Intelligent auxiliary method, electronic equipment, readable storage medium and program product

Abstract

The embodiment of the invention provides an intelligent auxiliary method, electronic equipment, a readable storage medium and a program product. The intelligent assistance method comprises the steps that according to session information of a first user, a domain generative model corresponding to the session information is determined, user information of a second user is obtained, and the user information of the second user at least comprises attribute information and / or historical session information; on the basis of the session information of the first user and the user information of the second user, generating prompt information and inputting the prompt information into a field generation type model so as to obtain corresponding generation information through the field generation type model; and performing semantic analysis on the generated information, and generating structured data responding to the session information of the first user according to a semantic analysis result. By determining and using the domain generative model corresponding to the session information of the first user, the domain generative model is a pre-trained model, so that the accuracy of model processing can be improved, the processing time of the model can be shortened, and the service efficiency in scenes such as customer service can be improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of computer technology, and in particular to an intelligent assistance method, an electronic device, a readable storage medium, and a program product. Background Art

[0002] Customer Relationship Management (CRM) is a strategy and technology that refers to the use of information technology to coordinate the interaction between enterprises and customers in sales, marketing and services in order to improve core competitiveness, provide services to customers, and thus promote the long-term development and profit growth of enterprises. The core concept of customer relationship management is to put customers at the core, pay attention to customer needs and opinions, and improve customer satisfaction.

[0003] Among these, it has become a common operating method for service personnel to communicate with customers through audio or video, recommend products to customers, record customer needs and opinions, and track customer service history.

[0004] Currently, CRM usually only supports establishing contact with customers through audio and video and recording the call process, while other service information with customers needs to be manually sorted out by service personnel based on the call process, resulting in low efficiency of service personnel. Summary of the invention

[0005] In view of this, embodiments of the present application provide an intelligent assistance method, an electronic device, a readable storage medium, and a program product to at least partially solve the above-mentioned problems.

[0006] According to a first aspect of an embodiment of the present application, an intelligent assistance method is provided, comprising: determining a domain generative model corresponding to session information of a first user according to the session information of the first user; and obtaining user information of a second user according to the session information of the first user, wherein the user information of the second user at least includes: attribute information of the second user and / or historical session information of the second user; generating prompt information based on the session information of the first user and the user information of the second user and inputting the prompt information into the domain generative model, so as to generate and process the prompt information through the domain generative model to obtain corresponding generated information; performing semantic parsing on the generated information, and generating structured data in response to the session information of the first user according to the result of the semantic parsing.

[0007] According to a second aspect of an embodiment of the present application, an intelligent assistance method is provided, comprising: determining a domain generative model corresponding to the session information of a first user according to the session information of the first user; and obtaining user information of a second user according to the session information of the first user, the user information of the second user including at least: attribute information of the second user and / or historical session information of the second user; obtaining a service assistance request of the first user; the service assistance request is used to indicate generation of recommendation words for the first user, or to indicate generation of key content information based on the session information of the first user and the user information of the second user; generating prompt information based on the session information of the first user, the user information of the second user and the service assistance request and inputting the prompt information into the domain generative model, so as to generate and process the prompt information through the domain generative model to obtain structured data in response to the session information of the first user.

[0008] According to the third aspect of an embodiment of the present application, there is provided an electronic device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; the memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform an operation corresponding to the method described in the first aspect or the second aspect.

[0009] According to a fourth aspect of an embodiment of the present application, a computer storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the method described in the first aspect or the second aspect is implemented.

[0010] According to a fifth aspect of an embodiment of the present application, a computer program product is provided, comprising computer instructions, wherein the computer instructions instruct a computing device to perform operations corresponding to the method described in the first aspect or the second aspect.

[0011] According to the scheme provided by the embodiment of the present application, the domain generative model corresponding to the session information is determined according to the session information of the first user, wherein the domain generative model can be a pre-trained model corresponding to the session information of the first user, the session information of the first user is more real-time, and the domain generative model processes the session information of the first user in a targeted manner, which can improve the accuracy of model processing and shorten the processing time. In addition, the user information of the second user is obtained according to the session information of the first user, and the user information of the second user at least includes the attribute information of the second user and / or the historical session information of the second user. Based on the session information of the first user and the user information of the second user, prompt information is generated and input into the domain generative model, so that the prompt information is generated and processed by the domain generative model to obtain the corresponding generated information. Since the domain generative model is based on the user information of the second user, the session information of the first user and the user information of the second user are comprehensively considered, which improves the accuracy of the domain generative model processing. The scheme provided by the embodiment of the present application not only ensures the accuracy of the model output results, but also shortens the model processing time, and improves the service efficiency when applied to scenarios such as customer service. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0013] Figure 1 A schematic diagram of an exemplary system applicable to the embodiment of the present application;

[0014] Figure 2 A flow chart of the intelligent assistance method provided in the embodiment of the present application;

[0015] Figure 3 Another flow chart of the intelligent assistance method provided in the embodiment of the present application;

[0016] Figure 4A to Figure 4B A group of schematic diagrams of a display interface displayed by a user device provided in an embodiment of the present application;

[0017] Figure 5 A schematic diagram of a preset template provided in an embodiment of the present application;

[0018] Figure 6A to Figure 6C A set of schematic diagrams of application examples of the intelligent assistance method provided in the embodiments of the present application;

[0019] Figure 7 It is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of 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 part of the embodiments of the present application, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the embodiments of the present application should fall within the scope of protection of the embodiments of the present application.

[0021] The specific implementation of the embodiment of the present application is further explained below in conjunction with the accompanying drawings of the embodiment of the present application.

[0022] Figure 1 An exemplary system applicable to the embodiment of the present application is shown. Figure 1 As shown, the system 100 may include a cloud service end 102, a communication network 104 and / or one or more user devices 106. Figure 1 The example in FIG. 1 is a plurality of user devices.

[0023] The cloud server 102 may be any suitable device for storing information, data, programs and / or any other suitable type of content, including but not limited to a distributed storage system device, a server cluster, a computing cloud server cluster, etc. In some embodiments, the cloud server 102 may perform any suitable function. For example, in some embodiments, a domain generative model is deployed on the cloud server 102 to process the session information of the first user. In some embodiments, the cloud server 102 determines a domain generative model corresponding to the session information based on the session information of the first user; and obtains user information of the second user based on the session information of the first user, the user information of the second user at least including: attribute information of the second user and / or historical session information of the second user; therefore, based on the session information of the first user and the user information of the second user, prompt information is generated and input into the domain generative model, so that the prompt information is generated and processed by the domain generative model to obtain corresponding generated information; semantic parsing is performed on the generated information, and structured data in response to the session information of the first user is generated according to the semantic parsing result. In some embodiments, the cloud service end 102 determines a domain generative model corresponding to the session information according to the session information of the first user; and obtains the user information of the second user according to the session information of the first user; obtains the service assistance request of the first user for the second user; generates prompt information based on the session information of the first user, the user information of the second user and the service assistance request, and inputs the domain generative model, so as to generate and process the prompt information through the domain generative model, and obtain service assistance information in response to the service assistance request. In some embodiments, the service assistance request is used to indicate the generation of a recommended speech for the first user, for example, the first user is a customer service staff, and the service assistance request is used to indicate the generation of a recommended speech for the customer service staff. In some embodiments, the service assistance request is used to indicate the generation of the session information of the first user and the key content information of the second user, for example, the first user is an after-sales staff, the second user is a customer, and in the scenario where the after-sales staff visits the customer, the service assistance request is used to indicate the generation of customer notes, product notes, sales notes, etc. about the after-sales staff and the customer. In some instances, the cloud service end 102 may also send relevant data to the user device 106, for example, text data converted according to the session information of the first user.

