Method, system and device for recommending insurance sales task scripts based on contextual semantic understanding

Abstract

A method for recommending insurance sales task scripts based on contextual semantic understanding includes the following steps: providing an insurance sales task script recommendation server to obtain the conversation context information between the current insurance agent and the user in real time; inputting the obtained current conversation context information into a script recommendation algorithm model; the script recommendation algorithm model executes the script recommendation algorithm to process the input conversation context information; the script recommendation algorithm model outputs sales prompt terms; providing an insurance sales task script recommendation agent terminal to feed back the sales prompt terms to the insurance agent at the insurance sales task script recommendation agent terminal in real time, and the insurance agent tells the insurance sales task content information according to the prompt; the insurance sales task script recommendation server obtains user feedback information; and judging whether the current sales task has been achieved. If not, the conversation context information is obtained and updated, and the script recommendation algorithm model is re-input until the sales task is achieved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a method, system, and device for recommending speech techniques for insurance sales tasks based on contextual semantic understanding. Background Art

[0002] In recent years, artificial intelligence has seen rapid growth in data, algorithms, and computing power, ushering in a new wave of development against the backdrop of the global digital transformation of the economy. This wave of AI has a far greater impact than any previous one, most notably by expanding its influence beyond specialized fields into the general public.

[0003] Artificial intelligence has also seen unprecedented growth in the financial sector, particularly in enhancing the insurance user experience and personalizing insurance services. In this regard, AI will enable fully automated and seamless integration with the policy user experience. For example, chatbots can access a customer's geographic and social data for personalized interactions. Insurance companies are also making a series of innovations in personalized services and user experience, such as allowing users to customize specific insurance targets and coverage (i.e., on-demand insurance).

[0004] In terms of personalized service and improving customer experience, AI technology is optimizing sales processes, standardizing sales procedures, and organizing and recommending insurance terminology and sales language. Recommended sales language is particularly important, helping salespeople introduce insurance products, recommend coverage plans, and sell them at the right time and in the right way. This streamlines communication and ensures that the communication process is more aligned with financial industry regulatory requirements, preventing the expansion of insurance product coverage and weakening of policyholders' claim limits. It also effectively helps salespeople avoid awkward silences, enabling them to communicate smoothly with diverse customer profiles and close sales.

[0005] However, although insurance companies currently require new sales personnel, namely insurance agents, to have professional qualification certificates issued by the China Insurance Regulatory Commission, have a certain understanding of basic insurance knowledge, have undergone long-term sales training in the company, have accumulated a large number of cases and experiences in the process of communicating with customers, and have also compiled a large number of professional dialogues, in the face of the current rapid development of the financial industry, the update and iteration cycle of insurance products is getting shorter and shorter, new products are emerging in an endless stream, and regulatory requirements are becoming more and more detailed. Faced with a complex working environment, basic business development training has become inadequate and cannot cope with the changing and complex sales scenarios. Summary of the Invention

[0006] One of the purposes of the present invention is to provide a method, system and device for recommending insurance sales task scripts based on contextual semantic understanding, which can be used to train response language scripts for various complex scenarios. According to the context of the current conversation, the sales task of the current conversation, the personality and emotional analysis of the current user, and the sales feature matching of the current salesperson, it recommends prompt scripts to help salespeople complete sales tasks quickly and effectively.

[0007] In order to achieve at least one of the inventive objectives of the present invention, the present invention provides a method for recommending insurance sales task scripts based on contextual semantic understanding, the method comprising the following steps:

[0008] Provides a service for recommending scripts for insurance sales tasks, and obtains real-time conversation context information between insurance agents and users.

[0009] Input the acquired current conversation context information into the speech recommendation algorithm model;

[0010] The speech recommendation algorithm model executes the speech recommendation algorithm and processes the input conversation context information;

[0011] The sales pitch recommendation algorithm model outputs sales tips;

[0012] Provide insurance sales task script recommendation agent side, and provide real-time feedback of sales prompts to the insurance agents on the insurance sales task script recommendation agent side. The insurance agents will then describe the content of the insurance sales task according to the prompts.

[0013] The insurance sales task script recommendation server obtains user feedback information; and

[0014] The insurance sales task script recommendation server determines whether the current sales task has been completed. If not, it obtains and updates the conversation context information and re-enters the script recommendation algorithm model until the sales task is completed.

[0015] In some embodiments, the method for recommending insurance sales task scripts based on contextual semantic understanding also includes the following steps: classifying the current conversation context information to obtain a classification list of all user information in the entire context; recommending language templates for the classification list of the conversation process; and filling in information for the recommended templates to obtain a text of recommended scripts to prompt sales staff, and outputting sales prompt terms.

