Multi-source fusion question and answer method and device, storage medium and electronic device

Abstract

The invention provides a multi-source fusion question answering method and device, a storage medium and an electronic device.The method comprises the steps that question data input by a user is received; state information of multiple question and answer models is obtained, priorities of the multiple question and answer models are obtained, and the multiple question and answer models comprise a local knowledge base, an artificial customer service model and a generative artificial intelligence AI GC model; calling a matched target question and answer model based on the state information and the priority; and inputting the question data into the target question and answer model, and returning reply data of the question data. Through the embodiment of the invention, the technical problem of low accuracy and efficiency of answers returned by a question answering system in the related technology is solved, and the speed and accuracy of obtaining the answers are improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a multi-source fusion question-answering method and device, a storage medium, and an electronic device. Background Art

[0002] In related technologies, with the development of Internet technology, the demand for customer service is growing, and the traditional customer service model can no longer meet the needs of modern enterprises. The development of artificial intelligence technology provides new possibilities for improving customer service. However, existing technologies are either a single knowledge base system that lacks flexibility and has rigid answers of low quality; or they combine certain AI capabilities, but the accuracy and efficiency are not high enough to quickly answer and handle more personalized questions; some also include the capabilities of manual customer service, but cannot solve the problem of efficiency in answering.

[0003] With respect to the above-mentioned problems existing in the related technologies, no efficient and accurate solutions have been found yet. Summary of the invention

[0004] The present invention provides a multi-source fusion question-answering method and device, a storage medium, and an electronic device to solve technical problems in related technologies.

[0005] According to one embodiment of the present invention, a multi-source fusion question and answer method is provided, including: receiving question data input by a user; obtaining status information of multiple question and answer models, and obtaining priorities of multiple question and answer models, wherein the multiple question and answer models include a local knowledge base, a manual customer service model, and a generative artificial intelligence AI GC model; calling a matching target question and answer model based on the status information and the priority; inputting the question data into the target question and answer model, and returning reply data of the question data.

[0006] Optionally, calling a matching target question and answer model based on the status information and the priority includes: selecting a question and answer model set with an online status from the multiple question and answer models based on the status information; sorting the question and answer model set based on the priority of each question and answer model in the question and answer model set to obtain a question and answer model sequence; and calling the question and answer model with the highest priority in the question and answer model sequence as the target question and answer model.

[0007] Optionally, the target question and answer model is a manual customer service model, the question data is input into the target question and answer model, and reply data to the question data is returned, including: inputting the question data to a first customer service agent in an idle state; while waiting for the first customer service agent, monitoring real-time questions of other occupied customer service agents; determining whether the real-time question is similar to the question data; if the real-time question is similar to the question data, returning the reply data to the real-time question as the reply data to the question data.

[0008] Optionally, obtaining the priorities of multiple question-answering models includes: parsing the question type of the question data; respectively calculating the first matching degrees between the question type and the multiple question-answering models; and configuring the priorities of the multiple question-answering models based on the first matching degrees.

[0009] Optionally, parsing the question type of the question data includes: performing semantic parsing on the question data to obtain multiple entity words; searching for a classification label that hits any of the entity words in a preset label library; and determining the label type of the classification label as the question type of the question data.

[0010] Optionally, obtaining the priorities of multiple question and answer models includes: reading the receiving time of the question data; locating the time period in which the receiving time is located; respectively calculating the second matching degrees between the time period and the multiple question and answer models; and configuring the priorities of the multiple question and answer models based on the second matching degrees.

[0011] Optionally, after inputting the question data into the target question and answer model, the method further includes: inputting the question data into other question and answer models other than the target question and answer model; monitoring the answer return status of the target question and answer model and other question and answer models; if there is an abnormal question and answer model that does not return an answer, after associating the question data and the reply data, synchronizing them to the database or training set of the abnormal question and answer model.

[0012] Optionally, the method also includes: obtaining question and answer data of the artificial customer service model, wherein the question and answer data includes question sample data and answer sample data; determining whether there is a hit classification label for the question and answer data; if there is no hit classification label for the question and answer data, adding a classification label to the question and answer data; using the question and answer data and the classification label to generate fine-tuning task samples; and using the fine-tuning task samples to perform fine-tuning training on the multiple question and answer models.

[0013] According to another embodiment of the present invention, a multi-source fusion question and answer device is provided, including: a receiving module for receiving question data input by a user; a first acquisition module for acquiring status information of multiple question and answer models, and acquiring priorities of multiple question and answer models, wherein the multiple question and answer models include a local knowledge base, a manual customer service model, and a generative artificial intelligence AI GC model; a calling module for calling a matching target question and answer model based on the status information and the priority; a question and answer module for inputting the question data into the target question and answer model, and returning reply data of the question data.

