data pairs based on the conversation content. The question and answer generation prompt words can be set according to the needs, and may include the definition of the role and skills of generating a large language model, the conversation content, the FAQ generated last time and the corresponding quality assessment results, FAQ examples, constraints, etc., and the corresponding templates can also be pre-set and then filled in with content, which is not limited by the present disclosure.

[0054] It should be noted that the generated agent can extract key information from the conversation content and automatically generate FAQ entries. Among them, the deep semantic understanding ability of the generated large language model is used to automatically identify key words and phrases in user questions, and by analyzing the context content, ensure that the generated FAQ entries are closely related to the user questions.

[0055] S102: Obtain a quality assessment result of the question and answer data pair according to the target question, target answer, question and answer data pair and quality assessment rules through the first deep learning model associated with the intelligent agent. The first deep learning model is used to perform quality assessment on the input question and answer data pair.

[0056] Among them, the first deep learning model can be a pre-trained large language model for quality assessment of input question and answer data pairs. For the sake of confusion, it is collectively referred to as the evaluation large language model below. The evaluation large language model can also be deployed in a multi-agent collaborative intelligent body, for example, deployed in the quality assessment Agent of the intelligent body. The structure and deployment method of the model can be selected according to needs, and the present disclosure does not limit this.

[0057] S103: When the quality assessment result of the question-answer data pair meets the preset quality condition, the question-answer data pair is stored, and the stored question-answer data pair is used to generate frequently asked questions and answer content.

[0058] By adopting the above method, the first deep learning model associated with the intelligent agent can automatically perform quality assessment on the question and answer data pairs, thereby improving the efficiency of quality assessment, and standardized quality assessment can be achieved based on the quality assessment rules, thereby ensuring the consistency of the quality assessment results, thereby ensuring the content quality of the frequently asked questions and answers content, and thus improving user experience and satisfaction.

[0059] In a possible manner, the quality assessment rules include multiple quality assessment rules for different dimensions, and the quality assessment results of the question and answer data pair are obtained through the first deep learning model associated with the intelligent agent according to the target question, target answer, question and answer data pair and the quality assessment rules, including: for the quality assessment rules for each target dimension, the quality assessment sub-result of the question and answer data pair in the target dimension is obtained through the first deep learning model according to the target question, target answer, question and answer data pair and the quality assessment rules for the target dimension; the quality assessment result of the question and answer data pair is obtained according to the multiple quality assessment sub-results of the question and answer data pair in all dimensions.

[0060] For example, a quality assessment rule may be preset, such as Figure 4 As shown, it can be a quantifiable scoring table, which can comprehensively evaluate the FAQ based on multiple dimensions such as accuracy, completeness, and relevance. The scoring items and corresponding scores in the scoring table can be set according to different business scenarios, and the present disclosure does not impose any restrictions on this.

[0061] For example, in practical applications, by evaluating the large language model, we can first calculate the quality assessment sub-results of the question-answer data pair in a single dimension, and then combine all the quality assessment sub-results to obtain a comprehensive quality assessment of the question-answer data pair, thereby achieving multi-dimensional quality assessment of FAQ.

[0062] In a possible manner, the quality assessment rules under each dimension correspond to multiple quality assessment levels, and the quality assessment results of the question and answer data pair are obtained through the first deep learning model according to the target question, target answer, question and answer data pair and the quality assessment rules, including: for each quality assessment level, a first prediction probability that the question and answer data pair meets the quality assessment level and a second prediction probability that the question and answer data pair does not meet the quality assessment level are obtained through the first deep learning model according to the target question, target answer, question and answer data pair and the quality assessment level, and the target prediction probability is determined based on the first prediction probability and the second prediction probability; the quality assessment results of the question and answer data pair are obtained according to the quality assessment levels corresponding to the quality assessment rules under all dimensions and the target prediction probabilities corresponding to each quality assessment level.

[0063] For example, in practical applications, a more fine-grained quality assessment can be implemented for question-answer data pairs based on the quality assessment level. Figure 4 As shown, the quality assessment level can be expressed in the form of a score, which can be set according to specific needs and is not limited in this disclosure.

[0064] In a possible manner, the number of the first prediction probability and the second prediction probability are both multiple, and determining the target prediction probability based on the first prediction probability and the second prediction probability includes: determining a preset number of first prediction probabilities from multiple first prediction probabilities, and determining a preset number of second prediction probabilities from multiple second prediction probabilities, wherein the preset number of first prediction probabilities are greater than other first prediction probabilities in the multiple first prediction probabilities except the preset number of first prediction probabilities, and the preset number of second prediction probabilities are greater than other second prediction probabilities in the multiple second prediction probabilities except the preset number of second prediction probabilities; determining a first target probability based on the preset number of first prediction probabilities; determining a second target probability based on the preset number of second prediction probabilities, and determining a difference probability between 1 and the second target probability; and taking the larger value between the first target probability and the difference probability as the target prediction probability.

