Generative question-answering system and method, and device, storage medium and program product

Abstract

Provided in the embodiments of the present application are a generative question-answering system and method, and a device, a storage medium and a program product. In the embodiments of the present application, a question-answering server end in the generative question-answering system cooperates with a question-answering model and a knowledge base to reply to question information raised by a user, and correct, with the assistance of a question-answering assistance end in the generative question-answering system, answer information for replying to the question information, such that a personalized reply is given to the user in a more intelligent, efficient and accurate manner, thereby meeting personalized requirements of the user. Especially when serving as a customer service robot, the question-answering server end cooperates with the question-answering model and the knowledge base to provide a more intelligent, more efficient and personalized automatic answering function during a reception process of the customer service robot, a humanlike consultation experience is provided for consumers by means of automatic replies, and by means of agent assistance and knowledge base linkage, the reception efficiency is improved for a seller.

Description

Generative question answering system, method, device, storage medium and program product

[0001] Cross-references

[0002] This application refers to Chinese Patent Application No. 2024100816742, filed on January 19, 2024, entitled “Generative Question Answering System, Method, Device, Storage Medium and Program Product”, which is incorporated into this application in its entirety by reference. Technical Field

[0003] The present application relates to the field of artificial intelligence technology, and in particular to a generative question-answering system, method, device, storage medium, and program product. Background Art

[0004] In the e-commerce sector, customer service robots can improve customer service efficiency and provide high-quality, intelligent customer service around the clock. Currently, the most common customer service robots on the market are retrieval-based question-and-answer robots. These rely on manually configured knowledge bases. When consumers ask questions of the robot, the robot retrieves the answers from the knowledge base and provides feedback. Existing customer service robots are prone to irrelevant answers, making it difficult to provide accurate responses, which negatively impacts the consumer experience.

[0005] Summary of the Invention

[0006] Various aspects of the present application provide a generative question-answering system, method, device, storage medium, and program product for providing personalized responses to users in a more intelligent, efficient, and accurate manner to meet their personalized needs.

[0007] The embodiment of the present application provides a generative question-answering system, comprising: a question-answering server and a question-answering assistant;

[0008] The question-and-answer server is configured to receive a question description, understand the intent of the question description information using a question-and-answer model, and obtain first question information; perform a search in a knowledge base based on the first question information to obtain first initial answer information; send the first initial answer information to the question-and-answer assistance terminal for correction, and output the first target answer information returned by the question-and-answer assistance terminal; the question-and-answer assistance terminal is configured to receive the first initial answer information, respond to a correction operation, correct the first initial answer information to obtain first target answer information, and return the information to the question-and-answer server.

[0009] An embodiment of the present application also provides a generative question-answering method, including: receiving first initial answer information sent by a question-answering server, where the first initial answer information is retrieved from a knowledge base by the question-answering server using a question-answering model to understand the intent of question description information; in response to a correction operation, correcting the first initial answer information to obtain first target answer information; and returning the first target answer information to the question-answering server so that the question-answering server can output the first target answer information.

[0010] An embodiment of the present application also provides an electronic device, including: a memory and a processor; the memory is used to store a computer program; the processor is coupled to the memory, and is used to execute the computer program to perform the steps in the generative question-answering method.

[0011] An embodiment of the present application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor is enabled to implement the steps in the generative question-answering method.

[0012] An embodiment of the present application also provides a computer program product, including a computer program / instruction. When the computer program / instruction is executed by a processor, the processor is enabled to implement the steps in the generative question answering method.

[0013] In an embodiment of the present application, the Q&A server in the generative Q&A system uses a joint Q&A model and knowledge base to respond to user questions, and with the assistance of the Q&A assistant in the generative Q&A system, it modifies the answer information to the question information, thereby providing more intelligent, efficient, and accurate personalized responses to users to meet their personalized needs. In particular, when the Q&A server acts as a customer service robot, the joint Q&A model and knowledge base provide a more intelligent, efficient, and personalized automatic response function during the customer service robot's reception process, providing consumers with a personalized consultation experience through automatic replies, and improving the efficiency of merchant reception through the linkage of agent assistance and knowledge base. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0015] FIG1 is a schematic diagram of the structure of a generative question-answering system provided in an embodiment of the present application;

[0016] FIG2 is an interaction diagram of a generative question-answering method provided in an embodiment of the present application;

[0017] FIG3 is an interaction diagram of another generative question-answering method provided in an embodiment of the present application;

[0018] FIG4 is a diagram of an exemplary application scenario provided by an embodiment of the present application;

[0019] FIG5 is a flow chart of a generative question-answering method provided in an embodiment of the present application;

[0020] FIG6 is a flowchart of another generative question-answering method provided in an embodiment of the present application;

[0021] FIG7 is a schematic diagram of the structure of a generative question-answering device provided in an embodiment of the present application;

[0022] FIG8 is a schematic diagram of the structure of another generative question-answering device provided in an embodiment of the present application;

[0023] FIG9 is a schematic structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0024] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0025] In the embodiments of the present application, "at least one" refers to one or more, and "more" refers to two or more. "And / or" describes the access relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent three situations: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. In the text description of the present application, the character " / " generally indicates that the previous and next associated objects are in an "or" relationship. In addition, in the embodiments of the present application, "first", "second", "third", etc. are only used to distinguish the contents of different objects and have no other special meanings.

[0026] In the e-commerce sector, customer service robots can improve customer service efficiency and provide high-quality, intelligent customer service around the clock. Currently, the most common customer service robots on the market are retrieval-based question-and-answer robots. These rely on manually configured knowledge bases. When consumers ask questions of the robot, the robot retrieves the answers from the knowledge base and provides feedback. Existing customer service robots are prone to irrelevant answers, making it difficult to provide accurate responses, which negatively impacts the consumer experience.

[0027] To this end, the embodiments of the present application provide a generative question-answering system, method, device, storage medium, and program product. In the embodiments of the present application, the question-answering server in the generative question-answering system jointly uses the question-answering model and the knowledge base to respond to the question information raised by the user, and with the assistance of the question-answering auxiliary terminal in the generative question-answering system, the answer information of the response question information is corrected, thereby providing more intelligent, efficient, and accurate personalized responses to the user to meet the user's personalized needs. In particular, when the question-answering server acts as a customer service robot, the combined question-answering model and the knowledge base provide a more intelligent, efficient, and personalized automatic response function during the customer service robot's reception process, providing consumers with a personified consultation experience through automatic replies, and improving the efficiency of merchant reception through the linkage of agent assistance and the knowledge base.

[0028] The following describes in detail the technical solutions provided by various embodiments of the present application in conjunction with the accompanying drawings.

[0029] FIG1 is a schematic diagram of the structure of a generative question-answering system provided in an embodiment of the present application. Referring to FIG1 , the generative question-answering system includes a question-answering server 10 and a question-answering assistant 20 .

[0030] In this embodiment, the Q&A server 10 can be any server-side device with Q&A interactive functionality, such as a cloud server, a conventional server, or a server cluster, without limitation. Supported by speech recognition and natural language processing technologies, the Q&A server 10 possesses multiple capabilities, including emotion recognition, personalized responses, and autonomous learning, fully meeting diverse Q&A interactive needs.