[0014] Optionally, the calling module includes: a selection unit, used to select a question and answer model set with an online status from the multiple question and answer models based on the status information; a sorting unit, used to sort the question and answer model set based on the priority of each question and answer model in the question and answer model set to obtain a question and answer model sequence; and a calling unit, used to call the question and answer model with the highest priority in the question and answer model sequence as the target question and answer model.

[0015] Optionally, the target question and answer model is a manual customer service model, and the question and answer module includes: an input unit, used to input the question data to a first customer service seat in an idle state; a monitoring unit, used to monitor real-time questions of other customer service seats in an occupied state while waiting for the first customer service seat; a judgment unit, used to judge whether the real-time question is similar to the question data; and a return unit, used to return the answer data of the real-time question as the answer data of the question data if the real-time question is similar to the question data.

[0016] Optionally, the first acquisition module includes: a parsing unit, used to parse the question type of the question data; a first calculation unit, used to respectively calculate the first matching degree between the question type and the multiple question and answer models; and a first configuration unit, used to configure the priority of the multiple question and answer models based on the first matching degree.

[0017] Optionally, the parsing unit includes: a parsing subunit, used to perform semantic parsing on the question data to obtain multiple entity words; a searching subunit, used to search for a classification label that hits any of the entity words in a preset label library; and a determining subunit, used to determine the label type of the classification label as the question type of the question data.

[0018] Optionally, the first acquisition module includes: a reading unit, used to read the receiving time of the question data; a positioning unit, used to locate the time period in which the receiving time is located; a second calculation unit, used to respectively calculate the second matching degree between the time period and the multiple question and answer models; and a second configuration unit, used to configure the priority of the multiple question and answer models based on the second matching degree.

[0019] Optionally, the device also includes: an input module, which is used to input the question data into other question and answer models other than the target question and answer model after the question and answer module inputs the question data into the target question and answer model; a monitoring module, which is used to monitor the answer return status of the target question and answer model and other question and answer models; and a synchronization module, which is used to synchronize the question data to the database or training set of the abnormal question and answer model after associating the question data with the reply data if there is an abnormal question and answer model that has not returned an answer.

[0020] Optionally, the device also includes: a second acquisition module, used to acquire the question and answer data of the manual customer service model, wherein the question and answer data includes question sample data and answer sample data; a judgment module, used to judge whether the question and answer data has a hit classification label; an adding module, used to add a classification label to the question and answer data if the question and answer data does not have a hit classification label; a generation module, used to generate fine-tuning task samples using the question and answer data and the classification labels; and a fine-tuning module, used to use the fine-tuning task samples to fine-tune the multiple question and answer models.

[0021] According to another aspect of an embodiment of the present application, a storage medium is further provided, which includes a stored program, and the above steps are executed when the program is run.

[0022] According to another aspect of an embodiment of the present application, there is also provided an electronic device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; wherein: the memory is used to store computer programs; and the processor is used to execute the steps in the above method by running the program stored in the memory.

[0023] The embodiment of the present application also provides a computer program product including instructions, which, when executed on a computer, enables the computer to execute the steps in the above method.

[0024] Beneficial effects of the present invention:

[0025] 1. Improve the speed and accuracy of answer acquisition: Through the large model optimization algorithm and intelligent scheduling module, the most suitable answer source can be quickly found according to the type of question and the availability of the current answer source, and the most accurate answer can be returned;

[0026] 2. Improve the efficiency of manual customer service: Data synchronization module and manual processing module can ensure that manual customer service can also get fast and accurate responses;

[0027] 3. High scalability: It can switch seamlessly, increase or decrease the answer source, and facilitate system maintenance and upgrade. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0029] Figure 1 is a hardware structure block diagram of a server according to an embodiment of the present invention;

[0030] Figure 2 is a flow chart of a multi-source fusion question-answering method according to an embodiment of the present invention;

[0031] Figure 3 is a workflow diagram of the fusion question-answering system in an embodiment of the present invention;

[0032] Figure 4 is a processing flow chart of a scheduling model in an embodiment of the present invention;

[0033] Figure 5 is a processing flow chart of an extensible protocol module in an embodiment of the present invention;

[0034] Figure 6 is a processing flow chart of data synchronization in an embodiment of the present invention;

[0035] Figure 7 It is a structural block diagram of a multi-source fusion question-answering device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0036] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only embodiments of a part of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present application. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0037] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0038] Example 1