[0065] For example, the quality assessment level can be reflected in the form of a score, which can be set according to the needs, and the present disclosure does not limit this. Figure 4 For the scoring items of each dimension in the scoring item list shown, the large language model is evaluated to predict the probability of the question and answer data meeting each quality assessment level and not meeting each quality assessment level, and multiple predictions can be made through the model, and then the value with a larger prediction probability is selected to determine the total target prediction probability of the scoring item. For example, the preset number can be 3, 5, etc., which can be set according to needs, and the present disclosure does not impose any restrictions on this.

[0066] It should be noted that since the probability of the model output may be relatively small, direct calculation may encounter the problem of numerical underflow. Therefore, when calculating whether the probability of meeting the quality assessment level is satisfied, the logarithmic probability can be calculated first, that is, the first predicted probability and the second predicted probability are logarithmic probabilities, and then converted into probabilities. The specific selection can be based on demand, and the present disclosure does not impose any restrictions on this.

[0067] For example, the process of quality assessment by the first deep learning model is as follows: Figure 5 As shown, an example is described below in conjunction with a specific calculation process.

[0068] For example, i represents the score of a certain scoring item, i∈[n,m], n is the lowest score of the scoring item, m is the highest score of the scoring item, and it is assumed that the preset number is 5.

[0069] First, determine the sum of the logarithmic probabilities of the top 5 tokens (representing the model’s answer) whose model prediction result is “yes”, which is the first target probability mentioned above:

[0070]

[0071] Among them, t represents Token, is the range of top 5 tokens whose model prediction result is “yes”. F(.) represents the quality assessment prompt words of the model input, in which v represents the model role and skill settings, c represents the conversation content including the target question and the target answer, s represents the FAQ being evaluated, and x i It represents the quality assessment rule that sets the scoring item to i points, and the sum of logarithmic probabilities lg P(yes|i)∈(-∞,0].

[0072] It should be noted that the following can be constructed based on the conversation content including the target question and the target answer, the question-answer data pair and the quality assessment rules: Figure 6 The quality assessment prompt words are shown, and then the question and answer assessment prompt words are input into the evaluation large language model to obtain the Top 5 answers and probabilities output by the evaluation large language model. The quality assessment prompt words can be set according to the needs, and the corresponding templates can be pre-set and then filled with content. This disclosure does not limit this.

[0073] For example, Figure 5 As shown in FIG. 1 , taking the accuracy scoring item in the quality assessment rule as an example, the prompt word 1 including “FAQ accuracy is 1, inaccurate”, the prompt word 2 including “FAQ accuracy is 2, a little inaccurate”, and so on are constructed, until the corresponding quality assessment prompt word is constructed for each scoring item. Among them, the quality assessment prompt word can also include the following: Figure 6 Other contents shown, such as model roles and skill settings, conversation content including target questions and target answers, evaluated FAQs, etc., can be determined based on specific needs, and this disclosure does not limit this.

[0074] Then, determine the sum of the logarithmic probabilities of the top 5 tokens whose model prediction result is "no", which is the second target probability mentioned above:

[0075]

[0076] in, The sum of logarithmic probabilities lgP(no|i)∈(-∞,0] is the range of top 5 tokens whose model prediction result is “no”.

[0077] Furthermore, the logarithmic probability lg P(yes|i) is converted to the probability P(yes|i), and the logarithmic probability lg P(no|i) is converted to the probability P(no|i), then the target prediction probability of the scoring item with i points can be expressed as P(i)=max(P(yes|i),1-P(no|i)), that is, the maximum value between the first target probability and the difference probability is taken.

[0078] Furthermore, continue to refer to Figure 5, respectively calculate the score and reason of each scoring item. The corresponding score and probability can be multiplied and then added to obtain the score of the scoring item, that is, the quality assessment sub-result, or the score of the scoring item can be calculated in combination with the weight corresponding to each score, which is not limited in the present disclosure.

[0079] For example, the score of the scoring item can be calculated by the following formula:

[0080]

[0081]

[0082] Among them, SingleScore represents the quality assessment sub-result of the scoring item, and Weight represents the weight. In order to ensure that the sum of the probabilities of all scores is 100%, the weight is a probability normalization parameter.

[0083] After obtaining the quality assessment sub-results of the question-answer data pair in each dimension, such as Figure 5 As shown, the total score after accumulating the scores of each scoring item can be used as the quality assessment result of the question and answer data pair, or the quality assessment result of the question and answer data pair can be calculated based on the weight corresponding to each scoring item. The present disclosure does not limit this.

[0084] For example, the quality assessment result of the question-answer data pair can be calculated by the following formula:

[0085]

[0086] Among them, Score represents the quality assessment result of the question-answer data pair, JudgeList is the list of scoring items, that is, the quality assessment rules, SingleScore j represents the SingleScore of the jth item in the list of rated items, W j represents the weight of the jth item in the rating item list and satisfies .