[0031] In this embodiment, the question-answering assistant terminal 20 can be either hardware or software. When the question-answering assistant terminal 20 is hardware, it can be, for example, a mobile phone, tablet computer, wearable device, or vehicle-mounted device. When the question-answering assistant terminal 20 is software, it can be installed in the hardware devices listed above. In this case, the question-answering assistant terminal 20 can be, for example, multiple software modules or a single software module, etc., and this embodiment of the application is not limited thereto.

[0032] In this embodiment, the question-answering server 10 combines the question-answering model and the knowledge base to respond to the question information raised by the user, and with the assistance of the question-answering auxiliary terminal 20, corrects the answer information of the response question information, thereby providing more intelligent, efficient and accurate personalized responses to the user to meet the user's personalized needs.

[0033] The generative question-answering system can replace manual work to solve most consulting problems, and can be applied in various scenarios such as intelligent online customer service, search engines, and voice dialogue interaction, without any restrictions. Taking the intelligent online customer service scenario in the e-commerce field as an example, merchants can provide consumers with 24-hour customer service. Consumers enter the question description content that describes the first question information in the user terminal 30, as shown in ① in Figure 1. The user terminal 30 sends the question description content to the question-answering server 10 that plays the role of the customer service robot. The question-answering server 10 uses the question-answering model and the knowledge base to perform response processing. Specifically, as shown in ② in Figure 1, the question-answering server 10 uses the question-answering model to understand the first question information described by the question description content. As shown in ③, ④ and ⑤ in Figure 1, the question-answering model retrieves the first initial answer information corresponding to the first question information in the knowledge base that stores each question information and its answer information, and returns it to the customer service robot. As shown in ⑥ in Figure 1, the customer service robot sends the first initial answer information to the question and answer assistance terminal 20. As shown in ⑦ and ⑧ in Figure 1, the question and answer assistance terminal 20 responds to the correction operation triggered by the question and answer assistant who plays the role of manual customer service, corrects the first initial answer information to obtain the first target answer information, and returns the first target answer information to the customer robot. The customer robot outputs the first target answer information to the user terminal 30, thereby completing the question and answer interaction between the customer service robot and the consumer.

[0034] It is explained here that in the scenario embodiment shown in Figure 2, the question and answer service end 10 is implemented as a customer service robot, and the question and answer auxiliary end 20 is implemented as a manual customer service end as an example, but it is not limited to this.

[0035] The working principle of the generative question answering system provided by the embodiment of the present application is described below in conjunction with Figure 2. Figure 2 is an interaction diagram of a generative question answering method provided by the embodiment of the present application. Referring to Figure 2, the method may include the following steps:

[0036] 201. The question-answering server receives the question description.

[0037] In this embodiment, when a user has a consultation need, the question description content is provided to the question-answering server, and the question description content is used to describe the first question information. For example, the user manually inputs or voice-inputs the question description content in natural language in various user terminals such as mobile phones or tablets, and the user terminal sends the question description content to the question-answering server. The question description content is the content that describes the question information. For ease of understanding, the question information described by the question description content is referred to as the first question information. Taking the user as a consumer as an example, the first question information may be, for example, "Why hasn't the order been shipped for so long?", "What kind of fabric is the clothes?", "Are there any discounts?", etc.

[0038] 202. The question-answering server uses the question-answering model to understand the intent of the question description information to obtain first question information; and searches the knowledge base based on the first question information to obtain first initial answer information.

[0039] In this embodiment, the Question Answering (QA) model has a question answering and response processing function, such as but not limited to: DBQA (Document-Based Question Answering) model, artificial intelligence generated content (AIGC) model or industry large model.

[0040] Among them, the industry large model refers to a large language model (LLM) applied to a specific industry field. The industry large model is pre-trained or instruction fine-tuned (IFT) on large-scale industry field data, and its effect in a specific industry field is better than that of a general large model. Therefore, the industry large model is usually deployed to serve various specific tasks downstream in the industry field. Preferably, the industry large model is fine-tuned with data of the scenario served by the generative question-answering system. For example, the industry large model is fine-tuned with the question and answer data of the customer service scenario in the e-commerce field to obtain a generative question and answer system for intelligent online customer service of e-commerce. For another example, the industry large model is fine-tuned with the question and answer data of the search engine in the search field to obtain a generative question and answer system for search engines. For another example, the industry large model is fine-tuned with the question and answer data of interacting with smart home devices in the smart home field to obtain a generative question and answer system for remote control of smart home devices.

[0041] In this embodiment, the knowledge base is a database that stores knowledge information and provides support for question answering and information retrieval.

[0042] In this embodiment, the joint question-answering model and the knowledge base perform question-answering interaction, which has the following advantages:

[0043] (1): The question-answering model can better understand the semantics of natural language processing tasks. The knowledge base stores rich structured or semi-structured knowledge. The combined question-answering model and the knowledge base can achieve more accurate semantic understanding and knowledge acquisition.

[0044] (2): The question-answering model can perform intelligent question-answering, but it may sometimes be limited by knowledge in some fields. The knowledge base provides rich domain knowledge. The combined question-answering model and the knowledge base can perform intelligent question-answering more accurately.

[0045] (3): The question-answering model has the ability to process multimodal data such as video, pictures, voice or text. The knowledge base stores the knowledge information of these multimodal data. The combined question-answering model and the knowledge base can perform multimodal analysis more comprehensively and achieve more accurate knowledge acquisition.

[0046] (4): The question-answering model has advantages in personalized responses. The knowledge information in the knowledge base can assist the question-answering model in providing more accurate personalized responses.

[0047] In this embodiment, when the question-answering server jointly conducts question-answering interaction with the question-answering model and the knowledge base, the question description content can be input into the question-answering model so that the question-answering model can understand the intention of the question description content to obtain the first question information; based on the first question information, a search is performed in the knowledge base to obtain the first initial answer information, and the first initial answer information is also the answer information of the first question information retrieved from the knowledge base.

[0048] Further optionally, in order to provide more accurate personalized responses, the question and answer server uses a question and answer model to understand the intent of the question description information. When the first question information is obtained, the question description content can be input into the question and answer model, and the question description content can be understood based on historical question information and its historical answer information to obtain the first question information.

[0049] Specifically, historical question information refers to questions answered by the generative question-answering system in the past, while historical answer information refers to the answers provided to these questions. Using this historical information, the question-answering model can more accurately understand the intent of the currently received question description, providing more precise and personalized responses.

[0050] In this embodiment, the knowledge base can store various question information and answer information thereof. After the question-answering model outputs the first question information, the first initial answer information of the first question information can be retrieved from the knowledge base.

[0051] In practical applications, there can be one knowledge base or multiple knowledge bases, and there is no limitation to this. Further, optionally, in order to provide more accurate personalized responses, the knowledge base includes a general knowledge base and at least one dedicated knowledge base, and different dedicated knowledge bases correspond to different service objects.