[0039] The method embodiment provided in the first embodiment of the present application can be executed in a car, a server, a processor, a controller or a similar processing device. Taking running on a server as an example, Figure 1 FIG. 1 is a hardware structure diagram of a server according to an embodiment of the present invention. Figure 1 As shown, the server may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data. Optionally, the server may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above server. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0040] The memory 104 can be used to store server programs, for example, software programs and modules of application software, such as a server program corresponding to a multi-source fusion question-answering method of a server in an embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the server program stored in the memory 104, that is, to implement the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the server via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0041] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the server. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0042] In this embodiment, a multi-source fusion question-answering method is provided. Figure 2 is a flow chart of a multi-source fusion question-answering method according to an embodiment of the present invention. Figure 2 As shown, the process includes the following steps:

[0043] Step S202, receiving question data input by the user;

[0044] The question data in this embodiment can be data in the format of text, picture, video, etc., integrating multiple input methods: integrating multiple question input methods, such as text, picture, video, document, audio and other common methods, and supporting the expansion of more input methods.

[0045] The input module of this embodiment supports multiple formats of question input. In addition to basic text, it also supports input of pictures, graphics, voice, video, documents and other types. It is encapsulated through a unified data protocol and supports expansion of other input types.

[0046] After the user enters the question input box, question recommendations will be made based on the user's previous question and answer history. In particular, for scenarios where manual customer service is not available, this question will be recommended first.

[0047] Step S204, obtaining status information of multiple question-answering models and obtaining priorities of multiple question-answering models, wherein the multiple question-answering models include a local knowledge base, a manual customer service model, and a generative artificial intelligence AI GC model;

[0048] The question and answer data of the manual customer service model is a labeled data set. The manual customer service model and the local knowledge base model are independent small models split from the full data model of the AI ​​GC model. Through distributed processing, the efficiency of model training can be improved.

[0049] The manual customer service model mainly processes data sets of manual questions and answers, and the local knowledge base mainly processes data sets of enterprise standardized questions and answers.

[0050] Step S206, calling a matching target question-answering model based on the state information and the priority;

[0051] Step S208, input the question data into the target question-answering model, and return the answer data of the question data.

[0052] Through the above steps, the question data input by the user is received; the status information of multiple question-answering models is obtained, and the priorities of multiple question-answering models are obtained, wherein the multiple question-answering models include a local knowledge base, a manual customer service model, and a generative artificial intelligence AI GC model; based on the status information and the priority, a matching target question-answering model is called; the question data is input into the target question-answering model, and the reply data of the question data is returned, thereby realizing a flexible scheduling scheme for the question-answering model, solving the technical problems of low accuracy and efficiency of answers returned by related technical question-answering systems, and improving the speed and accuracy of answer acquisition.

[0053] In one implementation of the present embodiment, calling a matching target question and answer model based on the status information and the priority includes: selecting a question and answer model set with an online status from the multiple question and answer models based on the status information; sorting the question and answer model set based on the priority of each question and answer model in the question and answer model set to obtain a question and answer model sequence; and calling the question and answer model with the highest priority in the question and answer model sequence as the target question and answer model.

[0054] In one example, the local knowledge base has the highest priority, followed by the AI ​​GC model, and finally the manual customer service model. The system defaults to scheduling in the order of local knowledge base, AI GC large model, and manual customer service. After the user enters the question, the scheduling module prioritizes scheduling the question to the local knowledge base for answer matching. If the match is successful, the answer is returned. If the match is unsuccessful, the question is scheduled to the AI ​​GC large model. If the match is successful and the answer is not returned successfully, the manual customer service is automatically connected for question and answer. If the service is found to be unavailable during the scheduling process, the currently available services are queried and the scheduling continues in the default order. The scheduling order supports user configuration and the scheduling strategy can be flexibly adjusted.

[0055] After calling the target question-and-answer model, you can also monitor the target question-and-answer model, connect to the answer source through the extensible protocol interface, and transmit the question to the appropriate data source through the question forwarding protocol, while monitoring the answer return and exception return of the data source. If the answer return is triggered, the answer is returned directly. If the exception callback is triggered, the forwarding interface is called again according to the question-and-answer model sequence to forward the answer to other answer sources, and then continue to monitor the answer return, repeating the above work until the process ends.

[0056] In an example of this embodiment, obtaining the priorities of multiple question-answering models includes: parsing the question type of the question data; respectively calculating the first matching degree between the question type and the multiple question-answering models; and configuring the priorities of the multiple question-answering models based on the first matching degree.

[0057] For example, question types include professional standardized questions, personalized questions, and unknown questions. Each question type matches one or more question-answering models. For example, the matching degrees of professional standardized questions and multiple question-answering models are local knowledge base > manual customer service model > AI GC model; the matching degrees of personalized questions and multiple question-answering models are AI GC model > local knowledge base > manual customer service model.

[0058] Optionally, parsing the question type of the question data includes: performing semantic parsing on the question data to obtain multiple entity words; searching for a classification label that hits any of the entity words in a preset label library; and determining the label type of the classification label as the question type of the question data.