[0087] Thus, an unsupervised quality assessment method for evaluating a large language model can be implemented based on the score probability of the scoring table, and the large language model used for quality assessment can be used not only to perform quality assessment on FAQs, but also to perform quality assessment on text content in other scenarios, and the present disclosure does not impose any restrictions on this.

[0088] In a possible manner, the method also includes: when the quality assessment result of the question and answer data pair does not meet the preset quality conditions, an optimized question and answer data pair is obtained according to the target question, target answer, question and answer data pair and the quality assessment result through a second deep learning model associated with the intelligent agent, and the second deep learning model can regenerate an optimized question and answer data pair based on the input question and answer based on the input question and answer.

[0089] Among them, the second deep learning model can be the same deep learning model as the above-mentioned fourth deep learning model, or it can be a different deep learning model. That is to say, the second deep learning model can be the above-mentioned generated large language model, and the present disclosure does not impose any restrictions on this.

[0090] For example, Figure 5 As shown, the quality assessment result of the question-answer data pair can be represented by a score, and the preset quality condition can be a score threshold. The question-answer data pairs above the score threshold are judged as high-quality FAQs and stored in the corresponding corpus, while the question-answer data pairs below the score threshold can be fed back to generate a large language model, that is, fed back to the generation Agent, and then the optimized question-answer data pairs are obtained according to the target question, target answer, question-answer data pair and quality assessment result.

[0091] In a possible manner, the method also includes: repeatedly executing the following steps until a preset number of executions is met or the quality assessment result of the optimized question and answer data pair meets the preset quality conditions: obtaining the quality assessment result of the optimized question and answer data pair according to the target question, the target answer, the optimized question and answer data pair and the quality assessment rule through a first deep learning model; when the quality assessment result of the optimized question and answer data pair does not meet the preset quality conditions, obtaining a new optimized question and answer data pair according to the target question, the target answer, the optimized question and answer data pair and the quality assessment result through a second deep learning model; when the quality assessment result of the optimized question and answer data pair meets the preset quality conditions, storing the optimized question and answer data pair; when the optimized question and answer data pairs corresponding to the preset number of executions do not meet the preset quality conditions, deleting the question and answer data pair and all optimized question and answer data pairs.

[0092] For example, Figure 5As shown, the quality of the optimized question-answer data pair can be further evaluated by evaluating the large language model. If the optimized question-answer data pair meets the preset quality conditions, the optimized question-answer data pair can be stored. If the question-answer data pairs generated multiple times for the same conversation content are all judged to not meet the preset quality conditions, that is, they are judged to be low-quality FAQs, the generated question-answer data pairs can be deleted, and question-answer data pairs will no longer be generated based on the conversation content. Among them, the preset number of executions can be set according to demand, and the present disclosure does not limit this. Thereby, it is possible to avoid generating question-answer data pairs based on invalid conversation content, wasting computing resources, and ensuring the content quality of the generated FAQ.

[0093] Using the above method, we can build a quality assessment agent based on the LLM unsupervised evaluation method of the score probability of the scoring table, such as Figure 5 As shown in the figure, this quality assessment method can automatically judge the quality of FAQs and filter low-scoring content without manual intervention, thereby greatly reducing the workload of manual annotation and maintenance of FAQs and improving the maintenance efficiency of FAQs. In addition, it can perform multi-dimensional quality assessment of FAQs based on multiple dimensions such as accuracy and completeness set by users, so that it can flexibly adapt to different business scenarios. In addition, the quality assessment results of FAQs can be fed back to the generation agent in real time, so that FAQ generation can be continuously optimized.

[0094] In a possible manner, the question and answer data pairs are stored, including: querying similar question and answer data pairs whose similarity with the question and answer data pairs is greater than a preset similarity from a corpus, the corpus including historically generated question and answer data pairs; fusing the question and answer data pairs and similar question and answer data pairs through a third deep learning model associated with the intelligent agent to obtain fused question and answer data pairs, the third deep learning model being used to fuse multiple input question and answer data pairs; deleting similar question and answer data pairs in the corpus, and storing the fused question and answer data pairs in the corpus.

[0095] The third deep learning model can be a pre-trained large language model for fusing multiple question-answer data pairs input. To avoid confusion, it is collectively referred to as a fused large language model below. The fused large language model can also be deployed in a multi-agent collaborative intelligent agent, such as in an intelligent agent fusion regeneration agent. The structure and deployment method of the model can be selected according to needs, and the present disclosure does not limit this.

[0096] For example, after the generated FAQ is determined to be a high-quality FAQ, the FAQ can be stored in the corpus. In order to avoid the corpus storing too many similar FAQs, such as Figure 7As shown, FAQ items with a similarity greater than a preset similarity can be retrieved from the corpus based on the generated FAQ, wherein the preset similarity can be set according to the requirements, and the present disclosure does not limit this. Retrieval and recall can also be performed based on the questions in the FAQ, and the present disclosure does not limit this.