[0052] In this embodiment, the general knowledge base can be understood as a knowledge base with universal applicability, and the dedicated knowledge base can be understood as a knowledge base dedicated to service objects, such as merchants, stores, search engines, smart home devices, etc.

[0053] Based on the above, the implementation method of searching in the knowledge base according to the first question information to obtain the first initial answer information is: searching in the target-specific knowledge base according to the first question information, the target-specific knowledge base is the dedicated knowledge base of the service object corresponding to the first question information; if the first initial answer information cannot be retrieved in the target-specific knowledge base, searching in the general knowledge base according to the first question information to obtain the first initial answer information.

[0054] It is worth noting that compared with the answer information retrieved from the general knowledge base, the answer information retrieved from the dedicated knowledge base is more accurate. The general knowledge base can provide a backup for the retrieved answer information and improve the reliability of the question-answering interaction.

[0055] 203. The question-answering server sends the first initial answer information to the question-answering auxiliary terminal.

[0056] In actual applications, the question-answering server may wait until all the first initial answer information is generated, and then send the first initial answer information to the question-answering auxiliary terminal.

[0057] Optionally, to improve Q&A efficiency, the Q&A server can use streaming to transmit data to the Q&A server. Streaming is a data transmission method in which data is transmitted continuously in a stream, rather than all at once. In streaming, data is divided into smaller blocks or frames and transmitted sequentially. Streaming can meet real-time requirements and reduce transmission delays and network congestion.

[0058] Based on the above, in the process of generating the first initial answer information, the question and answer server sends the first initial answer information to the question and answer assistance terminal in a streaming manner, so that the question and answer assistance terminal can modify the first initial answer information to obtain the first target answer information if it successfully receives the first initial answer information.

[0059] Further optionally, in the process of the question and answer server sending the first initial answer information to the question and answer assistance terminal in a streaming manner, the question and answer assistance terminal may also send a cancellation instruction to the question and answer server, and the question and answer assistance terminal terminates the generation and sending operation of the first initial answer information in response to the cancellation instruction.

[0060] Specifically, after the Q&A assistant on the Q&A assistant side views the streamed first initial answer information, he or she decides on demand whether to reply to the user based on the first initial answer information to reduce the probability of incorrectly replying to the user. If the Q&A assistant decides to give up replying to the user based on the first initial answer information, the Q&A assistant inputs a cancel operation on the Q&A assistant side, and the Q&A assistant side responds to the cancel operation and sends a cancel instruction to the Q&A server. The Q&A server terminates the generation and sending operations of the first initial answer information in response to the cancel instruction, that is, it no longer continues to generate the remaining part of the first initial answer information, nor does it send the first initial answer information to the Q&A assistant side.

[0061] 204. The question-answering auxiliary terminal receives first initial answer information corresponding to the first question information sent by the question-answering server terminal.

[0062] 205. The question-answering auxiliary terminal responds to the correction operation and corrects the first initial answer information to obtain the first target answer information.

[0063] Specifically, after the question-and-answer service end sends the complete first initial answer information to the question-and-answer assistance end, the question-and-answer assistance personnel on the question-and-answer assistance end correct the first initial answer information to reduce the probability of incorrectly replying to the user.

[0064] Optionally, to provide a faster and more accurate response to the user, the question-answering assistant terminal may perform step 205 by displaying the first initial answer information on the correction interface; if the first initial answer information does not meet the preset speech conditions, responding to the correction operation, correcting the first initial answer information to obtain the first target answer information. If the first initial answer information meets the preset speech conditions, the first initial answer information is directly used as the first target answer information.

[0065] Specifically, the preset speech condition is flexibly set as needed. For example, the preset speech condition is that sensitive words appear in the first initial answer information.

[0066] For another example, the preset speech condition is that the fluency of the first initial answer information is less than the preset fluency. Among them, a fluency evaluation model can be trained, and the fluency evaluation model is used to evaluate the fluency of the first initial answer information, and the preset fluency is flexibly set as needed. Fluency reflects the degree of sentence smoothness of the first initial answer information. When training the fluency evaluation model, a large number of Chinese sentences and their annotated fluency are prepared, and the Chinese sentences are input into the fluency evaluation model to obtain the predicted fluency output by the fluency evaluation model. The model parameters of the fluency evaluation model are adjusted according to the loss value between the annotated fluency and the predicted fluency, and the model training is repeated iteratively until the model training end condition is met.

[0067] For another example, the preset speech condition is that the friendliness of the first initial answer information is less than the preset friendliness. A friendliness evaluation model can be trained to evaluate the friendliness of the first initial answer information using the friendliness evaluation model, and the preset friendliness can be flexibly set as needed. The friendliness reflects whether the first initial answer information is easy for the user to accept and listen to. When training the friendliness evaluation model, a large number of Chinese sentences and their annotated friendliness are prepared, the Chinese sentences are input into the friendliness evaluation model, and the predicted friendliness output by the friendliness evaluation model is obtained. The model parameters of the friendliness evaluation model are adjusted according to the loss value between the annotated friendliness and the predicted friendliness, and the model training is repeated iteratively until the model training end condition is met.

[0068] For another example, the preset condition is that at least one of the following conditions is met: sensitive words appear in the first initial answer information, the fluency of the first initial answer information is less than the preset fluency, and the preset speech condition is that the friendliness of the first initial answer information is less than the preset friendliness.

[0069] 206. The question-answering auxiliary terminal returns the first target answer information to the question-answering server terminal.

[0070] 207. The question-answering server outputs the first target answer information.

[0071] Specifically, after receiving the first target answer information, the question-answering server outputs the first target answer information to the user who asked the first question information, so as to complete the question-answering interaction with the user.

[0072] In practical applications, a time period can be reserved for Q&A assistants, during which the Q&A assistant can request the Q&A server to withdraw the answer information already output to the user, thereby reducing the probability of incorrectly replying to the user. Based on this, further optionally, within a first preset time after outputting the first target answer information, the Q&A server responds to the withdrawal instruction sent by the Q&A assistant and withdraws the output first target answer information. The first preset time period can be flexibly set as needed, for example, 2 minutes.

[0073] If the Q&A assistant decides to withdraw the first target answer information that has been replied to the user, the Q&A assistant inputs the withdrawal operation on the Q&A assistant end, the Q&A assistant end responds to the withdrawal operation, and sends a withdrawal instruction to the Q&A server end. The Q&A server end responds to the withdrawal instruction sent by the Q&A assistant end and withdraws the first target answer information that has been output.

[0074] In some optional embodiments, the question-answering server further updates the first question information and the first target answer information as knowledge information into the knowledge base to expand the knowledge base.

[0075] In actual applications, a period of time can also be reserved for the question-and-answer assistant. During this period of time, the question-and-answer assistant can request the question-and-answer server to update the first target answer information of the first question information that has been saved in the knowledge base, so as to reduce the probability of incorrectly replying to the user. Based on this, it is further optional that after a second preset time after outputting the first target answer information, the question-and-answer server responds to the modification instruction sent by the question-and-answer assistant including the modified first target answer information, and replaces the first target answer information before modification stored in the knowledge base with the modified first target answer information. The second preset time is flexibly set as needed, for example, 2 minutes.