[0059] If the question does not match any of the classification labels in the preset label library, the question type is unknown.

[0060] In an example of this embodiment, obtaining the priorities of multiple question and answer models includes: reading the receiving time of the question data; locating the time period in which the receiving time is located; respectively calculating the second matching degrees between the time period and the multiple question and answer models; and configuring the priorities of the multiple question and answer models based on the second matching degrees.

[0061] In this example, each question-and-answer model corresponds to a time period, such as 0:00-8:00, 8:00-16:00, and 17:00-24:00. If the receiving time falls within the time period corresponding to the model, the matching degree is 100%. If it does not fall within the time period corresponding to the model, the time difference from the time period is calculated. The smaller the time difference, the higher the matching degree. When the time difference is 0 (falls within the time period corresponding to the model), the matching degree is the highest and the corresponding priority is also the highest.

[0062] Implement a time series analysis module to identify peak periods of user consultations and predict future consultation patterns. During peak periods, dynamically adjust resource allocation, such as increasing computing resources or optimizing model loading strategies, to ensure fast responses. Create a time-period-sensitive model that can automatically adjust its response strategy based on the current time point. For example, if customer service questions generally increase at night, the priority of the model training set for manual customer service will be appropriately increased, and more resources will be allocated to it to speed up the response.

[0063] In addition, you can prioritize common questions and high-frequency inquiries based on frequency, and build a set of high-frequency question models that are optimized specifically for common queries, such as holiday travel guides. Use natural language processing techniques (such as TF-IDF, Word2Vec, etc.) to identify and classify high-frequency questions. High-frequency question models can be updated regularly to adapt to emerging issues and trends.

[0064] In an implementation scenario of this embodiment, the target question and answer model is a manual customer service model, the question data is input into the target question and answer model, and the reply data of the question data is returned, including: inputting the question data to a first customer service agent in an idle state; while waiting for the first customer service agent, monitoring the real-time questions of other customer service agents in an occupied state; determining whether the real-time question is similar to the question data; if the real-time question is similar to the question data, returning the reply data of the real-time question as the reply data of the question data.

[0065] If the user is waiting to ask a question, the module will call the data synchronization module to obtain whether there are similar questions in the current real-time questions. If there are similar question and answer content, it will directly return the question and answer content.

[0066] After the user asks a question, the customer service is busy and times out. At this time, the question will be automatically added to the customer service message. After the manual customer service replies to the question in the message, the question and answer content will be automatically synchronized to each answer source through the data synchronization module to ensure that the next question can be answered quickly.

[0067] In an implementation scenario of this embodiment, after the question data is input into the target question and answer model, it also includes: inputting the question data into other question and answer models other than the target question and answer model; monitoring the answer return status of the target question and answer model and other question and answer models; if there is an abnormal question and answer model that has not returned an answer, after associating the question data and the reply data, synchronizing them to the database or training set of the abnormal question and answer model.

[0068] After the current Q&A process ends normally, the Q&A results will be synchronized to each Q&A model. If the answer to the local knowledge base is hit, the answer will be synchronized to the training set of AI GC. If other users' sources are switched to the AI ​​GC module at this time, and the questions are similar, the answer can be returned directly. If the answer to AI GC is hit, the synchronization module will synchronize the Q&A content to the local knowledge base. At this time, if the source of a user is the local knowledge base, the answer can also be quickly returned. If the manual customer service auxiliary module is already working and has replied to the message, the synchronization module will also forward the Q&A content to the local knowledge base and AI GC training set, and enter the question into the historical question recommendation module. The next time the user enters, he will quickly get the answer through the historical recommended questions. In the manual customer service scenario, after the user asks a question, the synchronization module will also synchronize similar questions and answers of other online users to the current session to maximize the response rate of manual customer service and the accuracy of the question.

[0069] In this embodiment, it also includes: obtaining question and answer data of the artificial customer service model, wherein the question and answer data includes question sample data and answer sample data; determining whether the question and answer data has a hit classification label; if the question and answer data does not have a hit classification label, adding a classification label to the question and answer data; using the question and answer data and the classification label to generate a fine-tuning task sample; and using the fine-tuning task sample to fine-tune the multiple question and answer models.

[0070] Adding custom labels to human customer service data allows you to quickly identify this data during the data preprocessing stage and assign it to a dedicated fine-tuning process. When fine-tuning, you can use smaller batch sizes and fewer training rounds because the human responses are already close to the final desired format and style.

[0071] The large model tag extracts key information from the original data to form a data carrier that can be directly used by the business. In this way, the large model tag can effectively organize and process massive data, improve data processing efficiency, and ensure that complex multi-table association and joint query logic processing will not affect the logical judgment of business data, thereby eliminating the performance impact caused by complex logic processing.