[0016] In some embodiments, the method for recommending speech scripts for insurance sales tasks based on contextual semantic understanding also includes a speech recommendation algorithm model training step: input data preparation and input format, wherein the input data includes current context data, text transcribed by ASR of the conversation process voice data, emotional identifiers extracted from the voice data, sales process steps of insurance products, customer information, auxiliary processes and step information; the input data is manually labeled as labeled classification data before being input, and is sequentially spliced ​​into a string of length not exceeding 1024, and training input and recommendation input are executed; and classification training is performed on the input text through the VDCNN algorithm, and specific scenario classification results are returned through the input to execute insurance product sales steps or insurance sales auxiliary steps.

[0017] In some embodiments, the method for recommending insurance sales task scripts based on contextual semantic understanding also includes the following steps: obtaining user information and classification data of the conversation text context; and performing recommendations based on the classification data of the conversation context using a user-based collaborative filtering algorithm, and providing real-time feedback of sales prompts to the insurance agent at the insurance sales task script recommendation agent end.

[0018] In some embodiments, the method for recommending insurance sales task speech based on contextual semantic understanding also includes the following steps: executing the filling of the language template, wherein an entity recognition algorithm is used to extract entities from the user conversation context, and the language template is filled by the user information, policy information, insurance product information, extracted entity data and other auxiliary information in the language template; obtaining the language text that needs to be prompted to the insurance salesperson, and feeding back the speech recommendation information under the sales task to the insurance agent at the insurance sales task speech recommendation agent end.

[0019] According to another aspect of the present invention, the present invention also provides an insurance sales task speech recommendation system based on contextual semantic understanding, the insurance sales task speech recommendation system based on contextual semantic understanding includes an insurance sales task speech recommendation service subsystem and an insurance sales task speech recommendation agent subsystem; the insurance sales task speech recommendation service subsystem is configured from the insurance sales task speech recommendation agent subsystem to: obtain the conversation context information between the current insurance agent and the user in real time; input the obtained current conversation context information into the speech recommendation algorithm model; the speech recommendation algorithm model executes the speech recommendation algorithm and processes the input conversation context information; the speech recommendation algorithm model outputs sales prompt terms; and provides real-time feedback The insurance sales task speech recommendation agent subsystem is provided with a human-computer interaction unit to obtain the sales prompt phrases fed back by the insurance sales task speech recommendation service subsystem in real time; the insurance sales task speech recommendation agent subsystem is further configured to send user feedback information to the insurance sales task speech recommendation service subsystem; the insurance sales task speech recommendation service subsystem also includes a sales task judgment unit, and the sales task judgment unit is configured to judge whether the current sales task has been achieved. If not, the conversation context information is obtained and updated, and the speech recommendation algorithm model is re-entered until the sales task is achieved.

[0020] In some embodiments, the insurance sales task speech recommendation service subsystem also includes a speech recommendation algorithm model unit, the speech recommendation algorithm model of the speech recommendation algorithm model unit executes the speech recommendation algorithm, processes the input dialogue context information, and outputs sales prompts; the speech recommendation algorithm model unit is also configured to: perform classification on the current dialogue context information, obtain a classification list of all user information in the entire context; perform language template recommendation for the classification list of the dialogue process; perform information filling on the recommended template, obtain the recommended speech text to the salesperson, and output sales prompts; wherein, the speech recommendation algorithm model unit is also configured to: prepare input data and input format, wherein the input data includes the current context data, the text of the ASR transcription of the dialogue process voice data, the emotion identifier extracted from the voice data, the sales process steps of the insurance product, customer information, auxiliary process and step information, the input data is manually marked as labeled classification data before being input, and is spliced ​​into a character string of no more than 1024 in length in sequence, and training input and recommendation input are performed.

[0021] In some embodiments, the speech recommendation algorithm model unit is further configured to: perform classification training on the input text through the VDCNN algorithm, return specific scenario classification results through input, and execute insurance product sales steps or insurance sales auxiliary steps; wherein, the speech recommendation algorithm model unit is further configured to: obtain user information and classification data of the conversation text context; based on the user's collaborative filtering algorithm, perform recommendations according to the classification data of the conversation context, and provide real-time feedback of sales tips to the insurance agent of the insurance sales task speech recommendation agent subsystem.

[0022] In some embodiments, the insurance sales task speech recommendation service subsystem also includes a language template filling module, which performs the filling of the language template, wherein an entity recognition algorithm is used to extract entities from the user conversation context, and the language template is filled by the user information, policy information, insurance product information, extracted entity data and other auxiliary information in the language template; the language text that needs to be prompted to the insurance salesperson is obtained, and the speech recommendation information under the sales task is fed back to the insurance agent of the insurance sales task speech recommendation agent subsystem.