[0024] In some embodiments, the communication network 104 can be any suitable combination of one or more wired and / or wireless networks. For example, the communication network 104 can include any one or more of the following: the Internet, an intranet, a wide area network (WAN), a local area network (LAN), a wireless network, a digital subscriber line (DSL) network, a frame relay network, an asynchronous transfer mode (ATM) network, a virtual private network (VPN) and / or any other suitable communication network. The user device 106 can be connected to the communication network 104 via one or more communication links (e.g., communication link 112), and the communication network 104 can be linked to the cloud service end 102 via one or more communication links (e.g., communication link 114). The communication link can be any communication link suitable for transmitting data between the user device 106 and the cloud service end 102, such as a network link, a dial-up link, a wireless link, a hard-wired link, any other suitable communication link, or any suitable combination of such links.

[0025] The user device 106 may include any one or more user devices suitable for interacting with a user and having a human-computer interaction interface. In some embodiments, the user device 106 may receive session information related to a second user conducted by a first user, and send it to the cloud service end 102, so that the cloud service end 102 can obtain corresponding generated information using a domain generative model corresponding to the session information, and then generate structured data in response to the session information of the first user based on the generated information. In some embodiments, the user device 106 may include any suitable type of device. For example, in some embodiments, the user device 106 may include a mobile device, a tablet computer, a laptop computer, a desktop computer, a wearable computer, a game console, a media player, a vehicle entertainment system, and / or any other suitable type of user device.

[0026] Based on the above system, the solution of the present application is described through multiple embodiments.

[0027] Reference Figure 2 , which shows a flow chart of the intelligent assistance method provided by the embodiment of the present application. The execution subject of this embodiment can be Figure 1 The cloud service end 102 is shown.

[0028] like Figure 2 As shown, the intelligent assistance method provided in this embodiment may include:

[0029] Step S202: Determine a domain generative model corresponding to the session information of the first user, and obtain user information of the second user according to the session information of the first user, wherein the user information of the second user at least includes: attribute information of the second user and / or historical session information of the second user.

[0030] Specifically, the conversation information of the first user involves the second user. For the convenience of explanation, this embodiment takes the first user as a service personnel, the second user as a customer, and the scenario in which the first user provides services to the second user as an example to illustrate the intelligent assistance method provided by this embodiment. For example, service personnel include but are not limited to: sales personnel, after-sales personnel, customer return personnel, etc., and this embodiment does not limit the names of personnel. Exemplary service scenarios include but are not limited to: sales personnel talk with customers through the CRM platform to promote related products to customers, or customer return personnel visit customers on-site to understand customers' product usage needs and feedback.

[0031] The session information of the first user may include audio data and / or video data. The session information may be input by the service personnel through the user device, and the user device sends the session information to the cloud service end; or, the session information may be obtained from the cloud storage device. Depending on the stage of the conversation between the service personnel and the customer, the session information includes but is not limited to any of the following: audio or video data generated in real time during the conversation between the service personnel and the customer, audio or video data of the conversation process after the service personnel and the customer have finished the conversation, and audio data about the customer input by the service personnel through the user device after contacting the customer online or offline.

[0032] Determine a domain generative model corresponding to the session information according to the session information of the first user. The domain generative model may be a pre-trained model corresponding to the session information of the first user. It is understandable that the session information of the first user is usually highly real-time, and therefore has certain requirements for processing latency. The domain generative model is pre-trained for the session information, and therefore, applying the domain generative model to process the session information of the first user can improve the accuracy of model processing and shorten the processing time of the model.

[0033] Optionally, the domain generative model includes a short text-specific model. The short text-specific model is a generative basic model pre-trained for short text input or short text output. Generally, the more input data the model has, or the more output data the model has, the longer the model processing time is. Since the first user's session information is usually highly real-time, the shorter the model processing time is, the better. By using a short text-specific model, the model processing time can be further shortened.

[0034] Optionally, the domain generation model may also include at least one of the following: a customer domain-specific model, a long text-specific model, and a preset function-specific model.

[0035] Among them, customer domain-specific models are pre-trained for the domain to which the customer belongs. Long text-specific models are pre-trained for long text input or long text output. Preset function-specific models are pre-trained for preset service functions, such as handling customer complaints and generating recommended words for service personnel.

[0036] It can be understood that different domain generative models have different effectiveness, performance, and cost. Generally, the more parameters a model has and the larger its scale, the more accurate its output results will be, but the longer the processing time will be. The more input data a model has, the longer the processing time will be. The more output data a model has, the longer the processing time will be. Therefore, by pre-training multiple types of domain generative models and selecting flexible calls based on the session information of the first user, the flexibility of model calls is improved, the efficiency of model processing is improved in different service requirements or service scenarios, and the model processing time is shortened.

[0037] The user information of the second user is also obtained according to the session information of the first user, and the user information of the second user at least includes: attribute information of the second user and / or historical session information of the second user.

[0038] Optionally, the attribute information of the second user includes at least: user portrait data of the second user.

[0039] User portrait usually refers to a comprehensive user virtual model built by collecting and analyzing various types of user data (such as age, gender, occupation, consumption habits, interest preferences, purchase history, etc.). In the service scenario, the second user is the customer, and the user portrait data of the second user can help enterprises and service personnel understand the characteristics and needs of customers more deeply, thereby achieving precision marketing and personalized services.

[0040] Optionally, the user portrait data of the second user includes basic information of the second user, such as, but not limited to, at least one of the following: user name, user industry, user address, activation time of the user's company, scale of the user's company, etc.

[0041] Optionally, in one implementation, obtaining user portrait data of the second user includes:

[0042] Obtaining an identifier of a second user;

[0043] The user portrait data corresponding to the identifier of the second user is retrieved from the external knowledge base through a retrieval-augmented generation (RAG) method.

[0044] The identifier of the second user can distinguish different users.

[0045] The user portrait data of the second user is retrieved from the external knowledge base according to the identifier of the second user through the RAG method. Since the data in the external knowledge base will be updated in a timely manner, accurate user portrait data of the second user is obtained, thereby providing rich and accurate input data for subsequent processing using the domain generative model, thereby improving the accuracy of the model output results.