[0072] Based on the big model algorithm, it is specially used to process these optimized customer service data to achieve rapid iteration and deployment, and at the same time form an artificial customer service model set. Similarly, the local knowledge base is also an approximate standard answer, and it can also be processed and beautified with a small amount of big model algorithm to form an independent local knowledge base model set.

[0073] On the existing training set, the previous model is further fine-tuned through the data set of manual customer service, and the time periods and question types when manual customer service frequently occurs are incorporated to further speed up the response speed of manual customer service related questions and answers. Because manual customer service already has relatively standard answers, there is no need for large-batch fine-tuning training, only smaller batch training is required.

[0074] The process of optimizing user question label classification includes:

[0075] Step 1: If the question raised by user 1 does not have any label in the data set, then a broad data query is performed to return the appropriate answer. At this time, the system labels the question and generates the first question label set [tag1]. Tag1 corresponds to a label in the answer data set, such as the manual customer service data set.

[0076] Step 2: Similarly, user 2 also goes through step (1) when asking a question, so the question tag set becomes [tag1, tag2];

[0077] Step 3: User 3 asks a question and hits tag1. Since tag1 exists and corresponds to data of a certain tag, the corresponding answer can be quickly returned;

[0078] Step 4: The question label set will also be added to the fine-tuning task of the large model to facilitate faster label matching later.

[0079] Questions and data are labeled, and the labels of questions and data are related to each other, such as manual customer service labels or local knowledge base labels. In massive data, data is divided into piles by labels, and query time can be greatly reduced by matching labels during query.

[0080] Figure 3 It is a workflow diagram of the fusion question-answering system in an embodiment of the present invention. The fusion question-answering system includes: a user input module, an intelligent scheduling module, a knowledge base fusion module, a data synchronization module, a manual assistance module, an extensible protocol module, a large model optimization algorithm, etc.

[0081] User input module: This module supports multiple input methods. In addition to traditional text input, it also supports common file input such as pictures, videos, documents, etc., and can support more input formats through abstract interfaces. It also supports historical manual question recommendations;

[0082] Intelligent scheduling module: This module is mainly responsible for scheduling questions to various answer sources while taking into account the high availability of the service;

[0083] Knowledge base fusion module: a module that integrates the local knowledge base, AI GC training set, and manual customer service system;

[0084] Data synchronization module: based on user feedback, timely synchronize question and answer data to various answer sources, and support data synchronization of various synchronization modules to ensure that answers can be responded to quickly;

[0085] Manual assistance module: used to solve the problem of waiting for manual service;

[0086] Extensible protocol module: through abstract interface, it supports the expansion of more answer sources and enriches the question-answering system;

[0087] Large model optimization algorithm: Optimize the model training algorithm through aggregation strategies such as labels, time periods, frequencies, and small models.

[0088] After the user enters the question-and-answer system page, the user input module will first recommend historical questions. After the user sends the question, the scheduling system will schedule the question to the appropriate answer source for processing, and perform unified monitoring of questions and answers through the extended protocol. After obtaining the answer from each answer source, the data synchronization module will synchronize the answer to each source to enrich the answer library. For the scenario of manual customer service, the manual assistance module will perform specific optimization processing for the scenario that is not responded to in time. Ultimately, it is guaranteed that users can get answers quickly and accurately.

[0089] Figure 4 It is a processing flow chart of the scheduling model in an embodiment of the present invention, including two tasks: scheduling strategy and question-and-answer monitoring.

[0090] Scheduling strategy: The system defaults to scheduling in the order of local knowledge base, AI GC large model, and manual customer service. After the user enters the question, the scheduling module prioritizes scheduling the question to the local knowledge base for answer matching. If the match is successful, the answer is returned. If the match is unsuccessful, the question is scheduled to the AI ​​GC large model. If the match is successful, the answer is returned. If unsuccessful, the manual customer service is automatically connected for question and answer. If the service is found to be unavailable during the scheduling process, the currently available services are queried and scheduling continues in the default order. The scheduling order supports user configuration, and the scheduling strategy can be flexibly adjusted.

[0091] Question and answer monitoring: The module connects to the answer source through an extensible protocol interface. Through the question forwarding protocol, the questions are transmitted to the appropriate data source, and the answer return and exception return of the data source are monitored. If the answer return is triggered, the answer is returned directly. If the exception callback is triggered, the forwarding interface is called again to forward the answer to other answer sources, and then continue to monitor the answer return, repeating the above work until the process ends.

[0092] Figure 5It is a processing flow chart of the extensible protocol module in the embodiment of the present invention. The module defines a series of extended protocols, and formulates a question forwarding protocol, an answer monitoring protocol, and an exception monitoring protocol for the scheduling system. For the answer source, it formulates a question receiving protocol, an answer sending protocol, and an exception sending protocol.