[0097] For example, after obtaining similar FAQ items, the following can be constructed based on the generated FAQ and similar FAQ items: Figure 8 The fusion regenerates the prompt words, and then inputs the fusion regenerates the prompt words into the third deep learning model, wherein the fusion regenerates the prompt words can be specifically set according to the needs, or the corresponding template can be pre-set, and then the content is filled, which is not limited by the present disclosure. Thus, the generated FAQ and similar FAQ items can be fused and reproduced using the fusion large language model to obtain the fused question and answer data pair.

[0098] It is worth noting that the above-mentioned process of recalling and fusing similar FAQs can be realized by using the Retrieval-Augmented Generation (RAG) technology. Thus, similar FAQs can be fused, reducing the FAQ expansion problem caused by the question and answer content of massive users and reducing the redundancy of FAQs. In addition, the FAQ generated by fusing similar FAQs with a large language model contains the question and answer information of multiple similar FAQs, and has higher content integrity, richness and accuracy.

[0099] It should be noted that the above-mentioned model for generating FAQ, the model for quality assessment, and the model for fusion and regeneration can be deployed and used separately, or can be deployed and integrated in a multi-agent collaborative intelligent body, and the present disclosure does not limit this. Thus, FAQ content is automatically generated from massive, user-authorized conversation content, and FAQ quality is automatically assessed and optimized through feedback mechanisms, fusion and regeneration, etc., which can eliminate low-quality FAQs and repeated and similar FAQs, realize a fully automatic editing and maintenance FAQ system, and ensure high quality and low redundancy of FAQ content.

[0100] Based on the same concept, the embodiment of the present disclosure also provides a question-answering processing device based on an intelligent agent, such as Fig. 9 As shown, the agent-based question-answering processing device 900 may include:

[0101] An acquisition module 901 is used to acquire a question-answer data pair and a quality assessment rule for performing quality assessment on the question-answer data pair, wherein the question-answer data pair is obtained based on a target question input by a user and a target answer corresponding to the target question;

[0102] The evaluation module 902 is used to obtain a quality evaluation result of the question and answer data pair according to the target question, the target answer, the question and answer data pair and the quality evaluation rules through the first deep learning model associated with the agent, and the first deep learning model is used to perform quality evaluation on the input question and answer data pair.

[0103] The storage module 903 is used to store the question and answer data pair when the quality assessment result of the question and answer data pair meets the preset quality condition, and the stored question and answer data pair is used to generate frequently asked questions and answer content.

[0104] Optionally, the quality assessment rule includes a plurality of quality assessment rules for different dimensions, and the assessment module 902 is used to:

[0105] For each target dimension quality assessment rule, the first deep learning model is used to obtain a quality assessment sub-result of the question-answer data pair under the target dimension according to the target question, the target answer, the question-answer data pair, and the quality assessment rule of the target dimension;

[0106] According to the multiple quality assessment sub-results of the question-answer data pair in all dimensions, a quality assessment result of the question-answer data pair is obtained.

[0107] Optionally, the quality assessment rule under each dimension corresponds to multiple quality assessment levels, and the assessment module 902 includes:

[0108] A first evaluation submodule is configured to obtain, for each quality assessment level, a first prediction probability that the question-answer data pair meets the quality assessment level and a second prediction probability that the question-answer data pair does not meet the quality assessment level according to the target question, the target answer, the question-answer data pair, and the quality assessment level through the first deep learning model, and determine a target prediction probability based on the first prediction probability and the second prediction probability;

[0109] The second evaluation submodule is used to obtain the quality evaluation result of the question and answer data pair according to the quality evaluation levels corresponding to the quality evaluation rules under all dimensions and the target prediction probability corresponding to each quality evaluation level.

[0110] Optionally, the number of the first predicted probability and the number of the second predicted probability are both plural, and the first evaluation submodule is used to:

[0111] Determine a preset number of first prediction probabilities from a plurality of first prediction probabilities, and determine a preset number of second prediction probabilities from a plurality of second prediction probabilities, wherein the preset number of first prediction probabilities is greater than other first prediction probabilities among the plurality of first prediction probabilities except the preset number of first prediction probabilities, and the preset number of second prediction probabilities is greater than other second prediction probabilities among the plurality of second prediction probabilities except the preset number of second prediction probabilities;

[0112] Determining a first target probability based on the preset number of first predicted probabilities;

[0113] Determine a second target probability based on the preset number of second predicted probabilities, and determine a difference probability between 1 and the second target probability;

[0114] The larger value between the first target probability and the difference probability is taken as the target prediction probability.

[0115] Optionally, the agent-based question-answering processing device 900 may further include:

[0116] An optimization module is used to obtain an optimized question-answer data pair based on the target question, the target answer, the question-answer data pair and the quality evaluation result through a second deep learning model associated with the agent when the quality evaluation result of the question-answer data pair does not meet the preset quality condition. The second deep learning model can regenerate an optimized question-answer data pair based on the input question and answer based on the input question and answer.