[0046] Optionally, the historical conversation information of the second user includes at least: key content information generated based on the historical conversation information of the second user, and / or the full text information of the historical conversation. For example, in some embodiments, the key content information can be stored in a preset format, such as json format. A variety of templates can also be pre-set, and the template includes multiple fields, and the key content information includes the content of each field in the preset template. Optionally, the preset templates include but are not limited to: customer note template, sales note template, product note template. Please refer to the subsequent Figure 3 Description of the illustrated embodiment. For example, in some embodiments, the historical conversation full text information includes text data generated based on the conversation information of the first user after the conversation between the first user and the second user ends.

[0047] By obtaining the historical session information of the second user, it can help enterprises and service personnel to have a deeper understanding of the customer's characteristics and needs, thereby achieving precise marketing and personalized services.

[0048] Step S204: Generate prompt information based on the session information of the first user and the user information of the second user and input the prompt information into the domain generation model, so as to generate and process the prompt information through the domain generation model to obtain corresponding generation information.

[0049] Since the prompt information is based on the session information of the first user and the user information of the second user, the accuracy of the generated prompt information is improved by the user information of the second user. Furthermore, the prompt information is generated and processed by the domain generation model to obtain the corresponding generated information, thereby improving the accuracy of the generated information. Therefore, the accuracy of the model output results is guaranteed and the model processing time is shortened.

[0050] Step S206: semantically parse the generated information, and generate structured data in response to the session information of the first user according to the semantic parsing result.

[0051] Structured data generally refers to data stored in a preset format. Optionally, structured data can be stored in json format. By semantically parsing the generated information, structured data in response to the session information of the first user is obtained according to the generated information, thereby avoiding the first user manually arranging the structured data according to the generated information, thereby further improving the processing efficiency of the session information of the first user and shortening the processing delay.

[0052] It can be seen that the intelligent assistance method provided in this embodiment determines the domain generative model corresponding to the session information according to the session information of the first user, wherein the domain generative model can be a pre-trained model corresponding to the session information of the first user, and the session information of the first user is more real-time. The domain generative model processes the session information of the first user in a targeted manner, which can improve the accuracy of model processing and shorten the processing time. In addition, the user information of the second user is obtained according to the session information of the first user, and the user information of the second user at least includes the attribute information of the second user and / or the historical session information of the second user. Based on the session information of the first user and the user information of the second user, prompt information is generated and input into the domain generative model, so that the prompt information is generated and processed by the domain generative model to obtain the corresponding generated information. Since the domain generative model is based on the user information of the second user, the accuracy of the output result of the domain generative model is improved. The solution provided in the embodiment of the present application not only ensures the accuracy of the model output result, but also shortens the model processing time, thereby improving service efficiency in scenarios such as customer service.

[0053] Optionally, in one implementation, the intelligent assistance method provided in this embodiment may further include:

[0054] The user information of the second user is displayed on the display interface of the session information of the first user.

[0055] By displaying the user information of the second user in the display interface, the first user can timely understand the relevant information of the second user. For example, in the customer service scenario, during the conversation between the first user and the second user, the first user can timely understand the basic information of the second user or the historical feedback on the demand information or usage opinions of a specific product model, which is conducive to the first user to better provide services to the second user. Figure 4B .

[0056] Optionally, in one implementation, the intelligent assistance method provided in this embodiment may further include:

[0057] Obtaining corresponding domain knowledge according to the session information of the first user;

[0058] Correspondingly, in step S204, based on the session information of the first user and the user information of the second user, prompt information is generated and input into the domain generative model, including: based on the session information of the first user, the user information of the second user, and domain knowledge, prompt information is generated and input into the domain generative model.

[0059] It can be understood that when the second user belongs to a different industry, the first user's session information involves different fields. For example, if the second user is an employee of an Internet company, then his field is the Internet field, and the first user's session information will involve the Internet field. Similarly, when the second user is an employee of a power plant, then his field is the power generation industry field, and the first user's session information may involve power generation equipment. Therefore, by obtaining the corresponding domain knowledge through the first user's session information, and generating prompt information based on the domain knowledge on the basis of the first user's session information and the second user's user information, the accuracy of the generated prompt information can be improved, and thus, the accuracy of the generated information obtained by applying the domain generative model to obtain the model output is improved.

[0060] Optionally, the domain knowledge may include product information of products related to the second user.

[0061] Optionally, in one implementation, the intelligent assistance method provided in this embodiment may further include:

[0062] Obtaining user information of the first user, where the user information of the first user at least includes: service level information of the first user, and / or idiomatic phrase information of the first user;

[0063] Correspondingly, in step S204, based on the session information of the first user and the user information of the second user, prompt information is generated and input into the domain generation model, including:

[0064] Based on the session information of the first user, the user information of the second user, and the user information of the first user, prompt information is generated and input into a domain generative model.

[0065] It is understandable that different first users usually have different service levels and idioms. For example, for customer service personnel, the longer the working time and the richer the work experience, the higher the service level. On the contrary, for customer service personnel with shorter working time, the service level is lower. Everyone's idioms are also different. Therefore, by obtaining the user information of the first user, and generating prompt information based on the user information of the first user on the basis of the session information of the first user and the user information of the second user, the accuracy of the generated prompt information can be improved, and the matching degree between the prompt information and the first user can be improved. In turn, the accuracy of the generated information output by the application domain generative model is improved, and the matching degree between the generated information output by the model and the first user is improved.

[0066] Optionally, in one implementation, the intelligent assistance method provided in this embodiment may further include:

[0067] Acquiring emotion information of a second user according to the session information of the first user;

[0068] Correspondingly, in step S204, based on the session information of the first user and the user information of the second user, prompt information is generated and input into the domain generation model, including:

[0069] Based on the session information of the first user, the user information of the second user, and the emotion information of the second user, prompt information is generated and input into a domain generation model.

[0070] Specifically, during the communication between service personnel and customers, the customer's mood is crucial. If the customer is in a good mood, the service personnel's success rate in recommending products or performing service tasks will increase. On the contrary, if the customer is in a bad mood, the service personnel need to handle it in time and appease the customer. Customer's emotional information includes but is not limited to happy, pleasant, satisfied, angry, busy, no response, etc.

[0071] In this implementation, by obtaining the customer's emotional information, the customer's emotional information is further considered when generating prompt information, so that the generated prompt information is more suitable for the customer, and the structured data output by the domain generative model based on the prompt information is also more suitable for the customer. For example, when the service assistance request is used to indicate the generation of a recommended speech for the first user, the recommended speech takes into account the customer's current emotions and is more suitable for the customer, thereby improving the service quality and the customer's service experience.

[0072] Furthermore, acquiring the emotion information of the second user according to the conversation information of the first user may include: segmenting the conversation information to obtain emotion words representing the customer's emotions, and acquiring the emotion information of the second user according to the emotion words.

[0073] Furthermore, obtaining the emotion information of the second user according to the conversation information of the first user may include: inputting the conversation information into a pre-trained emotion model to obtain the emotion information of the second user. This embodiment does not limit the type of the emotion model.

[0074] Optionally, in one implementation, in step S204, based on the session information of the first user and the user information of the second user, generating prompt information and inputting it into a domain generation model includes:

[0075] According to a preset metadata template, the session information of the first user and the user information of the second user are filled into the metadata template to generate prompt information, and the prompt information is input into the domain generation model.