[0093] Before the user asks a question, the answer source needs to implement the question receiving protocol first. After the user asks the question, the scheduling module calls the question forwarding protocol to uniformly forward the question to the answer source and implement the answer and exception monitoring protocol at the same time.

[0094] After the question is forwarded to the answer source, it is matched, and after the answer is hit, the answer sending protocol is called to send the answer to the answer listening protocol of the scheduling system to obtain the answer;

[0095] If no answer is found or other exceptions occur, the exception sending protocol is called to send the exception to the exception listener registered by the scheduling system. When the scheduling module receives the exception information, it dispatches the question to other answer sources;

[0096] If there is a new answer source that needs to be connected, you only need to implement the question receiving protocol, answer sending protocol, and exception sending protocol, and add them to the scheduling system;

[0097] If you want to delete a source, just remove it from the scheduling pool.

[0098] Figure 6 It is a processing flow chart of data synchronization in an embodiment of the present invention. This module is used to synchronize the question and answer results to various data sources. If the answer of the local knowledge base is hit, the answer will be synchronized to the training set of AI GC. If the source of other users is switched to the AI ​​GC module at this time, and the question is similar, then the answer can be directly returned. If the answer of AI GC is hit, the synchronization module will synchronize the question and answer content to the local knowledge base. At this time, if the source of a user is the local knowledge base, the answer can also be quickly returned. If the manual customer service auxiliary module has been working and has replied to the message, the synchronization module will also forward the question and answer content to the local knowledge base and the AI ​​GC training set, and enter the question into the historical question recommendation module. The next time the user enters, he will quickly get the answer through the historical recommended questions. In the manual customer service scenario, after the user asks a question, the synchronization module will also synchronize similar questions and answers of other online users to the current session to ensure the response rate of manual customer service and the accuracy of the question as much as possible.

[0099] This embodiment proposes a multi-source fusion question-and-answer system, which provides users with an efficient, intelligent, and personalized customer service solution by integrating the advantages of AI GC, local knowledge base, and manual customer service. The background design of the system focuses on scalability, can adapt to changing business needs and technological advances, and provides enterprises with a powerful customer service support platform. This embodiment implements multi-source fusion question-and-answer technology, in which the fusion is mainly reflected in the unification of external interfaces and the synchronization and fusion of internal data. Through labeling, the internal data is divided into three types: manual customer service data, local knowledge base data, and other data. Different types of APPs have different sensitivities to different types of data. For example, car companies' APPs are more sensitive to the data of their own vehicle-related knowledge bases and user feedback data. Then increasing the weight of the local knowledge base of the manual customer service data set is a more normal processing method. Because there is a data synchronization mechanism, the data of the manual customer service and the local knowledge base will be synchronized with each other, and after fusion, they will be synchronized to the general AI GC data set, and then further integrated.

[0100] Integration of multiple input methods: It integrates multiple question input methods, such as text, pictures, videos, documents, audio and other common methods, and supports the expansion of more input methods. Flexible switching of answer sources: You can flexibly switch between the local knowledge base, AI GC large model and manual customer service question and answer system. Timely synchronization of question and answer data: According to the user's feedback on the answer, the question and answer data is synchronized to the local knowledge base, AI large model and manual customer service system in a timely manner, so that similar questions can be answered quickly in the future. Faster manual customer service: It can maximize the guarantee that manual customer service can respond to the answers that users are concerned about in a timely manner, eliminating the trouble of waiting. Scalability design: The system design takes into account the future development of business and technology, and supports the adjustment and expansion of the knowledge base at any time.

[0101] The solution of this embodiment improves the speed and accuracy of answer acquisition: through the large model optimization algorithm and intelligent scheduling module, the most suitable answer source can be quickly found according to the type of question and the available status of the current answer source, and the most accurate answer can be returned. The efficiency of manual customer service is improved: the data synchronization module and the manual processing module can ensure that manual customer service can also get fast and accurate replies. High scalability: It can switch, increase, and reduce answer sources without feeling, which is convenient for system maintenance and upgrade.

[0102] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0103] Example 2

[0104] In this embodiment, a multi-source fusion question-answering device is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and will not be repeated hereafter. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0105] Figure 7 is a structural block diagram of a multi-source fusion question-answering device according to an embodiment of the present invention. Figure 7 As shown, the device comprises:

[0106] The receiving module 70 is used to receive question data input by the user;

[0107] A first acquisition module 72, configured to acquire status information of a plurality of question-answering models and to acquire priorities of the plurality of question-answering models, wherein the plurality of question-answering models include a local knowledge base, an artificial customer service model, and a generative artificial intelligence AIGC model;

[0108] A calling module 74, configured to call a matching target question-answering model based on the state information and the priority;

[0109] The question-answering module 76 is used to input the question data into the target question-answering model and return answer data of the question data.