[0117] Optionally, the agent-based question-answering processing device 900 may further include an execution module, which is used to:

[0118] Repeat the following steps until a preset number of executions is met or the quality assessment result of the optimized question and answer data pair meets the preset quality condition: obtain the quality assessment result of the optimized question and answer data pair according to the target question, the target answer, the optimized question and answer data pair and the quality assessment rule through the first deep learning model; if the quality assessment result of the optimized question and answer data pair does not meet the preset quality condition, obtain a new optimized question and answer data pair according to the target question, the target answer, the optimized question and answer data pair and the quality assessment result through the second deep learning model;

[0119] When the quality evaluation result of the optimized question-answer data pair meets the preset quality condition, storing the optimized question-answer data pair;

[0120] When the optimized question-answer data pairs corresponding to the preset number of executions do not satisfy the preset quality condition, the question-answer data pair and all the optimized question-answer data pairs are deleted.

[0121] Optionally, the storage module is used for:

[0122] Querying similar question-answer data pairs whose similarity to the question-answer data pairs is greater than a preset similarity from a corpus, wherein the corpus includes historically generated question-answer data pairs;

[0123] The question-answer data pair and the similar question-answer data pair are fused by a third deep learning model associated with the agent to obtain a fused question-answer data pair, wherein the third deep learning model is used to fuse multiple input question-answer data pairs;

[0124] Similar question-answer data pairs in the corpus are deleted, and the fused question-answer data pairs are stored in the corpus.

[0125] Optionally, the question-answer data pair is obtained in the following manner:

[0126] Obtaining a target question input by the user and a target answer corresponding to the target question;

[0127] Constructing a question-answer generation prompt word based on the target question and the target answer, wherein the question-answer generation prompt word is used to instruct a fourth deep learning model associated with the agent to generate a question-answer data pair based on the target question and the target answer, wherein the fourth deep learning model is capable of generating a question-answer data pair based on at least an input question and an answer;

[0128] The question and answer generation prompt words are input into the fourth deep learning model to obtain the question and answer data pair.

[0129] Based on the same concept, an embodiment of the present disclosure also provides a computer-readable medium on which a computer program is stored, and when the program is executed by a processing device, the steps of any of the above-mentioned agent-based question and answer processing methods are implemented.

[0130] Based on the same concept, the present disclosure also provides an electronic device, which may include:

[0131] a storage device having a computer program stored thereon;

[0132] A processing device is used to execute a computer program in a storage device to implement the steps of any of the above-mentioned agent-based question-answering processing methods.

[0133] Based on the same concept, an embodiment of the present disclosure also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of any of the above-mentioned agent-based question-answering processing methods.

[0134] Reference below Fig.10 , which shows a schematic diagram of the structure of an electronic device 1000 suitable for implementing the embodiment of the present disclosure. The terminal device in the embodiment of the present disclosure may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle terminals (such as vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Fig.10 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0135] like Fig.10 As shown, the electronic device 1000 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1008 to a random access memory (RAM) 1003. In the RAM 1003, various programs and data required for the operation of the electronic device 1000 are also stored. The processing device 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0136] Typically, the following devices may be connected to the I / O interface 1005: an input device 1006 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 1007 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 1008 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 may allow the electronic device 1000 to communicate with other devices wirelessly or by wire to exchange data. Although Fig.10 The electronic device 1000 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0137] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device 1009, or installed from a storage device 1008, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed.

[0138] It should be noted that the computer-readable medium mentioned above in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. Computer readable signal media may also be any computer readable medium other than computer readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0139] In some embodiments, any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol) can be used for communication, and can be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0140] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0141] The above-mentioned computer-readable medium carries one or more programs. When the above-mentioned one or more programs are executed by the electronic device, the electronic device: obtains a question and answer data pair and a quality assessment rule for performing quality assessment on the question and answer data pair, wherein the question and answer data pair is obtained based on a target question input by a user and a target answer corresponding to the target question; obtains a quality assessment result of the question and answer data pair according to the target question, the target answer, the question and answer data pair and the quality assessment rule through a first deep learning model associated with the intelligent agent, wherein the first deep learning model is used to perform quality assessment on the input question and answer data pair; and when the quality assessment result of the question and answer data pair meets a preset quality condition, the question and answer data pair is stored, and the stored question and answer data pair is used to generate frequently asked questions and answer content.

[0142] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including, but not limited to, object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0143] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0144] The modules involved in the embodiments described in the present disclosure may be implemented by software or hardware, wherein the name of a module does not, in some cases, limit the module itself.