[0023] According to another aspect of the present invention, the present invention further provides a device for recommending insurance sales task scripts based on contextual semantic understanding, comprising:

[0024] Memory, for storing software applications,

[0025] A processor is used to execute the software application, and each program of the software application correspondingly executes the steps in the method for recommending insurance sales task speech based on contextual semantic understanding. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 The present invention is a flowchart of the steps of a method for recommending insurance sales task scripts based on contextual semantic understanding according to an embodiment of the present invention.

[0027] Figure 2 This is a flowchart of the steps of the method for recommending insurance sales task scripts based on contextual semantic understanding according to the above embodiment of the present invention.

[0028] Figure 3 It is a schematic diagram of the algorithm structure of the insurance sales task speech recommendation method based on contextual semantic understanding according to the above embodiment of the present invention.

[0029] Figure 4 This is a flowchart of the steps of the method for recommending insurance sales task scripts based on contextual semantic understanding according to the above embodiment of the present invention. DETAILED DESCRIPTION

[0030] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are for illustrative purposes only, and those skilled in the art will readily appreciate other obvious variations. The basic principles of the present invention defined in the following description may be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the present invention.

[0031] It is to be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the elements may be multiple, and the term "one" should not be understood as a limitation on the quantity.

[0032] The present invention is an invention related to a computer program. Figure 1 The flowchart of the method for recommending insurance sales task speech based on contextual semantic understanding according to the present invention is shown, which describes a solution for solving the problem proposed by the present invention, based on the computer program processing flow, by executing a computer program compiled according to the above flow to control or process the external object or internal object of the computer. Through the method for recommending insurance sales task speech based on contextual semantic understanding of the present invention, it is possible to utilize a computer system, integrate manual experience and machine learning results, and train response language speech for various complex scenarios. According to the context of the current conversation, the sales task of the current conversation, the personality and emotion analysis of the current user, and the sales feature matching of the current salesperson, it is possible to recommend prompt speech to help salespeople in the insurance industry complete sales tasks quickly and effectively. It is understandable that the "computer" referred to in the present invention not only refers to desktop computers, laptops, tablets and other devices, but also includes other intelligent electronic devices that can run according to programs and process data.

[0033] Specifically, the method for recommending insurance sales task scripts based on contextual semantic understanding includes the following steps:

[0034] Provides a service for recommending scripts for insurance sales tasks, and obtains real-time conversation context information between insurance agents and users.

[0035] Input the acquired current conversation context information into the speech recommendation algorithm model;

[0036] The speech recommendation algorithm model executes the speech recommendation algorithm and processes the input conversation context information;

[0037] The sales pitch recommendation algorithm model outputs sales tips;

[0038] Provide insurance sales task script recommendation agent side, and provide real-time feedback of sales prompts to the insurance agents on the insurance sales task script recommendation agent side. The insurance agents will then describe the content of the insurance sales task according to the prompts.

[0039] The insurance sales task script recommendation server obtains user feedback information;

[0040] The insurance sales task script recommendation server determines whether the current sales task has been completed. If not, it obtains and updates the conversation context information and re-enters the script recommendation algorithm model until the sales task is completed.

[0041] In a specific embodiment, when insurance sales personnel are completing sales tasks and talking with users, the conversation information can be input into a speech recommendation algorithm model. The model uses deep learning and machine learning algorithms, pre-labels a large amount of data and then trains it into a general model, which is then used to predict the language text of the next reply in the conversation process.

[0042] More specifically, executing the speech recommendation algorithm in the speech recommendation algorithm model further includes the following steps:

[0043] Perform classification on the current conversation context information to obtain a classified list of all user information in the entire context;

[0044] Recommendation of language templates based on the categorized list of conversation processes;

[0045] Fill in information for the recommended template, obtain the recommendation text for the salesperson, and output the sales prompt wording.

[0046] More specifically, the example diagram of the text classification algorithm in the speech recommendation algorithm model is as follows Figure 2 The method for recommending speech techniques for insurance sales tasks based on contextual semantic understanding also includes the following steps:

[0047] Input data preparation and input format. The input data includes current context data, ASR-transcribed text from conversational speech data, emotion tags extracted from speech data, insurance product sales process steps, customer information, and auxiliary process and step information. The input data is manually annotated as labeled categorized data before input and sequentially concatenated into strings no longer than 1024 characters for training and recommendation input.