[0076] Specifically, multiple metadata templates may be pre-set, and the metadata templates may include fields. This embodiment does not limit the number and content of the fields included in each metadata template. In one implementation, the fields in the metadata template include: service personnel (i.e., the first user) role, service personnel capabilities, customer (i.e., the second user) user portrait data, product information, etc. The various fields in the metadata template are filled according to the session information of the first user and the user information of the second user, and then prompt information is generated based on the metadata template filled with data, which improves the accuracy of the generated prompt information and is more suitable for the first user and the second user, thereby improving the matching degree between the generated information output by the model based on the prompt information and the first user and the second user.

[0077] Optionally, in one implementation, in step S202, determining, according to the session information of the first user, a domain generative model corresponding to the session information includes:

[0078] Determining the openness of the session information according to the session information of the first user;

[0079] If the degree of openness is less than or equal to a first threshold, determining to use a domain generative model corresponding to the session information;

[0080] If the degree of openness is greater than the first threshold, it is determined to use the general generative model.

[0081] The degree of openness of the session information is used to indicate the degree of uncertainty of the session information or to indicate the difficulty of applying the domain generative model to obtain the accuracy of the generated information based on the session information. The domain generative model has a strong processing capability, but there are still differences in the accuracy and processing time of different information. Generally, in scenarios involving calculations, time analysis, etc. that require precise mathematical calculations, the domain generative model is not good at it, and the accuracy of the generated information may be low, and the processing time of the generated information may be long.

[0082] Below, the first user's session information including time information is used as an example for explanation. Assume that the first user's session information includes the time agreed upon by the first user and the second user for the next time the first user calls the second user. Case 1: The first user will call the second user at 8 a.m. next Tuesday. Case 2: The first user will call the second user later. Then, the openness of the session information corresponding to case 1 will be less than the openness of the session information corresponding to case 2, that is, it is easier to determine the exact agreed time in case 1, or in other words, the uncertainty of the first user's session information is lower in case 1.

[0083] In this implementation, the degree of openness of the session information is compared with a first threshold value, and this embodiment does not limit the value of the first threshold value. If the degree of openness is less than or equal to the first threshold value, it is determined to use the domain generative model corresponding to the session information. Since the domain generative model is a pre-trained model corresponding to the session information of the first user, when the degree of openness of the session information of the first user is low, the accuracy of the generated information obtained can be improved and the processing time can be shortened by using the domain generative model. On the contrary, if the degree of openness is greater than the first threshold value, it is determined to use the general generative model. Since the general generative model is more universal, when the degree of openness of the session information of the first user is high, the accuracy of the generated information obtained can be improved by using the general generative model.

[0084] Optionally, in one implementation, the intelligent assistance method provided in this embodiment may further include:

[0085] The backfill information is obtained according to the session information of the first user and the user information of the second user, so as to modify and supplement the structured data using the backfill information.

[0086] Specifically, the processing capability of the domain generative model is very strong, but the domain generative model usually has more parameters, the amount of calculation is large during use, and the processing time is long. Moreover, the generation principle of the domain generative model is based on vector similarity, and the generation results of the domain generative model are usually unstable or biased towards probability. The processing of precise calculation information such as calculation and time may take a long time, but the results obtained are not necessarily accurate. In the present application, it is applied to the CRM scenario, and the structured data obtained in response to the session information of the first user has the requirements of predictability and robustness. The generated information generated by the domain generative model can be disassembled into various fields in json format, or it can be understood that the structured data in response to the session information of the first user will be preset with multiple fields. Please refer to the subsequent about Figure 5For the customer note template shown, the structured data should include the content of the predefined fields in the customer note template. For some of the fields, for example, the results obtained by using the domain generative model may be inaccurate or the domain generative model may take a long time to generate. For example, for the field "appointment time", it is often not easy and time-consuming to use the domain generative model to obtain the time scheduled by the service staff and the customer. For example, in the service scenario, the customer agrees that the next communication time is next Tuesday night. The domain generative model may not be able to determine the specific time of the next communication because it cannot obtain the current time. Moreover, for the customer relationship management system, there are many service staff and many customers. There may be multiple service staff talking to multiple customers at the same time. In this way, the domain generative model is called many times. If the domain generative model called multiple times processes this type of information, the overall time consumption is long, resulting in low service efficiency of the entire system.

[0087] In this implementation, the supplementary information is obtained based on the session information of the first user and the user information of the second user. The supplementary information is obtained by using other methods other than the domain generative model. For information that takes a long time to process or generates inaccurate information using the domain generative model, the supplementary information is directly used to correct and supplement, such as the processing of time information. By combining the information generated using the domain generative model with the supplementary information obtained by other methods, and then performing semantic parsing on the generated information, structured data responsive to the session information of the first user is generated according to the semantic parsing result, thereby further shortening the processing time for obtaining structured data and improving the service efficiency of service personnel.

[0088] Further, in step S204, based on the session information of the first user and the user information of the second user, prompt information is generated and input into the domain generative model, including: based on the session information of the first user, the user information of the second user and the backfill information, prompt information is generated and input into the domain generative model. The prompt information includes the first information related to the backfill information or does not include the second information related to the backfill information.

[0089] Optionally, the first information is complement information.

[0090] Let's take an example. Assume that the first user's session information includes that the customer agreed that the next communication time is next Tuesday night. The domain generative model may not be able to determine the specific time of the next communication because it cannot obtain the current time. The supplementary information is that the customer agreed that the next communication time is 8 pm on August 3. In one example, the first information is the supplementary information, that is, the prompt information includes "the customer agreed that the next communication time is 8 pm on August 3." The determined time information is added to the prompt information through the supplementary information. Therefore, using the domain generative model, it can be obtained that the customer agreed that the next communication time is 8 pm on August 3. In another example, the second information is "the customer agreed that the next communication time is next Tuesday night", that is, the prompt information does not include "the customer agreed that the next communication time is next Tuesday night". The domain generative model does not need to generate and process the second information. Assume that the structured data in response to the first user's session information is the subsequent Figure 5 The customer note form shown includes the field "appointment time". Then, the field "appointment time" can be directly filled in according to the replenishment information, that is, "August 3, 8 o'clock in the evening".

[0091] It can be seen that after obtaining the backfill information, one implementation method is to generate prompt information based on the backfill information, and the prompt information includes the first information related to the backfill information. Since the backfill information supplements the information that is inaccurately processed or takes a long time to process by the domain generative model, that is, the prompt information supplements the information that is inaccurately processed or takes a long time to process by the domain generative model, the processing time of using the domain generative model can be reduced through the first information for the information that was originally processed inaccurately or takes a long time to process using the domain generative model. Another implementation method is to generate prompt information based on the backfill information, and the prompt information does not include the second information related to the backfill information. For information that was originally processed inaccurately or takes a long time to process using the domain generative model, the domain generative model is not used for processing, but the backfill information is used to directly fill in and correct the structured data of the session information obtained later in response to the first user. Since the information that is inaccurately processed or takes a long time to process using the domain generative model is avoided, the processing time is also reduced.