[0145] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0146] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0147] According to one or more embodiments of the present disclosure, Example 1 provides an agent-based question and answer processing method, the method comprising: obtaining a question and answer data pair and a quality assessment rule for performing quality assessment on the question and answer data pair, the question and answer data pair being obtained based on a target question input by a user and a target answer corresponding to the target question; obtaining a quality assessment result of the question and answer data pair according to the target question, the target answer, the question and answer data pair and the quality assessment rule through a first deep learning model associated with the agent, the first deep learning model being used to perform quality assessment on the input question and answer data pair; storing the question and answer data pair when the quality assessment result of the question and answer data pair meets a preset quality condition, and the stored question and answer data pair is used to generate frequently asked questions content.

[0148] According to one or more embodiments of the present disclosure, Example 2 provides the method of Example 1, wherein the quality assessment rules include multiple quality assessment rules for different dimensions, and the first deep learning model associated with the agent obtains a quality assessment result of the question and answer data pair according to the target question, the target answer, the question and answer data pair and the quality assessment rules, including: for the quality assessment rules for each target dimension, obtaining a quality assessment sub-result of the question and answer data pair in the target dimension through the first deep learning model according to the target question, the target answer, the question and answer data pair and the quality assessment rules of the target dimension; obtaining the quality assessment result of the question and answer data pair according to multiple quality assessment sub-results of the question and answer data pair in all dimensions.

[0149] According to one or more embodiments of the present disclosure, Example 3 provides the method of Example 1, wherein the quality assessment rules under each dimension correspond to multiple quality assessment levels, and the first deep learning model associated with the intelligent agent obtains the quality assessment result of the question and answer data pair according to the target question, the target answer, the question and answer data pair and the quality assessment rules, including: for each quality assessment level, obtaining a first prediction probability that the question and answer data pair meets the quality assessment level and a second prediction probability that the question and answer data pair does not meet the quality assessment level according to the target question, the target answer, the question and answer data pair and the quality assessment level through the first deep learning model, and determining the target prediction probability based on the first prediction probability and the second prediction probability; obtaining the quality assessment result of the question and answer data pair according to each quality assessment level corresponding to the quality assessment rules under all dimensions and the target prediction probability corresponding to each quality assessment level.

[0150] According to one or more embodiments of the present disclosure, Example 4 provides the method of Example 3, wherein the number of the first prediction probability and the number of the second prediction probability are both multiple, and determining the target prediction probability based on the first prediction probability and the second prediction probability includes: determining a preset number of first prediction probabilities from multiple first prediction probabilities, and determining a preset number of second prediction probabilities from multiple second prediction probabilities, wherein the preset number of first prediction probabilities are greater than other first prediction probabilities in the multiple first prediction probabilities except the preset number of first prediction probabilities, and the preset number of second prediction probabilities are greater than other second prediction probabilities in the multiple second prediction probabilities except the preset number of second prediction probabilities; determining a first target probability based on the preset number of first prediction probabilities; determining a second target probability based on the preset number of second prediction probabilities, and determining the difference probability between 1 and the second target probability; and taking the larger value between the first target probability and the difference probability as the target prediction probability.

[0151] According to one or more embodiments of the present disclosure, Example 5 provides the method of any one of Examples 1-4, which further includes: when the quality assessment result of the question and answer data pair does not meet the preset quality condition, an optimized question and answer data pair is obtained according to the target question, the target answer, the question and answer data pair and the quality assessment result through a second deep learning model associated with the agent, and the second deep learning model can regenerate an optimized question and answer data pair based on the input question and answer based on the input question and answer.

[0152] According to one or more embodiments of the present disclosure, Example 6 provides the method of Example 5, which further includes: repeatedly performing the following steps until a preset number of executions is met or the quality assessment result of the optimized question and answer data pair meets the preset quality condition: obtaining the quality assessment result of the optimized question and answer data pair according to the target question, the target answer, the optimized question and answer data pair and the quality assessment rule through the first deep learning model; if the quality assessment result of the optimized question and answer data pair does not meet the preset quality condition, obtaining a new optimized question and answer data pair according to the target question, the target answer, the optimized question and answer data pair and the quality assessment result through the second deep learning model; if the quality assessment result of the optimized question and answer data pair meets the preset quality condition, storing the optimized question and answer data pair; if the optimized question and answer data pairs corresponding to the preset number of executions do not meet the preset quality condition, deleting the question and answer data pair and all the optimized question and answer data pairs.

[0153] According to one or more embodiments of the present disclosure, Example 7 provides the method of any one of Examples 1-4, wherein the storing of the question and answer data pairs comprises: querying from a corpus similar question and answer data pairs whose similarity with the question and answer data pairs is greater than a preset similarity, wherein the corpus comprises historically generated question and answer data pairs; fusing the question and answer data pairs and the similar question and answer data pairs through a third deep learning model associated with the agent to obtain a fused question and answer data pair, wherein the third deep learning model is used to fuse multiple input question and answer data pairs; deleting similar question and answer data pairs in the corpus, and storing the fused question and answer data pairs in the corpus.