[0110] Optionally, the calling module includes: a selection unit, used to select a question and answer model set with an online status from the multiple question and answer models based on the status information; a sorting unit, used to sort the question and answer model set based on the priority of each question and answer model in the question and answer model set to obtain a question and answer model sequence; and a calling unit, used to call the question and answer model with the highest priority in the question and answer model sequence as the target question and answer model.

[0111] Optionally, the target question and answer model is a manual customer service model, and the question and answer module includes: an input unit, used to input the question data to a first customer service seat in an idle state; a monitoring unit, used to monitor real-time questions of other customer service seats in an occupied state while waiting for the first customer service seat; a judgment unit, used to judge whether the real-time question is similar to the question data; and a return unit, used to return the answer data of the real-time question as the answer data of the question data if the real-time question is similar to the question data.

[0112] Optionally, the first acquisition module includes: a parsing unit, used to parse the question type of the question data; a first calculation unit, used to respectively calculate the first matching degree between the question type and the multiple question and answer models; and a first configuration unit, used to configure the priority of the multiple question and answer models based on the first matching degree.

[0113] Optionally, the parsing unit includes: a parsing subunit, used to perform semantic parsing on the question data to obtain multiple entity words; a searching subunit, used to search for a classification label that hits any of the entity words in a preset label library; and a determining subunit, used to determine the label type of the classification label as the question type of the question data.

[0114] Optionally, the first acquisition module includes: a reading unit, used to read the receiving time of the question data; a positioning unit, used to locate the time period in which the receiving time is located; a second calculation unit, used to respectively calculate the second matching degree between the time period and the multiple question and answer models; and a second configuration unit, used to configure the priority of the multiple question and answer models based on the second matching degree.

[0115] Optionally, the device also includes: an input module, which is used to input the question data into other question and answer models other than the target question and answer model after the question and answer module inputs the question data into the target question and answer model; a monitoring module, which is used to monitor the answer return status of the target question and answer model and other question and answer models; and a synchronization module, which is used to synchronize the question data to the database or training set of the abnormal question and answer model after associating the question data with the reply data if there is an abnormal question and answer model that has not returned an answer.

[0116] Optionally, the device also includes: a second acquisition module, used to acquire the question and answer data of the manual customer service model, wherein the question and answer data includes question sample data and answer sample data; a judgment module, used to judge whether the question and answer data has a hit classification label; an adding module, used to add a classification label to the question and answer data if the question and answer data does not have a hit classification label; a generation module, used to generate fine-tuning task samples using the question and answer data and the classification labels; and a fine-tuning module, used to use the fine-tuning task samples to fine-tune the multiple question and answer models.

[0117] It should be noted that the above modules can be implemented by software or hardware. For the latter, it can be implemented in the following ways, but not limited to: the above modules are all located in the same processor; or the above modules are located in different processors in any combination.

[0118] Example 3

[0119] An embodiment of the present invention further provides a storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0120] Optionally, in this embodiment, the storage medium may be configured to store a computer program for performing the following steps:

[0121] S1, receiving question data input by the user;

[0122] S2, obtaining status information of multiple question-answering models and obtaining priorities of multiple question-answering models, wherein the multiple question-answering models include a local knowledge base, an artificial customer service model, and a generative artificial intelligence AI GC model;

[0123] S3, calling a matching target question-answering model based on the state information and the priority;

[0124] S4, inputting the question data into the target question-answering model, and returning answer data of the question data.

[0125] Optionally, in this embodiment, the above-mentioned storage medium may include but is not limited to: a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store computer programs.

[0126] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0127] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0128] Optionally, in this embodiment, the processor may be configured to perform the following steps through a computer program:

[0129] S1, receiving question data input by the user;

[0130] S2, obtaining status information of multiple question-answering models and obtaining priorities of multiple question-answering models, wherein the multiple question-answering models include a local knowledge base, an artificial customer service model, and a generative artificial intelligence AI GC model;

[0131] S3, calling a matching target question-answering model based on the state information and the priority;

[0132] S4, inputting the question data into the target question-answering model, and returning answer data of the question data.

[0133] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments and optional implementation modes, and this embodiment will not be described in detail here.

[0134] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0135] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0136] It should be understood that the terms used herein are only for the purpose of describing specific example embodiments and are not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms "one", "an" and "said" as used herein may also be meant to include plural forms. The terms "include", "comprise", "contain", and "have" are inclusive, and therefore specify the existence of stated features, steps, operations, elements and / or parts, but do not exclude the existence or addition of one or more other features, steps, operations, elements, parts, and / or combinations thereof. The method steps, processes, and operations described herein are not interpreted as necessarily requiring them to be performed in the specific order described or illustrated, unless the execution order is clearly indicated. It should also be understood that additional or alternative steps may be used.