[0092] Optionally, in an implementation, obtaining the replenishment information according to the session information of the first user and the user information of the second user includes:

[0093] Processing the session information of the first user and the user information of the second user by using a natural language processing (NLP) method to obtain information to be processed;

[0094] Match and process according to the information to be processed and the preset thesaurus to obtain the standard information corresponding to the information to be processed as the backfill information; wherein, the preset thesaurus includes the corresponding relationship between replacement words and standard words.

[0095] Optionally, the NLP methods include but are not limited to word segmentation, part-of-speech tagging, semantic role labeling, sentiment analysis, keyword extraction, etc.

[0096] Optionally, the preset thesaurus includes at least one of the following: a thesaurus of synonyms, a thesaurus of misspelled words, a thesaurus of time words, and a thesaurus of regular words.

[0097] The thesaurus of synonyms includes the corresponding relationship between synonyms and standard words. For example, the standard word is "attendance management", and the corresponding synonyms include "attendance, attendance system". Another example, the standard word is "A", which is the product name, and the corresponding synonyms include "A1, A2, A3, A4". Usually, when a customer says the product name, they cannot say it completely, and habitually say the abbreviation of the product. For the filling of the product name field, it can be determined through the thesaurus of synonyms. For example, perform word segmentation on the conversation information of the first user to get "A2", match A2 with the thesaurus of synonyms, and directly obtain the product name as A, without the need to use a domain generation model for generation processing, greatly shortening the processing time.

[0098] The thesaurus of misspelled words includes the corresponding relationship between misspelled words and standard words. For example, the standard word is "bill", and the corresponding misspelled word includes "帐单".

[0099] The thesaurus of time words includes the corresponding relationship between words describing time and standard words, and the standard word can be a time calculation rule. For example. The word describing time is "afternoon", and the corresponding standard word is "14:00". The word describing time is "tomorrow", and the corresponding standard word is "+24 hours". Suppose the customer's statement is "I have some time tomorrow afternoon, you can contact me again". By performing word segmentation on the text data, the information to be processed "tomorrow" and "afternoon" are obtained. According to the thesaurus of time words, the standard word corresponding to "tomorrow" is "+24 hours", which is a time calculation rule. Suppose the current time is 17:30:00 on June 3, 2024. According to the time calculation rule, the time "17:30:00 on June 4, 2024" is obtained, and tomorrow is June 4, 2024. The standard word corresponding to "afternoon" is "14:00". Then, through the matching process of the thesaurus of time words, the next appointment time between the service staff and the customer can be obtained as "14:00 on June 4, 2024", without the need to use a domain generation model for generation processing, greatly shortening the processing time.

[0100] The rule word library includes the correspondence between the words describing the rules and the standard words. For example, the words describing the rules include "every day, daily, by day, day dimension, which day", and the corresponding standard word is "daily dimension".

[0101] It should be noted that in actual use, a preset vocabulary library can be designed according to dimensions such as customer field and product field, and the above example does not limit the preset vocabulary library.

[0102] Optionally, in one implementation, in step S206, generating structured data in response to the session information of the first user according to the semantic parsing result includes:

[0103] Determine the target template according to the semantic analysis result, where the target template is a preset template or a template generated by dynamically calling a component;

[0104] Structured data responsive to the session information of the first user is generated according to the semantic parsing result and the target template.

[0105] Specifically, multiple templates may be pre-set, and the templates may include multiple components, such as fields. According to the semantic analysis result, the preset template may be called, and structured data may be obtained by filling the template with content. Since the preset template format is fixed, the implementation method is simple. Alternatively, according to the semantic analysis result, a template may be generated by dynamically calling a component, and structured data may be obtained by filling the generated template with content. Since the method of calling a component to generate a template is more flexible, the matching degree between the generated structured data and the session information of the first user is improved.

[0106] Reference Figure 3 , shows another flow chart of the intelligent assistance method provided in the embodiment of the present application. The execution subject of this embodiment can be Figure 1 The cloud service end 102 is shown.

[0107] like Figure 3 As shown, the intelligent assistance method provided in this embodiment may include:

[0108] Step S302: Determine a domain generative model corresponding to the session information according to the session information of the first user. Also, obtain user information of the second user according to the session information of the first user, wherein the user information of the second user at least includes: attribute information of the second user and / or historical session information of the second user.

[0109] See also Figure 2 The description of step S202 in the illustrated embodiment will not be repeated here.

[0110] Step S304: Obtain a service assistance request from the first user, wherein the service assistance request is used to instruct generation of a recommendation phrase for the first user, or to instruct generation of key content information based on the session information of the first user and the user information of the second user.

[0111] The service assistance request is sent by the user device. In the service scenario, the user device can display the relevant interface to the service personnel, and the service personnel can operate or make voice input on the relevant interface. The user device generates a service assistance request according to the user operation or the user's voice input.

[0112] In one implementation, the conversation information of the first user is conversation information corresponding to a real-time conversation between a service person and a customer, and the service assistance request is used to instruct generation of a recommended speech for the first user.

[0113] In this implementation, service personnel hope to get recommended speech during the conversation with customers, so as to better complete the communication with customers through prompts, improve service efficiency and customer service satisfaction. For example, if the service personnel are inexperienced or have little experience, the recommended speech can guide the personnel to communicate better with customers. It can be understood that since the service personnel and customers are talking in real time, this scenario has certain requirements for processing latency.

[0114] For example, Figure 4A A schematic diagram of a display interface displayed by a user device provided in an embodiment of the present application. Assume that a service staff is talking to a customer. Figure 4A As shown, the user device displays a display interface 30, which includes a dialogue window 31 and a speech prompt window 32. In the dialogue window 31, the conversation content between the service staff and the customer can be displayed in real time. In the speech prompt window 32, the recommended speech pushed by the service staff in real time can be displayed, such as "How is your trial experience", "There is a discount now", "Issuing a coupon of 180 yuan".

[0115] For example, Figure 4B Another schematic diagram of a display interface displayed by a user device provided in an embodiment of the present application. Figure 4BIn the display interface 30, the user information 36 of the second user is included. The user information 36 of the second user includes the user portrait data of the customer, such as "XX position in HH company". The user information 36 of the second user also includes the historical conversation information of the second user. For example, the historical conversation information includes key content information, that is, the customer notes generated based on the historical call data. For example, the customer notes include multiple fields, and the multiple fields include: service personnel, communication methods, communication time, product usage, intended functions, recommended functions, risk warnings, next actions, etc. For example, the content of product usage is "XX function abnormality". This information is displayed in the display interface 30, so that the service personnel can timely understand the historical communication content with the customer, which is conducive to the service personnel to adjust the communication with the customer in a timely and accurate manner and improve the service quality.

[0116] In another implementation, the conversation information of the first user includes the conversation information corresponding to the conversation between the service personnel and the customer after the conversation ends, or the voice data (also called audio data) related to the customer input by the service personnel through the user device. The service assistance request is used to instruct to generate key content information based on the conversation information of the first user and the user information of the second user.