[0154] According to one or more embodiments of the present disclosure, Example 8 provides the method of any one of Examples 1-4, wherein the question and answer data pair is obtained in the following manner: obtaining a target question input by the user and a target answer corresponding to the target question; constructing a question and answer generation prompt word based on the target question and the target answer, the question and answer generation prompt word being used to instruct a fourth deep learning model associated with the agent to generate a question and answer data pair based on the target question and the target answer, the fourth deep learning model being capable of generating a question and answer data pair based on at least the input question and answer; inputting the question and answer generation prompt word into the fourth deep learning model to obtain the question and answer data pair.

[0155] According to one or more embodiments of the present disclosure, Example 9 provides an agent-based question and answer processing device, the device comprising: an acquisition module, used to acquire question and answer data pairs and quality assessment rules for performing quality assessment on the question and answer data pairs, the question and answer data pairs being obtained based on a target question input by a user and a target answer corresponding to the target question; an evaluation module, used to obtain a quality assessment result of the question and answer data pair according to the target question, the target answer, the question and answer data pair and the quality assessment rules through a first deep learning model associated with the agent, the first deep learning model being used to perform quality assessment on the input question and answer data pairs; a storage module, used to store the question and answer data pairs when the quality assessment result of the question and answer data pairs meets a preset quality condition, the stored question and answer data pairs being used to generate frequently asked questions content.

[0156] According to one or more embodiments of the present disclosure, Example 10 provides a computer-readable medium having a computer program stored thereon, and when the computer program is executed by a processing device, the steps of the method described in any one of Examples 1-8 are implemented.

[0157] According to one or more embodiments of the present disclosure, Example 11 provides an electronic device, comprising: a storage device on which a computer program is stored; and a processing device for executing the computer program in the storage device to implement the steps of any one of the methods described in Examples 1-8.

[0158] According to one or more embodiments of the present disclosure, Example 12 provides a computer program product, including a computer program, which implements the steps of any one of the methods of Examples 1-8 when executed by a processor.

[0159] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present disclosure (but not limited to) by each other to form a technical solution.

[0160] In addition, although each operation is described in a specific order, this should not be understood as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0161] Although the subject matter has been described in language specific to structural features and / or method logic actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing the claims. Regarding the device in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.

Claims

1. An agent-based question-answering method, characterized in that: The method comprises: Obtaining a question-answer data pair and a quality assessment rule for performing quality assessment on the question-answer data pair, wherein the question-answer data pair is obtained based on a target question input by a user and a target answer corresponding to the target question; Obtaining, by means of a first deep learning model associated with the agent, a quality assessment result of the question-answer data pair according to the target question, the target answer, the question-answer data pair, and the quality assessment rule, wherein the first deep learning model is used to perform quality assessment on the input question-answer data pair; When the quality assessment result of the question-answer data pair meets the preset quality condition, the question-answer data pair is stored, and the stored question-answer data pair is used to generate frequently asked questions and answer content; The quality assessment rules under each dimension correspond to multiple quality assessment levels. The first deep learning model associated with the agent obtains a quality assessment result of the question and answer data pair according to the target question, the target answer, the question and answer data pair and the quality assessment rules, including: For each quality assessment level, obtaining, by the first deep learning model, according to the target question, the target answer, the question-answer data pair, and the quality assessment level, a first predicted probability that the question-answer data pair satisfies the quality assessment level and a second predicted probability that the question-answer data pair does not satisfy the quality assessment level, wherein the number of the first predicted probability and the number of the second predicted probability are both multiple; Determine a preset number of first prediction probabilities from a plurality of first prediction probabilities, and determine a preset number of second prediction probabilities from a plurality of second prediction probabilities, wherein the preset number of first prediction probabilities is greater than other first prediction probabilities among the plurality of first prediction probabilities except the preset number of first prediction probabilities, and the preset number of second prediction probabilities is greater than other second prediction probabilities among the plurality of second prediction probabilities except the preset number of second prediction probabilities; Determining a first target probability based on the preset number of first predicted probabilities; Determine a second target probability based on the preset number of second predicted probabilities, and determine a difference probability between 1 and the second target probability; Taking the larger value between the first target probability and the difference probability as the target prediction probability; According to the quality assessment levels corresponding to the quality assessment rules under all dimensions and the target prediction probability corresponding to each quality assessment level, the quality assessment result of the question and answer data pair is obtained.

2. The agent-based question-answering processing method according to claim 1, characterized in that: The quality assessment rules include a plurality of quality assessment rules for different dimensions, and the first deep learning model associated with the agent obtains a quality assessment result of the question-answer data pair according to the target question, the target answer, the question-answer data pair, and the quality assessment rules, including: For each target dimension quality assessment rule, the first deep learning model is used to obtain a quality assessment sub-result of the question-answer data pair under the target dimension according to the target question, the target answer, the question-answer data pair, and the quality assessment rule of the target dimension; According to the multiple quality assessment sub-results of the question-answer data pair in all dimensions, a quality assessment result of the question-answer data pair is obtained.