[0076] In actual applications, the question-and-answer assistant inputs modification operations on the question-and-answer assistant end, the question-and-answer assistant end responds to the modification operations and sends modification instructions to the question-and-answer server end. The question-and-answer server end responds to the modification instructions sent by the question-and-answer assistant end and replaces the first target answer information before the modification stored in the knowledge base with the modified first target answer information.

[0077] In some optional embodiments, after the question-answering server outputs the first target answer information to the user, it can also optimize the question-answering model based on the user's feedback information on the first target answer information to improve the accurate answer performance of the question-answering model.

[0078] In this embodiment, the feedback information is used to feedback the user's satisfaction with the first target answer information. In practical applications, any existing method can be used to optimize the question-answering model based on the feedback information, and there is no limitation on this.

[0079] Further optionally, in order to better optimize the question-answering model, a reward model (RM) can be trained based on the user's feedback on the first target answer information, and the reward model can be used to fine-tune the question-answering model using reinforcement learning. Exemplarily, when fine-tuning the question-answering model, the action space of the question-answering model to be optimized is the prediction vocabulary, the state is the currently generated content, and the feedback information of the reward model is transmitted to the question-answering model for optimization through the PPO (Proximal Policy Optimization) algorithm. The feedback information of the reward model reflects the characteristics of e-commerce customer service question-answering tasks. The question-answering model will adjust parameters and structure based on the characteristics of e-commerce customer service question-answering tasks provided by the reward model, improving the output performance of the model while ensuring performance.

[0080] The technical solution provided by the embodiment of the present application is that the question-answering server in the generative question-answering system combines the question-answering model and the knowledge base to respond to the question information raised by the user, and with the assistance of the question-answering auxiliary terminal in the generative question-answering system, the answer information of the response question information is corrected, thereby providing more intelligent, efficient, and accurate personalized responses to users to meet their personalized needs. In particular, when the question-answering server acts as a customer service robot, the combined question-answering model and the knowledge base provide a more intelligent, efficient, and personalized automatic response function during the customer service robot's reception process, providing consumers with a humanized consultation experience through automatic replies, and improving the efficiency of merchant reception through the linkage of agent assistance and knowledge base.

[0081] FIG3 is an interaction diagram of another generative question-answering method provided in an embodiment of the present application. Referring to FIG3 , the method may include the following steps:

[0082] 301. The question-answering server receives the question description.

[0083] The implementation of step 301 can refer to the implementation of step 201 in the above embodiment, and will not be repeated here.

[0084] 302. The Q&A server determines whether the question description satisfies the diversion conditions based on at least one of the service target information, user information, and a preset diversion ratio corresponding to the question description. If yes, proceed to step 303; otherwise, proceed to step 309.

[0085] In this embodiment, during the question-and-answer interaction, the question-and-answer server can respond to user questions using a combination of the question-and-answer model and the knowledge base, or it can solely utilize the question-and-answer model. Specifically, if the question description satisfies the diversion criteria, the question description is responded to by combining the question-and-answer model and the knowledge base, i.e., the joint generation of answer information using the question-and-answer model and the knowledge base is determined. If the question description does not meet the diversion criteria, the question description is responded to by solely utilizing the question-and-answer model.

[0086] In this embodiment, the question and answer server can determine whether the question description content meets the diversion conditions based on at least one of the service object information, user information and a preset diversion ratio corresponding to the question description content. The diversion conditions are mainly used to divert the received question description content. Part of the question description content is diverted to the link where the question and answer model and the knowledge base are combined for processing, and part of the question description content is diverted to the link that the question and answer model is solely responsible for for processing. Among them, using two links and diverting the received question description content to different links for processing based on the diversion conditions can reduce the processing pressure on a single link. Of course, it is not limited to two links. For example, it can also include a manual processing link and a link that the knowledge base is solely responsible for. In the embodiment of the present application, the specific implementation of the diversion conditions is not limited and can be flexibly determined according to application requirements.

[0087] Among them, the service object information is mainly related information of the service object. The service object refers to the object to which the user initiates the question description content, such as a merchant, a store or an e-commerce platform. Taking the service object as an example, the service object information is, for example, the merchant type and merchant level. Merchant types can be divided from different angles, and merchant types can be, for example: small merchants, medium-sized merchants and large merchants, etc. Merchant types can also be, for example: fresh food merchants, daily necessities merchants, home appliance merchants, etc. Merchant levels can be, for example, 1, 2, 3, 4, 5 and other different levels, and can also be, for example, A, B, C and other different levels.

[0088] Exemplarily, one diversion condition is to divert question description content involving preset merchant types to the link where the question-answering model and the knowledge base are combined, and other question description content is diverted to the link where the question-answering model is solely responsible; the preset merchant types are, for example, individual business owners or household appliance merchants. Another diversion condition is to divert question description content involving preset merchant levels to the link where the question-answering model and the knowledge base are combined, and other question description content is diverted to the link where the question-answering model is solely responsible; the preset merchant levels are, for example, 5 stars, 4 diamonds, and 5 diamonds. Yet another diversion condition is to divert question description content involving preset merchant types and merchant levels greater than a set level threshold to the link where the question-answering model and the knowledge base are combined, and other question description content is diverted to the link where the question-answering model is solely responsible; for example, question description content involving large merchants and merchant levels above 3 stars is diverted to the link where the question-answering model and the knowledge base are combined, and other question description content is diverted to the link where the question-answering model is solely responsible.

[0089] The information of the user who enters the question description (hereinafter referred to as user information) includes, but is not limited to, the user's location information, payment information, or user level. Payment information can reflect whether the user has made large purchases, and the user level can be, for example, an ordinary user or a VIP (very important person) user.

[0090] For example, one diversion condition is to divert question description content submitted by users in a specific geographical area to the link where the question-answering model and the knowledge base are combined, while other question description content is diverted to the link where the question-answering model is solely responsible. Another diversion condition is that if the user has made large purchases, then the question description content submitted by users who have made large purchases can be diverted to the link where the question-answering model and the knowledge base are combined, while other question description content is diverted to the link where the question-answering model is solely responsible. Yet another diversion condition is that if the user is a VIP user, then the question description content submitted by VIP users can be diverted to the link where the question-answering model and the knowledge base are combined, while other question description content is diverted to the link where the question-answering model is solely responsible.

[0091] Among them, the preset diversion ratio refers to the proportion of question description content received within the set time period that is diverted to the link where the question-answering model and the knowledge base are combined. For example, the preset diversion ratio is 70%, which means that 70% of the question description content received within the set time period is diverted to the link where the question-answering model and the knowledge base are combined, and the other 30% of the question description content is diverted to the link that the question-answering model is solely responsible for.

[0092] It should be noted that the aforementioned diversion conditions may be used individually or in any combination, without limitation. For example, combining diversion conditions based on user information with a preset diversion ratio allows question descriptions submitted by users within a specific geographic area to be diverted to the link combining the question-answering model and the knowledge base. Furthermore, the question descriptions diverted to this link within a set time period cannot exceed 70% of all question descriptions received within the set time period.