[0137] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A multi-source fusion question answering method, It is characterized in that include: Receive question data input by the user; Obtaining status information of multiple question-answering models and obtaining priorities of multiple question-answering models, wherein the multiple question-answering models include a local knowledge base, an artificial customer service model, and a generative artificial intelligence AIGC model; Invoke a matching target question-answering model based on the state information and the priority; The question data is input into the target question answering model, and answer data of the question data is returned.

2. The method according to claim 1, It is characterized in that Calling a matching target question-answering model based on the state information and the priority includes: Selecting a question-answering model set of online status from the multiple question-answering models based on the status information; Sorting the question-answering model set based on the priority of each question-answering model in the question-answering model set to obtain a question-answering model sequence; The question-answering model with the highest priority in the question-answering model sequence is called as the target question-answering model.

3. The method according to claim 1, It is characterized in that The target question-answering model is a manual customer service model, the question data is input into the target question-answering model, and the answer data of the question data is returned, including: Inputting the problem data into a first customer service agent who is in an idle state; While waiting for the first customer service agent, monitoring the real-time problems of other customer service agents in occupied states; Determining whether the real-time problem is similar to the problem data; If the real-time question is similar to the question data, the answer data of the real-time question is returned as the answer data of the question data.

4. The method according to claim 1, It is characterized in that The priorities for obtaining multiple question-answering models include: Analyze the question type of the question data; Calculating first matching degrees between the question type and the multiple question-answering models respectively; Configure the priorities of the multiple question-answering models based on the first matching degree.

5. The method according to claim 4, It is characterized in that The types of questions for parsing the question data include: Performing semantic analysis on the question data to obtain multiple entity words; Searching for a classification label that matches any of the entity words in a preset label library; The label type of the classification label is determined as the question type of the question data.

6. The method according to claim 1, It is characterized in that The priorities for obtaining multiple question-answering models include: Read the receiving time of the problem data; The time period in which the positioning was received; respectively calculating a second degree of matching between the time period and the plurality of question-answering models; Configure the priorities of the multiple question-answering models based on the second matching degree.

7. The method according to claim 1, It is characterized in that After inputting the question data into the target question answering model, the method further includes: Inputting the question data into other question answering models except the target question answering model; Monitoring the answer return status of the target question-answering model and other question-answering models; If there is an abnormal question-and-answer model that does not return an answer, the question data and the answer data are associated and then synchronized to the database or training set of the abnormal question-and-answer model.

8. The method according to claim 1, It is characterized in that The method further comprises: Acquire question and answer data of the artificial customer service model, wherein the question and answer data includes question sample data and answer sample data; Determine whether there is a matching classification label in the question and answer data; If there is no matching classification label for the question and answer data, add a classification label to the question and answer data; Generate a fine-tuning task sample using the question-answering data and the classification label; The multiple question-answering models are fine-tuned and trained using the fine-tuning task samples.

9. A multi-source fusion question-answering device, It is characterized in that include: A receiving module, used for receiving question data input by a user; A first acquisition module is used to obtain status information of multiple question-answering models and obtain priorities of multiple question-answering models, wherein the multiple question-answering models include a local knowledge base, an artificial customer service model, and a generative artificial intelligence AIGC model; A calling module, configured to call a matching target question-answering model based on the state information and the priority; The question-answering module is used to input the question data into the target question-answering model and return the answer data of the question data.

10. The device according to claim 9, It is characterized in that The calling module includes: A selection unit, configured to select a question-answering model set in an online state from among the multiple question-answering models based on the state information; A sorting unit, used to sort the question and answer model set based on the priority of each question and answer model in the question and answer model set to obtain a question and answer model sequence; A calling unit is used to call the question-answering model with the highest priority in the question-answering model sequence as the target question-answering model.

11. The device according to claim 9, It is characterized in that The device also includes: A second acquisition module is used to acquire question and answer data of the artificial customer service model, wherein the question and answer data includes question sample data and answer sample data; A judgment module, used to judge whether the question and answer data has a matching classification label; An adding module, used for adding a classification label to the question and answer data if there is no matching classification label for the question and answer data; A generation module, used to generate a fine-tuning task sample using the question-answering data and the classification label; A fine-tuning module is used to perform fine-tuning training on the multiple question-answering models using the fine-tuning task samples.

12. A storage medium, It is characterized in that The storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 8 when executed.

13. An electronic device comprising a memory and a processor, It is characterized in that A computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 8.

Images