3. The agent-based question-answering processing method according to claim 1 or 2, characterized in that: The method further comprises: When the quality assessment result of the question and answer data pair does not meet the preset quality condition, an optimized question and answer data pair is obtained according to the target question, the target answer, the question and answer data pair and the quality assessment result through a second deep learning model associated with the agent. The second deep learning model can regenerate an optimized question and answer data pair based on the input question and answer based on the input question and answer.

4. The agent-based question-answering method according to claim 3, characterized in that: The method further comprises: Repeat the following steps until a preset number of executions is met or the quality assessment result of the optimized question and answer data pair meets the preset quality condition: obtain the quality assessment result of the optimized question and answer data pair according to the target question, the target answer, the optimized question and answer data pair and the quality assessment rule through the first deep learning model; if the quality assessment result of the optimized question and answer data pair does not meet the preset quality condition, obtain a new optimized question and answer data pair according to the target question, the target answer, the optimized question and answer data pair and the quality assessment result through the second deep learning model; When the quality evaluation result of the optimized question-answer data pair meets the preset quality condition, storing the optimized question-answer data pair; When the optimized question-answer data pairs corresponding to the preset number of executions do not satisfy the preset quality condition, the question-answer data pair and all the optimized question-answer data pairs are deleted.

5. The agent-based question-answering processing method according to claim 1 or 2, characterized in that: The storing of the question-answer data pair comprises: Querying similar question-answer data pairs whose similarity to the question-answer data pairs is greater than a preset similarity from a corpus, wherein the corpus includes historically generated question-answer data pairs; The question-answer data pair and the similar question-answer data pair are fused by a third deep learning model associated with the agent to obtain a fused question-answer data pair, wherein the third deep learning model is used to fuse multiple input question-answer data pairs; Similar question-answer data pairs in the corpus are deleted, and the fused question-answer data pairs are stored in the corpus.

6. The agent-based question-answering method according to claim 1 or 2, characterized in that: The question-answer data pair is obtained in the following manner: Obtaining a target question input by the user and a target answer corresponding to the target question; Constructing a question-answer generation prompt word based on the target question and the target answer, wherein the question-answer generation prompt word is used to instruct a fourth deep learning model associated with the agent to generate a question-answer data pair based on the target question and the target answer, wherein the fourth deep learning model is capable of generating a question-answer data pair based on at least an input question and an answer; The question and answer generation prompt words are input into the fourth deep learning model to obtain the question and answer data pair.

7. An agent-based question-answering processing device, characterized in that: The device comprises: An acquisition module, used to acquire a question-answer data pair and a quality assessment rule for performing quality assessment on the question-answer data pair, wherein the question-answer data pair is obtained based on a target question input by a user and a target answer corresponding to the target question; An evaluation module, configured to obtain a quality evaluation result of the question-answer data pair according to the target question, the target answer, the question-answer data pair and the quality evaluation rule through a first deep learning model associated with the agent, wherein the first deep learning model is used to perform quality evaluation on the input question-answer data pair; A storage module, configured to store the question-answer data pair if the quality assessment result of the question-answer data pair meets a preset quality condition, and the stored question-answer data pair is used to generate frequently asked questions and answer content; The quality assessment rules under each dimension correspond to multiple quality assessment levels, and the assessment module includes: A first evaluation submodule is used for obtaining, for each quality assessment level, a first prediction probability that the question and answer data pair meets the quality assessment level and a second prediction probability that the question and answer data pair does not meet the quality assessment level through the first deep learning model according to the target question, the target answer, the question and answer data pair and the quality assessment level, wherein the number of the first prediction probability and the number of the second prediction probability are both multiple; determining a preset number of first prediction probabilities from the multiple first prediction probabilities, and determining a preset number of second prediction probabilities from the multiple second prediction probabilities, wherein the preset number of first prediction probabilities are greater than other first prediction probabilities in the multiple first prediction probabilities except the preset number of first prediction probabilities, and the preset number of second prediction probabilities are greater than other second prediction probabilities in the multiple second prediction probabilities except the preset number of second prediction probabilities; determining a first target probability based on the preset number of first prediction probabilities; determining a second target probability based on the preset number of second prediction probabilities, and determining a difference probability between 1 and the second target probability; taking the larger value between the first target probability and the difference probability as the target prediction probability; The second evaluation submodule is used to obtain the quality evaluation result of the question and answer data pair according to the quality evaluation levels corresponding to the quality evaluation rules under all dimensions and the target prediction probability corresponding to each quality evaluation level.

8. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processing device, the steps of the method described in any one of claims 1 to 6 are implemented.

9. An electronic device, characterized in that: include: a storage device having at least one computer program stored thereon; At least one processing device, configured to execute the at least one computer program in the storage device to implement the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images