[0093] 303. The question-answering server uses the question-answering model to understand the intent of the question description information to obtain first question information; and searches the knowledge base based on the first question information to obtain first initial answer information.

[0094] For the question description content that is diverted to the link between the question-answering model and the knowledge base, the question-answering model and the knowledge base can be used to jointly generate the first initial answer information corresponding to the first question information based on the question description content. The implementation method of step 303 can be referred to the implementation method of step 202 in the above embodiment and will not be repeated here.

[0095] 304. The question-answering server sends the first initial answer information to the question-answering auxiliary terminal.

[0096] The implementation of step 304 can refer to the implementation of step 203 in the above embodiment, and will not be repeated here.

[0097] 305. The question-answering auxiliary terminal receives first initial answer information corresponding to the first question information sent by the question-answering server.

[0098] The implementation of step 305 can refer to the implementation of step 204 in the above embodiment, and will not be repeated here.

[0099] 306. The question-answering auxiliary terminal responds to the correction operation and corrects the first initial answer information to obtain the first target answer information.

[0100] The implementation of step 306 can refer to the implementation of step 205 in the above embodiment, and will not be repeated here.

[0101] 307. The question-answering auxiliary terminal returns the first target answer information to the question-answering server terminal.

[0102] The implementation of step 307 can refer to the implementation of step 206 in the above embodiment, which will not be repeated here.

[0103] 308. The question-answering server outputs the first target answer information.

[0104] The implementation of step 308 can refer to the implementation of step 207 in the above embodiment, and will not be repeated here.

[0105] 309. The question-answering server generates second initial answer information corresponding to the first question information based on the question description content using the question-answering model.

[0106] Specifically, the question-answering model has a question-answering function, which can answer the first question information and obtain the second initial answer information corresponding to the first question information.

[0107] 308. The question-answering server sends the second initial answer information to the question-answering auxiliary terminal, so that the question-answering auxiliary terminal can modify the second initial answer information to obtain the second target answer information.

[0108] Among them, the implementation method of the question and answer server sending the second initial answer information to the question and answer auxiliary terminal is similar to the implementation method of the question and answer server sending the first initial answer information to the question and answer auxiliary terminal, which will not be repeated here.

[0109] The implementation method of the question-and-answer assistance terminal to correct the second initial answer information is similar to the implementation method of the question-and-answer assistance terminal to correct the first initial answer information, and will not be repeated here.

[0110] 309. Receive the second target answer information returned by the question-answering auxiliary terminal, and output the second target answer information.

[0111] The technical solution provided by the embodiment of the present application is that the question-answering server in the generative question-answering system uses diversion conditions so that during the question-answering interaction process, the question-answering server can jointly use the question-answering model and the knowledge base to answer the question information raised by the user, or can only use the question-answering model to answer the question information raised by the user, and with the assistance of the question-answering auxiliary terminal in the generative question-answering system, the answer information of the reply question information is corrected, thereby providing more intelligent, efficient and accurate personalized replies to users to meet the personalized needs of users.

[0112] In some optional embodiments, it also supports expanding the knowledge base and enriching the knowledge information of the knowledge base, thereby supporting more intelligent, efficient, and accurate personalized responses to users. Specifically, the question-answering service determines the second question information, which includes question information with a frequency greater than a set frequency threshold and / or customized question information; uses the question-answering model to generate the third initial answer information corresponding to the second question information, that is, the answer information obtained by the question-answering model after answering the second question information is used as the third initial answer information; sends the third initial answer information to the question-answering auxiliary terminal, so that the question-answering auxiliary terminal can modify the third initial answer information to obtain the third target answer information; receives the third target answer information returned by the question-answering auxiliary terminal, and updates the second question information and the third target answer information as knowledge information to the knowledge base.

[0113] In order to better understand the technical solution provided by the embodiment of the present application, a scenario embodiment is introduced below in conjunction with Figure 4.

[0114] Figure 4 shows an example application scenario. Taking the intelligent customer service in the e-commerce field as an example, the customer service robot 50 includes a customer service messaging system, a customer service delivery system, a large model, and a knowledge base. When shopping online, consumers can ask questions to merchants.

[0115] As shown in ① in Figure 4, the consumer's terminal device 70 (such as a mobile phone) sends a consultation question to the customer service robot 50. For example, "What kind of fabric is this?" After receiving the consultation question, the customer service message system decides the customer service mode through customer service diversion. The customer service mode includes manual customer service reception mode, robot fully automatic reception mode, and robot assisted reception mode. In this way, it supports providing multiple reception modes without the consumer's perception. The legitimacy of the consultation question is verified through message authentication. If the consultation question is legitimate, the consultation question is distributed based on the diversion result. If the diversion result is the manual customer service reception mode, the customer service robot 50 will output the consultation question to the merchant's manual customer service terminal device (such as a mobile phone), and return the reply of the manual customer service to the consumer. If the diversion result is fully automated reception mode or robot-assisted reception mode, as shown in Figure 4 (2), the customer service messaging system synchronizes the inquiry question to the large model. As shown in Figure 4 (3), the large model, in conjunction with the knowledge base, responds to the inquiry question, dynamically generates an answer, and returns it to the customer service messaging system. The customer service messaging system monitors the answer generation and periodically verifies the answer. As shown in Figure 4 (4), the customer service messaging system pushes the answer to the customer service delivery system. The customer service messaging system also caches the answer generated by the large model. As shown in Figure 4 (5), during the answer generation process, the customer service delivery system streams the generated answer to the merchant's human customer service terminal device based on an end-to-end communication protocol, where the human customer service terminal displays a visual representation of the generated answer word by word. As shown in Figure 4 (6), after the answer is generated, the customer service delivery system automatically sends the cached complete answer to the inquiry question to the consumer.

[0116] For the human customer service representative, after the answer to the inquiry question begins to be generated, a specific interface pops up on the display interface of the human customer service terminal device 60. The specific interface can be displayed above the input box, and the answer content is dynamically updated. The specific interface displays the answer information. The answer information may be, for example, "AI (Artificial Intelligence) generating: The fabric of this dress is..." The specific interface may also include a cancel button, allowing the human customer service representative to click the cancel button to stop the customer service robot 50 from generating the answer and cancel the display of the specific interface.

[0117] On the human customer service side, after the answer is generated, the customer service robot 50 can automatically reply to the consumer. Of course, the human customer service can also modify the answer automatically generated by the customer service robot 50 and trigger the customer service robot 50 to send the modified answer to the consumer. In addition, within two minutes of the answer being sent to the consumer, the human customer service can withdraw the answer that has been sent to the consumer. After two minutes, the human customer service can correct the answer to the consultation question that has been saved in the knowledge base, which is called knowledge base correction.

[0118] On the consumer side, consumers receive the answer information returned by the big model and are supported to provide positive or negative feedback on the returned answer information. The customer service robot 50 synchronizes the feedback results with the big model to optimize the big model based on the feedback results.