[0117] In this implementation, by generating key content information, it is convenient to timely record the customer service process, record the customer's product needs and feedback, etc., so as to provide more accurate and timely information to the customer or other service personnel who serve the customer.

[0118] Optionally, the key content information may be stored in a preset format, such as a json format. A variety of templates may be preset, each template including a plurality of fields, and the key content information includes the content of each field in the preset template.

[0119] Optionally, the preset templates include but are not limited to: customer note template, sales note template, product note template. Different templates focus on different information. For example, the customer note template focuses on customer information, including customer name, customer size, customer product demand, etc. The product note template focuses on product information, including product name, product sales, product profit analysis, product problems, product improvement needs, etc. This embodiment does not limit the number of preset templates, the name of the preset template, the number and name of the fields in the preset template, etc.

[0120] For example, Figure 5 A schematic diagram of a preset template provided in an embodiment of the present application shows a form of a customer note template displayed on a user device. Figure 5As shown, the customer note template includes: template type, customer, contact person, preferred product, communication record, appointment time and appointment method. Some fields also have a drop-down menu control 41. For example, for "preferred product", the drop-down menu may include multiple saleable products, so that the service staff can manually fill in and modify the content in the template. Figure 5 In the example shown, after visiting XX Company, service personnel Xiao Wang inputs audio data about the visit of the customer (XX Company) through the application (APP) installed on the mobile phone. Xiao Wang's service assistance request is to generate key content information based on the audio data. After the key content information is generated on the cloud service side, it is sent to the user device, so that the user device can display the key content information as shown in the figure. Figure 5 The customer note template shown.

[0121] Step S306: Generate prompt information based on the session information of the first user, the user information of the second user and the service assistance request and input it into the domain generative model, so as to generate and process the prompt information through the domain generative model to obtain structured data in response to the session information of the first user.

[0122] Among them, the domain generation model can be found in Figure 2 Description of the illustrated embodiment.

[0123] Since the service assistance request is considered in addition to the session information of the first user and the user information of the second user when generating the prompt information, that is, different service scenarios are considered, the accuracy of generating the prompt information is improved, and thus the accuracy of the output result obtained by using the domain generative model is also improved.

[0124] The intelligent assistance method provided in this embodiment is suitable for the scenario of generating a recommendation speech for the first user, or the scenario of generating key content information according to the session information of the first user and the user information of the second user. Specifically, the domain generative model corresponding to the session information is determined according to the session information of the first user, wherein the domain generative model can be a pre-trained model corresponding to the session information of the first user, and the session information of the first user is more real-time. The domain generative model processes the session information of the first user in a targeted manner, which can improve the accuracy of model processing and shorten the processing time. In addition, the user information of the second user is obtained according to the session information of the first user, and the user information of the second user at least includes the attribute information of the second user and / or the historical session information of the second user. The service assistance request of the first user is also obtained. Based on the session information of the first user, the user information of the second user and the service assistance request, prompt information is generated and input into the domain generative model, so that the prompt information is generated and processed by the domain generative model to obtain structured data in response to the session information of the first user. The solution provided in the embodiment of the present application not only ensures the accuracy of the model output results, but also shortens the model processing time, thereby improving service efficiency in scenarios such as customer service.

[0125] Optionally, in step S306, the prompt information is generated and processed by a domain generation model to obtain structured data in response to the session information of the first user, including:

[0126] Generate and process the prompt information through the domain generation model to obtain corresponding generated information;

[0127] The generated information is semantically parsed, and structured data responsive to the session information of the first user is generated according to the semantic parsing result.

[0128] See also Figure 2 The description of step S206 in the illustrated embodiment will not be repeated here.

[0129] Optionally, in one implementation, the intelligent assistance method provided in this embodiment further includes:

[0130] Receive target text information sent by a user device and selected by a first user according to session information of the first user.

[0131] Correspondingly, in step S306, prompt information is generated based on the session information of the first user, the user information of the second user and the service assistance request, including:

[0132] Prompt information is generated according to the service assistance request, the session information of the first user, the user information of the second user and the target text information.

[0133] Combine the following Figure 4ALet's take an example. Assume that a service person is talking to a customer, and the service assistance request is used to instruct the service person to generate a recommended speech. Figure 4A In the example, the user device displays a display interface 30, which includes a dialogue window 31, a speech prompt window 32 and a keyword window 33. In the dialogue window 31, the conversation content between the service personnel and the customer can be displayed in real time. Assume that the service personnel operates in the dialogue window 31 and selects the target text information as "YYYYY" 34. The user device responds to the user operation and displays the target text information "YYYYY" in the keyword window 33. The user device sends the target text information "YYYYY" to the cloud service end, and correspondingly, the cloud service end receives the target text information "YYYYY", and can generate prompt information based on the target text information. Among them, if the user device is a mobile phone or the like with a touch display screen, the operation performed by the service personnel in the dialogue window 31 can be a touch operation, such as a long press operation; if the user device is a computer or the like with a normal display screen, the operation performed by the service personnel in the dialogue window 31 can be an operation performed by a mouse.

[0134] In this implementation, by obtaining the target text information selected by the service personnel in the text data of the conversation, the target text information selected by the service personnel is further considered when generating prompt information, so that the generated prompt information is more suitable for the customer. The subsequent recommended words output by the domain generative model based on the prompt information are more suitable for the customer, thereby improving the service quality and improving the customer's service experience.

[0135] Below, refer to Figure 6A to Figure 6C , the process of the intelligent assistance method provided in this application is illustrated in the form of an example.

[0136] like Fig. 6A As shown, the system includes a user device and a cloud server.

[0137] An APP for implementing CRM is installed on the user device. The service personnel (i.e., the first user) uses the functions provided by the APP to initiate a call with the customer (i.e., the second user). During the call between the service personnel and the customer, or after the service personnel ends the call with the customer, the user device obtains the customer-related session information conducted by the service personnel and sends the session information to the cloud service end. The user device also displays an interface, which includes controls that can be operated by the service personnel. This example does not limit the layout of the interface, the number, location, and function of the controls in the interface. For example, the controls include a generate note control 520 and a speech recommendation control 521. In one scenario, during the call between the service personnel and the customer, the service personnel can click on the speech recommendation control 521. Accordingly, the user device generates a service assistance request in response to the service personnel clicking on the speech recommendation control 521, and the service assistance request is used to indicate the generation of recommended speech for the service personnel. In this scenario, the service personnel requests to obtain recommended speech skills, so as to obtain timely and accurate speech skills guidance when communicating with customers, which can improve the success rate of successfully serving customers, for example, improve the success rate of successfully establishing contact with customers so as to subsequently maintain the customers, or improve the success rate of successfully selling products to customers. For example, in another scenario, after the service personnel finishes the call with the customer, the service personnel can click on the generate note control 520. Accordingly, the user device generates a service assistance request in response to the service personnel clicking on the generate note control 520. The service assistance request is used to instruct the generation of key content information about the session information and the customer, that is, to generate a note for subsequent maintenance of the customer. The user device obtains the service assistance request of the service personnel and sends the service assistance request to the cloud service end.