[0119] Regarding the knowledge base: the knowledge base is updated dynamically, and the reply scripts in the knowledge base are not fixed, which reduces the cost of knowledge base operation and maintenance and reduces the quality of answers caused by differences in the level of manual customer service. During customer service reception, manual customer service can trigger the customer service robot 50 to automatically generate reply scripts for high-frequency questions in the customer service link using a large model. After the manual customer service can modify the reply scripts for high-frequency questions, the high-frequency questions and their final reply scripts can be saved to the knowledge base to add new knowledge information to the knowledge base and expand the knowledge base. During customer service reception, manual customer service can customize questions. Manual customer service can trigger the customer service robot 50 to automatically generate reply scripts for customized questions in the customer service link using a large model. After the manual customer service can modify the reply scripts for customized questions, the customized questions and their final reply scripts can be saved to the knowledge base to add new knowledge information to the knowledge base and expand the knowledge base.

[0120] In this application scenario, a closed loop is formed from the generation of answers from the big model, real-time push notifications to consumers or manual customer service representatives, and positive and negative feedback from consumers. Based on an end-to-end communication protocol, data streaming is provided for the customer service reception domain. Furthermore, the big model leverages the fully automatic / assisted reception process of the customer service robot 50 to provide a more intelligent, efficient, and personalized automatic response function. Automatic replies provide consumers with a personalized consultation experience, while agent assistance and knowledge base linkage improve merchant reception efficiency. During the customer service reception process, the big model generates recommended responses based on consumer questions and pushes them to manual customer service representatives in real time, enabling agent assistance and helping improve customer service reception efficiency.

[0121] FIG5 is a flow chart of a generative question-answering method provided in an embodiment of the present application. Referring to FIG5 , the method may include the following steps:

[0122] 501. Receive the problem description.

[0123] 502. Use the question-answering model to understand the intention of the question description information and obtain the first question information.

[0124] 503. Searching the knowledge base according to the first question information to obtain first initial answer information;

[0125] 504. Modify the first initial answer information to obtain first target answer information, and output the first target answer information.

[0126] Further optionally, the question description information is understood in intent using a question-answering model to obtain first question information, including: inputting the question description content into the question-answering model, and based on historical question information and its historical answer information, understanding the intent of the question description content to obtain the first question information.

[0127] Further optionally, the knowledge base includes a general knowledge base and at least one special knowledge base, and different special knowledge bases correspond to different service objects; searching in the knowledge base according to the first question information to obtain first initial answer information includes: searching in a target special knowledge base according to the first question information, and the target special knowledge base is a special knowledge base of the service object corresponding to the first question information; if the first initial answer information cannot be retrieved in the target special knowledge base, searching in the general knowledge base according to the first question information to obtain the first initial answer information.

[0128] Further optionally, before using the question-answering model to understand the intent of the question description information to obtain the first question information, it also includes: judging whether the question description content meets the diversion conditions based on at least one of the service object information, user information and a preset diversion ratio corresponding to the question description content; if the question description content meets the diversion conditions, determining a mode for jointly generating answer information using the question-answering model and the knowledge base.

[0129] Further optionally, when the question description content does not meet the diversion conditions, the question-answering model is used to generate second initial answer information corresponding to the first question information based on the question description content; the second initial answer information is corrected to obtain second target answer information, and the second target answer information is output.

[0130] Further optionally, the above method also includes: updating the first question information and the first target answer information as knowledge information into the knowledge base; and / or training a reward model based on user feedback information on the first target answer information, and using the reward model to perform reinforcement learning-style fine-tuning on the question-answering model.

[0131] Further optionally, correcting the first initial answer information to obtain first target answer information includes: in the process of generating the first initial answer information, sending the first initial answer information to the question-and-answer assistance terminal in a streaming manner, so that the question-and-answer assistance terminal can correct the first initial answer information to obtain the first target answer information if the first initial answer information is successfully received; and

[0132] If a cancellation instruction is received from the question-and-answer assistance terminal during the process of sending the first initial answer information to the question-and-answer assistance terminal in a streaming manner, the generation and sending operations of the first initial answer information are terminated.

[0133] Optionally, the above method further includes:

[0134] within a first preset time after outputting the first target answer information, responding to a withdrawal instruction sent by the question-answering assistance terminal to withdraw the outputted first target answer information;

[0135] and / or

[0136] After a second preset time after outputting the first target answer information, in response to the modification instruction including the modified first target answer information sent by the question-and-answer assistance terminal, the first target answer information before modification stored in the knowledge base is replaced with the modified first target answer information.

[0137] Further optionally, the above method also includes: determining second question information, the second question information including question information with an occurrence frequency greater than a set frequency threshold and / or customized question information; using a question-answering model to generate third initial answer information corresponding to the second question information; correcting the third initial answer information to obtain third target answer information; and updating the second question information and the third target answer information as knowledge information into the knowledge base.

[0138] For the implementation of each step of the embodiment shown in FIG5 , reference can be made to the relevant contents of the aforementioned embodiment, which will not be described in detail here.

[0139] Each step of the method provided in the above embodiment can be performed by the same device, or the method can be performed by different devices. For example, steps 501 to 504 can be performed by device A; for another example, steps 501 and 502 can be performed by device A, while steps 503 and 504 can be performed by device B; and so on.

[0140] The technical solution provided in the embodiment of the present application combines the question-answering model and the knowledge base to respond to the questions raised by the user, and can also correct the answer information of the response question information, thereby providing personalized responses to the user in a more intelligent, efficient and accurate manner to meet the user's personalized needs.

[0141] FIG6 is a flow chart of a generative question-answering method provided in an embodiment of the present application. Referring to FIG6 , the method may include the following steps:

[0142] 601. Receive first initial answer information sent by the question-answering server. The first initial answer information is obtained by searching the knowledge base for first question information obtained by the question-answering server using a question-answering model to understand the intent of question description information.

[0143] 602. In response to the correction operation, correct the first initial answer information to obtain first target answer information.

[0144] 603. Return the first target answer information to the question and answer server, so that the question and answer server can output the first target answer information.

[0145] Optionally, to provide a faster and more accurate response to the user, step 602 may be implemented by displaying the first initial answer information on the correction interface; if the first initial answer information does not meet the preset speech conditions, the first initial answer information is corrected in response to a correction operation to obtain the first target answer information. If the first initial answer information meets the preset speech conditions, the first initial answer information is directly used as the first target answer information.

[0146] For the implementation of each step of the embodiment shown in FIG6 , reference can be made to the relevant contents of the aforementioned embodiment, which will not be described in detail here.

[0147] Each step of the method provided in the above embodiment can be performed by the same device, or the method can be performed by different devices. For example, steps 601 to 603 can be performed by device A; for another example, steps 601 and 602 can be performed by device A, while step 603 can be performed by device B; and so on.

[0148] The technical solution provided in the embodiment of the present application combines the question-answering model and the knowledge base to respond to the questions raised by the user, and can also correct the answer information of the response question information, thereby providing personalized responses to the user in a more intelligent, efficient and accurate manner to meet the user's personalized needs.

[0149] FIG7 is a schematic diagram of the structure of a generative question-answering device provided in an embodiment of the present application. Referring to FIG7 , the device may include:

[0150] The receiving module 71 is used to receive the problem description content.

[0151] The intention understanding module 72 is used to use the question-answering model to understand the intention of the question description information and obtain the first question information.

[0152] A retrieval module 73 is configured to search the knowledge base according to the first question information to obtain first initial answer information;

[0153] The correction module 74 is configured to correct the first initial answer information to obtain first target answer information and output the first target answer information.

[0154] Further optionally, the intention understanding module 72 is specifically configured to: input the question description content into the question-answering model, and perform intention understanding on the question description content based on historical question information and historical answer information to obtain the first question information.

[0155] Further optionally, the knowledge base includes a general knowledge base and at least one special knowledge base, and different special knowledge bases correspond to different service objects; the retrieval module 73 is specifically used to: search in the target special knowledge base according to the first question information, and the target special knowledge base is the special knowledge base of the service object corresponding to the first question information; if the first initial answer information cannot be retrieved in the target special knowledge base, search in the general knowledge base according to the first question information to obtain the first initial answer information.

[0156] Further optionally, the above-mentioned device also includes: a diversion module, which is used to determine whether the question description content meets the diversion conditions based on at least one of the service object information, user information and a preset diversion ratio corresponding to the question description content; when the question description content meets the diversion conditions, determine a mode of jointly generating answer information using a question-answering model and a knowledge base.

[0157] Further optionally, the diversion module is also used to generate second initial answer information corresponding to the first question information based on the question description content using the question-answering model when the question description content does not meet the diversion conditions; correct the second initial answer information to obtain second target answer information, and output the second target answer information.

[0158] Further optionally, the above-mentioned device also includes: an update module and / or a fine-tuning module: an update module, used to update the first question information and the first target answer information as knowledge information into the knowledge base; and / or a fine-tuning module, used to train a reward model based on user feedback information on the first target answer information, and use the reward model to perform reinforcement learning-style fine-tuning on the question-answering model.

[0159] Further optionally, the correction module 74 is specifically configured to: in the process of generating the first initial answer information, send the first initial answer information to the question-answering assistance terminal in a streaming manner, so that the question-answering assistance terminal can correct the first initial answer information to obtain the first target answer information if the first initial answer information is successfully received; and

[0160] If a cancellation instruction is received from the question-and-answer assistance terminal during the process of sending the first initial answer information to the question-and-answer assistance terminal in a streaming manner, the generation and sending operations of the first initial answer information are terminated.

[0161] Further optionally, the above device further includes a withdrawal module for:

[0162] within a first preset time after outputting the first target answer information, responding to a withdrawal instruction sent by the question-answering assistance terminal to withdraw the outputted first target answer information;

[0163] and / or

[0164] The correction module 74 is also used to: after a second preset time after outputting the first target answer information, respond to the modification instruction including the modified first target answer information sent by the question and answer assistance terminal, and replace the first target answer information before modification stored in the knowledge base with the modified first target answer information.

[0165] Further optionally, the correction module 74 is also used to: determine second question information, the second question information including question information with an occurrence frequency greater than a set frequency threshold and / or customized question information; generate third initial answer information corresponding to the second question information using a question-answering model; correct the third initial answer information to obtain third target answer information; and update the second question information and the third target answer information as knowledge information into the knowledge base.

[0166] The implementation principle of the device shown in FIG7 can be found in the relevant description of the aforementioned embodiment and will not be described in detail.

[0167] The technical solution provided in the embodiment of the present application combines the question-answering model and the knowledge base to respond to the questions raised by the user, and can also correct the answer information of the response question information, thereby providing personalized responses to the user in a more intelligent, efficient and accurate manner to meet the user's personalized needs.

[0168] FIG8 is a schematic diagram of the structure of another generative question-answering device provided in an embodiment of the present application. Referring to FIG8 , the device may include:

[0169] A receiving module 81 is configured to receive first initial answer information sent by a question-answering server, where the first initial answer information is retrieved from a knowledge base by the question-answering server using a question-answering model to understand the intent of the question description information.

[0170] a correction module 82, configured to correct the first initial answer information in response to a correction operation to obtain first target answer information;

[0171] The returning module 83 is configured to return the first target answer information to the question-answering server, so that the question-answering server can output the first target answer information.

[0172] Optionally, to provide a faster and more accurate response to the user, the correction module 82 is specifically configured to: display the first initial answer information on the correction interface; if the first initial answer information does not meet the preset speech conditions, respond to the correction operation and correct the first initial answer information to obtain the first target answer information. If the first initial answer information meets the preset speech conditions, the first initial answer information is directly used as the first target answer information.

[0173] For the implementation of each step of the embodiment shown in FIG8 , reference can be made to the relevant contents of the aforementioned embodiment, which will not be described in detail here.

[0174] The technical solution provided in the embodiment of the present application combines the question-answering model and the knowledge base to respond to the questions raised by the user, and can also correct the answer information of the response question information, thereby providing personalized responses to the user in a more intelligent, efficient and accurate manner to meet the user's personalized needs.

[0175] In addition, in some of the processes described in the above embodiments and the accompanying drawings, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or may be executed in parallel. The sequence numbers of the operations, such as 401, 402, etc., are only used to distinguish between different operations, and the sequence numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0176] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0177] FIG9 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. As shown in FIG9 , the electronic device includes: a memory 91 and a processor 92;

[0178] The memory 91 is used to store computer programs and can be configured to store various other data to support operations on the computing platform. Examples of such data include instructions for any application or method operating on the computing platform, contact data, phone book data, messages, pictures, videos, etc.

[0179] The memory 91 can be implemented by any type of volatile or non-volatile memory device or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read only memory (EEPROM), erasable programmable read only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0180] The processor 92 is coupled to the memory 91 and is configured to execute the computer program in the memory 91 to perform the steps in the generative question answering method.

[0181] Further optionally, as shown in Figure 9, the electronic device also includes: a communication component 93, a display 94, a power supply component 95, an audio component 96 and other components. Figure 9 only schematically shows some components, which does not mean that the electronic device only includes the components shown in Figure 9. In addition, the components in the dotted box in Figure 9 are optional components, not mandatory components, and the specific product form of the electronic device may depend on the product form. The electronic device of this embodiment can be implemented as a terminal device such as a desktop computer, a laptop computer, a smart phone or an IOT (Internet of things) device, or it can be a server-side device such as a conventional server, a cloud server or a server array. If the electronic device of this embodiment is implemented as a terminal device such as a desktop computer, a laptop computer, a smart phone, etc., it may include the components in the dotted box in Figure 9; if the electronic device of this embodiment is implemented as a server-side device such as a conventional server, a cloud server or a server array, it may not include the components in the dotted box in Figure 9.

[0182] The detailed implementation process of the processor executing each action can be found in the relevant description in the aforementioned method embodiment or device embodiment, and will not be repeated here.

[0183] Accordingly, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, which, when executed, can implement the steps that can be performed by the electronic device in the above method embodiment.

[0184] Accordingly, an embodiment of the present application also provides a computer program product, including a computer program / instruction. When the computer program / instruction is executed by a processor, the processor is enabled to implement the steps in the above method embodiment that can be performed by an electronic device.