[0138] The cloud service end receives the session information and service assistance request sent by the user device, and also obtains the user information of the client from the external knowledge base. The cloud service end generates prompt information based on the session information, the service assistance request and the user information of the client, and uses the domain generative model to generate and process the prompt information to obtain corresponding generated information. The generated information is semantically parsed, and structured data in response to the session information of the first user is generated according to the semantic parsing result, and the structured data is sent to the user device.

[0139] After receiving the structured data sent by the cloud server, the user device displays the structured data. For example, when the service assistance request is used to instruct the service personnel to generate recommended words, the structured data includes the recommended words. The display effect can be seen in Figure 4A ; When the service assistance request is used to instruct the generation of key content information about the session information and the customer, the structured data includes the content of each field in the note template. The display effect can be seen in Figure 5 .

[0140] Figure 6BThe following shows the effect that CRM can achieve by using the intelligent assistance method provided in the embodiment of the present application. Figure 6B As shown, the text content of the call can be displayed in real time based on the real-time voice and text stream between the service personnel and the customer, so that the service personnel can view it in time. Knowledge recommendations can be obtained from the external knowledge base, and the knowledge recommendations include but are not limited to: industry solutions, product documents, market information, etc. Talking skills recommendations can be obtained so that service personnel can be guided to improve service efficiency and service quality. The intention of customer service can be identified based on the content of the call between the service personnel and the customer. The identified customer intentions include but are not limited to: discounts, query information, ordering, etc., and then, corresponding operations can be performed according to the customer intentions, such as issuing coupons, order push, etc. The customer's emotions can be identified based on the content of the call between the service personnel and the customer. The identified customer emotions include but are not limited to: angry, busy, satisfied, etc., and then, risk warnings or customer reminders can be performed based on the customer's emotions. The call quality score can be obtained, and the scoring criteria include but are not limited to: consistency of speech skills, emotional control, etc. The call recording recommendation between the service personnel and the customer can be obtained so that the customer can continue to be served in the future.

[0141] Figure 6C A processing process using the intelligent assistance method provided in an embodiment of the present application is shown.

[0142] Step 501: The user device provides a CRM management platform, and the service personnel establish a connection with the customer through audio and video cloud communication via the CRM management platform provided by the user device.

[0143] Step 502: The user device collects audio of the audio and video call between the service personnel and the customer, separates the audio channels, obtains the service personnel's audio 1 and the customer's audio 2, and sends them to the cloud service end.

[0144] Step 503: The cloud service end obtains the service personnel's audio 1 to obtain a real-time audio stream.

[0145] Step 504: The cloud server obtains the client's audio 2 and obtains a real-time audio stream.

[0146] Step 505: The cloud service performs real-time voice-to-text operations on the service personnel's audio 1 and the customer's audio 2 to obtain text data.

[0147] Step 506: The cloud server performs block text separation and filtering on the text data to obtain segmented text.

[0148] Block text separation is the process of breaking down large blocks of text into smaller, more manageable segments. By splitting text into blocks of preset sizes, the accuracy of content recalled from the vector database can be optimized, and the domain generative model can more effectively analyze information within its context window, which helps improve the efficiency and accuracy of domain generative model-based processing.

[0149] Step 507: After the service personnel hangs up the audio and video call with the customer, the user device generates a complete audio file, that is, an offline audio file, and sends the offline audio file to the cloud service end.

[0150] Among them, step 507 and step 514 can be applied to the scenario where the service personnel requests to generate key content information about the session information and the customer, that is, requests to generate a note.

[0151] Step 508: The cloud server generates prompt information for inputting the generative basic model through prompt engineering.

[0152] For domain generative models, if the prompts are different, the content generated by the domain generative model will also be different. In order to improve the accuracy of the results generated by the domain generative model, it is necessary to construct accurate prompts. Among them, prompt engineering is a method that can guide the model output results without adjusting the parameters or weights of the domain generative model.

[0153] In this example, dynamic prompt word engineering technology is implemented based on diversified data and multiple templates. Diversified data includes: text data corresponding to customer-related conversation information conducted by service personnel, service assistance requests from service personnel, and user profile data of customers recalled through RAG (see Figure 5 509), product domain knowledge (i.e. product information, see Figure 5 In 510), user information of the service personnel, emotional information of the customer extracted from the text data or conversation information, and target text information selected by the service personnel based on the text data of the conversation with the customer.

[0154] Step 511: The cloud server generates corresponding generation information by calling the model based on the prompt information at the model processing layer.

[0155] Among them, the model includes a general generative model and a domain generative model. The general domain knowledge large model is the general generative model, and the domain generative model is a short text customization model. It also includes a small model customized for customer domains, a long text customization model, and a special capability model. This example does not limit the model name. For example, the short text customization model is also called a short text exclusive model, the customer domain customization small model is also called a customer domain exclusive model, the long text customization model is also called a long text exclusive model, and the special capability model is also called a preset function exclusive model.

[0156] Afterwards, semantic analysis is performed on the generated information, and structured data responsive to the session information of the first user is generated according to the semantic analysis result.

[0157] Step 512: The cloud service end obtains the replenishment information at the system engineering replenishment processing layer according to the call information and the user information of the customer, and uses the replenishment information to correct and supplement the structured data.

[0158] Among them, the example of time NLP slicing processing generation can be found in the relevant description of the time word library, and the example of product synonym word library filtering can be found in the relevant description of the synonym word library.

[0159] Step 513: When the service assistance request is used to instruct the generation of recommended words for the service personnel, the cloud service generates the recommended words.

[0160] Technical effects can be seen in Figure 4A The relevant description will not be repeated here.

[0161] Step 514: When the service assistance request is used to instruct generation of key content information about the session information and the customer, the cloud service generates a note.

[0162] Technical effects can be seen in Figure 5 The relevant description will not be repeated here.

[0163] Reference Figure 7 , shows a schematic diagram of the structure of an electronic device according to an embodiment of the present application. The specific embodiment of the present application does not limit the specific implementation of the electronic device.

[0164] like Figure 7 As shown, the electronic device may include: a processor (processor) 602 , a communication interface (Communications Interface) 604 , a memory (memory) 606 , and a communication bus 608 .

[0165] in:

[0166] The processor 602 , the communication interface 604 , and the memory 606 communicate with each other via a communication bus 608 .

[0167] The communication interface 604 is used to communicate with other electronic devices or servers.

[0168] The processor 602 is used to execute the program 610, and specifically can execute the relevant steps in any of the above method embodiments.

[0169] Specifically, the program 610 may include program codes, which include computer operation instructions.

[0170] The processor 602 may be a CPU, a GPU (Graphic Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present application. The one or more processors included in the smart device may be processors of the same type, such as one or more CPUs; or may be processors of different types, such as one or more CPUs and one or more ASICs.