[0185] The above-mentioned communication component is configured to facilitate wired or wireless communication between the device where the communication component is located and other devices. The device where the communication component is located can access a wireless network based on a communication standard, such as WiFi (Wireless Fidelity), 2G (2Generation, 2nd generation), 3G (3Generation, 3rd generation), 4G (4Generation, 4th generation) / LTE (long Term Evolution, long term evolution), 5G (5Generation, 5th generation) and other mobile communication networks, or a combination thereof. In an exemplary embodiment, the communication component receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wide band (UWB) technology, Bluetooth (BT) technology and other technologies.

[0186] The above-mentioned display includes a screen, which may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor can not only sense the boundary of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation.

[0187] The power supply assembly provides power to various components of the device in which the power supply assembly is located. The power supply assembly may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device in which the power supply assembly is located.

[0188] The above-mentioned audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC), and when the device where the audio component is located is in an operating mode, such as call mode, recording mode, and voice recognition mode, the microphone is configured to receive external audio signals. The received audio signal can be further stored in a memory or sent via a communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0189] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0190] The present application is described with reference to the flow chart and / or block diagram of the method, device (system), and computer program product according to the embodiment of the present application. It should be understood that each flow process and / or box in the flow chart and / or block diagram and the combination of the flow process and / or box in the flow chart and / or block diagram can be realized by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processing machine or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for realizing the function specified in one flow chart flow or multiple flows and / or one box or multiple boxes of the block diagram.

[0191] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0192] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0193] In a typical configuration, a computing device includes one or more processors (Central Processing Unit, CPU), input / output interfaces, network interfaces, and memory.

[0194] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0195] Computer-readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, Phase Change RAM (PRAM), Static Random-Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically-Erasable Programmable Read-Only Memory (EEPROM), flash memory or other memory technology, CD-ROM, Digital Versatile Disc (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device. According to the definition in this article, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

[0196] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0197] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A generative question-answering system, characterized in that Including: A question-and-answer server and a question-and-answer assistant; The question-and-answer server is used to receive the question description content, use a question-and-answer model to perform intent understanding on the question description information to obtain first question information; retrieve in a knowledge base according to the first question information to obtain first initial answer information; Send the first initial answer information to the question-and-answer assistant for correction, and output the first target answer information returned by the question-and-answer assistant; The question-and-answer assistant is used to receive the first initial answer information, respond to the correction operation, correct the first initial answer information to obtain first target answer information, and return it to the question-and-answer server.

2. The system according to claim 1, wherein The question-and-answer server is further used for: Updating the first question information and the first target answer information to the knowledge base as knowledge information; and / or For second question information, using a question-and-answer model to generate third initial answer information corresponding to the second question information; Sending the third initial answer information to the question-and-answer assistant for correction to obtain third target answer information, and updating the second question information and the third target answer information to the knowledge base as knowledge information; wherein, the second question information includes question information with an appearance frequency greater than a set frequency threshold and / or custom question information.

3. A generative question-answering method, characterized in that, Including: Receiving the question description content; Using a question-and-answer model to perform intent understanding on the question description information to obtain first question information; Retrieving in a knowledge base according to the first question information to obtain first initial answer information; Correcting the first initial answer information to obtain first target answer information, and outputting the first target answer information.

4. The method according to claim 3, wherein Using a question-and-answer model to perform intent understanding on the question description information to obtain first question information, including: Inputting the question description content into the question-and-answer model, and performing intent understanding on the question description content based on historical question information and its historical answer information to obtain the first question information.

5. The method according to claim 4, wherein The knowledge base includes a general knowledge base and at least one dedicated knowledge base, and different dedicated knowledge bases correspond to different service objects; Retrieving in a knowledge base according to the first question information to obtain first initial answer information, including: Retrieving in a target dedicated knowledge base according to the first question information, and the target dedicated knowledge base is the dedicated knowledge base of the service object corresponding to the first question information; If the first initial answer information cannot be retrieved in the target dedicated knowledge base, retrieving in the general knowledge base according to the first question information to obtain the first initial answer information.

6. The method according to claim 3, wherein Before using a question-and-answer model to perform intent understanding on the question description information to obtain first question information, it further includes: Judging whether the question description content meets the shunting condition according to at least one of the service object information, user information, and preset shunting ratio corresponding to the question description content; In the case where the question description content meets the shunting condition, determining to adopt a mode of jointly generating answer information by using a question-and-answer model and a knowledge base.

7. The method according to claim 4, characterized in that, It further includes: In the case that the problem description content does not meet the shunting conditions, generate the second initial answer information corresponding to the first question information based on the problem description content by using the Q&A model; Correct the second initial answer information to obtain the second target answer information, and output the second target answer information.

8. The method according to claim 3, wherein Further comprising: Update the first question information and the first target answer information as knowledge information into the knowledge base; and / or Train a reward model according to the feedback information of the user on the first target answer information, and perform reinforcement learning-based fine-tuning on the Q&A model by using the reward model.

9. The method according to claim 3, wherein Correct the first initial answer information to obtain the first target answer information, including: During the process of generating the first initial answer information, send the first initial answer information to the Q&A assistance end in a streaming manner, so that the Q&A assistance end can correct the first initial answer information to obtain the first target answer information when the first initial answer information is successfully received; and If a cancellation instruction sent by the Q&A assistance end is received during the process of sending the first initial answer information to the Q&A assistance end in a streaming manner, terminate the generation and sending operations of the first initial answer information.

10. The method according to claim 9, wherein Further comprising: Within a first preset time after outputting the first target answer information, respond to the withdrawal instruction sent by the Q&A assistance end and withdraw the output first target answer information; and / or After a second preset time after outputting the first target answer information, respond to the modification instruction including the modified first target answer information sent by the Q&A assistance end, and replace the previously stored unmodified first target answer information in the knowledge base with the modified first target answer information.

11. The method according to any one of claims 3-10, characterized in that, Further comprising: Determine the second question information, where the second question information includes question information with an occurrence frequency greater than a set frequency threshold and / or custom question information; Generate the third initial answer information corresponding to the second question information by using the Q&A model; Correct the third initial answer information to obtain the third target answer information; Update the second question information and the third target answer information as knowledge information into the knowledge base.

12. A generative question-answering method, characterized in that, Comprising: Receive the first initial answer information sent by the Q&A service end, where the first initial answer information is retrieved from the knowledge base for the first question information obtained by the Q&A service end through intention understanding of the question description information by using the Q&A model; Respond to the correction operation, correct the first initial answer information to obtain the first target answer information; Return the first target answer information to the Q&A service end for the Q&A service end to output the first target answer information.

13. An electronic device, characterized in that, Comprising: A memory and a processor; The memory is used for storing a computer program; the processor is coupled to the memory and is used for executing the computer program to execute the steps in any one of claims 3-11 and claim 12.

14. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it causes the processor to be able to implement the steps in the method according to any one of claims 3-11 and claim 12.

15. A computer program product, characterized in that, Comprising a computer program / instructions, when the computer program / instructions are executed by a processor, it causes the processor to be able to implement the steps in the method according to any one of claims 3-11 and claim 12.

Images