[0171] The memory 606 is used to store the program 610. The memory 606 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0172] The program 610 may include multiple computer instructions. Specifically, the program 610 may enable the processor 602 to execute operations corresponding to the method described in any of the aforementioned method embodiments through the multiple computer instructions.

[0173] The specific implementation of each step in program 610 can refer to the corresponding description of the corresponding steps and units in the above method embodiment, and has corresponding beneficial effects, which will not be repeated here. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the above-described devices and modules can refer to the corresponding process description in the above method embodiment, which will not be repeated here.

[0174] The present application also provides a computer storage medium on which a computer program is stored, and when the program is executed by a processor, the method described in any of the above-mentioned multiple method embodiments is implemented. The computer storage medium includes but is not limited to: a compact disc read-only memory (CD-ROM), a random access memory (RAM), a floppy disk, a hard disk or a magneto-optical disk, etc.

[0175] An embodiment of the present application also provides a computer program product, including computer instructions, which instruct a computing device to execute operations corresponding to any one of the above-mentioned multiple method embodiments.

[0176] In addition, it should be noted that the user-related information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to request data input by the user, sample data for training the model, data for analysis, stored data, displayed data, etc.) involved in the embodiments of the present 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 need to comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0177] It should be pointed out that, according to the needs of implementation, the various components / steps described in the embodiments of the present application can be split into more components / steps, or two or more components / steps or partial operations of components / steps can be combined into new components / steps to achieve the purpose of the embodiments of the present application.

[0178] The above-mentioned method according to the embodiment of the present application can be implemented in hardware, firmware, or implemented as software or computer code that can be stored in a recording medium (such as a CD-ROM, RAM, floppy disk, hard disk or magneto-optical disk), or implemented as a computer code originally stored in a remote recording medium or a non-temporary machine-readable medium downloaded through a network and stored in a local recording medium, so that the method described herein can be stored in such software processing on a recording medium using a general-purpose computer, a dedicated processor or programmable or dedicated hardware (such as an application-specific integrated circuit (ASIC) or a field programmable gate array (FPGA)). It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component (e.g., a random access memory (RAM), a read-only memory (ROM), a flash memory, etc.) that can store or receive software or computer code, and when the software or computer code is accessed and executed by a computer, a processor or hardware, the method described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the method shown here, the execution of the code converts the general-purpose computer into a dedicated computer for executing the method shown here.

[0179] Those of ordinary skill in the art will appreciate that the units and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for specific applications, but such implementation should not be considered to be beyond the scope of the embodiments of the present application.

[0180] The above implementation methods are only used to illustrate the embodiments of the present application, and are not limitations on the embodiments of the present application. Ordinary technicians in the relevant technical field can make various changes and modifications without departing from the spirit and scope of the embodiments of the present application. Therefore, all equivalent technical solutions also belong to the scope of the embodiments of the present application. The scope of patent protection of the embodiments of the present application should be limited by the claims.

Claims

1. An intelligent assistance method, comprising: Determining, according to the session information of the first user, a domain generative model corresponding to the session information; Furthermore, according to the session information of the first user, user information of the second user is obtained, where the user information of the second user at least includes: attribute information of the second user and / or historical session information of the second user; Based on the session information of the first user and the user information of the second user, generating prompt information and inputting the prompt information into the domain generative model, so as to generate and process the prompt information through the domain generative model to obtain corresponding generated information; The generated information is semantically parsed, and structured data responsive to the session information of the first user is generated according to the semantic parsing result.

2. The method according to claim 1, wherein: The method further comprises: The user information of the second user is displayed on the display interface of the session information of the first user.

3. The method according to claim 1 or 2, wherein: The attribute information of the second user includes at least: user portrait data of the second user; The historical conversation information of the second user at least includes: key content information generated according to the historical conversation information of the second user, and / or historical conversation full text information.

4. The method according to claim 1 or 2, wherein: The method further includes: obtaining corresponding domain knowledge according to the session information of the first user; The generating prompt information based on the session information of the first user and the user information of the second user and inputting it into the domain generative model includes: generating prompt information based on the session information of the first user, the user information of the second user, and the domain knowledge and inputting it into the domain generative model.

5. The method according to claim 1 or 2, wherein: The method further comprises: obtaining user information of the first user, the user information of the first user comprising at least: service level information of the first user, and / or idiom information of the first user; The generating prompt information based on the session information of the first user and the user information of the second user and inputting the prompt information into the domain generative model includes: generating prompt information based on the session information of the first user, the user information of the second user, and the user information of the first user and inputting the prompt information into the domain generative model.

6. The method according to claim 1 or 2, wherein: The generating prompt information based on the session information of the first user and the user information of the second user and inputting the prompt information into the domain generative model includes: According to a preset metadata template, the session information of the first user and the user information of the second user are filled into the metadata template to generate prompt information, and the prompt information is input into the domain generation model.

7. The method according to claim 1 or 2, wherein: The determining, according to the session information of the first user, a domain generative model corresponding to the session information includes: Determining, based on the session information of the first user, a degree of openness of the session information; If the degree of openness is less than or equal to a first threshold, determining to use a domain generative model corresponding to the session information; If the degree of openness is greater than the first threshold, it is determined to use a general generative model.

8. The method according to claim 7, wherein: The method further comprises: The backfill information is obtained according to the session information of the first user and the user information of the second user, so as to modify and supplement the structured data using the backfill information.

9. The method according to claim 8, wherein: The obtaining the replenishment information according to the session information of the first user and the user information of the second user includes: The session information of the first user and the user information of the second user are processed by natural language processing (NLP) to obtain information to be processed; Matching processing is performed based on the information to be processed and a preset word library to obtain standard information corresponding to the information to be processed as the complement information; wherein the preset word library includes a correspondence between replacement words and standard words.

10. The method according to claim 9, wherein: The preset word library includes at least one of the following: a synonym word library, an error word library, a time word library and a rule word library.

11. The method according to claim 1 or 2, wherein: The generating structured data in response to the session information of the first user according to the semantic parsing result includes: Determining a target template according to the semantic analysis result, wherein the target template is a preset template or a template generated by dynamically calling a component; Structured data responsive to the session information of the first user is generated according to the semantic parsing result and the target template.

12. An intelligent assistance method, comprising: Determining, according to the session information of the first user, a domain generative model corresponding to the session information; Furthermore, according to the session information of the first user, user information of the second user is obtained, where the user information of the second user at least includes: attribute information of the second user and / or historical session information of the second user; Obtaining a service assistance request of the first user; the service assistance request is used to instruct generation of a recommendation phrase for the first user, or to instruct generation of key content information based on session information of the first user and user information of the second user; Based on the session information of the first user, the user information of the second user and the service assistance request, prompt information is generated and input into the domain generative model, so that the prompt information is generated and processed by the domain generative model to obtain structured data in response to the session information of the first user.

13. An electronic device comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform an operation corresponding to the method according to any one of claims 1 to 12.

14. A computer storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the method according to any one of claims 1 to 12 is implemented.

15. A computer program product, comprising computer instructions, wherein the computer instructions instruct a computing device to execute operations corresponding to the method according to any one of claims 1 to 12.

Images