[0048] as well as

[0049] The VDCNN algorithm is used to perform classification training on the input text, and specific scene classification results are returned through the input to execute insurance product sales steps or insurance sales auxiliary steps.

[0050] In a specific embodiment, Figure 3 The model architecture diagram shown in the figure shows that the classification algorithm uses the VDCNN algorithm. The model input of this algorithm is in characters, and the input length is fixed at s = 1024. Therefore, it solves the actual text classification prediction task and solves the return from input data to speech classification by using this algorithm.

[0051] Furthermore, if Figure 4 As shown, after classification, the recommended language template is executed. The method for recommending insurance sales task scripts based on contextual semantic understanding also includes the following steps:

[0052] Obtain classification data of user information and conversation text context;

[0053] Based on the user's collaborative filtering algorithm, recommendations are made according to the classification data of the conversation context, and real-time feedback of sales prompts is given to the insurance agents on the agent side who recommend the insurance sales task scripts.

[0054] In a specific embodiment, the recommendation algorithm is based on a user-based collaborative filtering algorithm. Based on the categorized data in the conversation context, the recommendation is made, and the most appropriate line of dialogue is provided to the salesperson for prompting. This algorithm addresses the role of recommending language templates in the actual conversation context, and solves the problem of predicting categorized data and converting it into a language template. For example, the similarity calculation formula used for calculating conversation similarity is:

[0055]

[0056] Furthermore, the method for recommending insurance sales task scripts based on contextual semantic understanding also includes the following steps: executing the filling of the language template.

[0057] Use entity recognition algorithms to extract entities from the user conversation context, and fill in the language template by using user information, policy information, insurance product information, extracted entity data, and other auxiliary information in the language template; obtain the language text that needs to be prompted to the insurance salesperson, and feedback the recommended sales script information under the sales task to the insurance agent on the insurance sales task sales script recommendation agent side.

[0058] Through the insurance sales task speech recommendation method based on contextual semantic understanding of the present invention, after using the above algorithm, it can effectively solve the problems of sales personnel's unclear conversations and ineffective communication during the sales process, and at the same time, it can also improve the fluency of communication and increase the probability of successfully signing insurance transaction contracts.

[0059] Through the insurance sales task script recommendation method based on contextual semantic understanding of the present invention, a task-based dialogue system is constructed through speech recognition technology (ASR), speech sentiment analysis technology, short text classification, entity recognition technology (NER), and knowledge graph technology. The text that prompts sales scripts is synthesized in combination with the calculation results, which greatly improves the efficiency of effective communication in the sales process and helps insurance companies improve the final success rate of signing insurance contracts. In addition, it can serve as training for insurance agents, helping them optimize their communication skills and improve their business communication capabilities.

[0060] It's understandable that in addition to using the aforementioned pre-labeled data to train an algorithmic model, multi-dimensional weighted analysis of the current conversation scenario can be used to match the corresponding sales pitch. Sales pitches can also be recommended through multi-level filtering. For example, based on the previous conversation text, the next possible response can be selected. Furthermore, multiple levels of filtering can be performed based on other dimensional parameters, ultimately selecting a single text as the next sales pitch to recommend to the insurance agent.

[0061] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided in the form of methods, systems, or computer program products. Thus, the present invention may take the form of an all-hardware embodiment, an all-software embodiment, or a combination of software and hardware embodiments.

[0062] Those skilled in the art will appreciate that the methods of the present invention can be implemented using hardware, software, or a combination of hardware and software. The present invention can be implemented in a centralized manner in at least one computer system, or in a decentralized manner by different components distributed across several interconnected computer systems. Any computer system or other device capable of implementing the methods is applicable. A conventional combination of hardware and software can be a general-purpose computer system with a computer program installed, where the computer system is controlled by the installation and execution of the program to perform the methods.

[0063] The present invention may be embedded in a computer program product, which includes all the features that enable the method described herein to be implemented. The computer program product is contained in one or more computer-readable storage media, and the computer-readable storage medium has a computer-readable program code contained therein. According to another aspect of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method of the present invention are performed. A computer storage medium is a medium in a computer memory for storing a certain discontinuous physical quantity. Computer storage media include but are not limited to semiconductors, disk memories, magnetic cores, magnetic drums, magnetic tapes, laser discs, etc. It will be understood by those skilled in the art that computer storage media are not limited to the aforementioned examples, which are merely examples and are not limited to the present invention.

[0064] According to another aspect of the present invention, a device for recommending insurance sales task scripts based on contextual semantic understanding is provided. The device includes a software application, a memory for storing the software application, and a processor for executing the software application. Each program in the software application is capable of correspondingly executing the steps of the method for recommending insurance sales task scripts based on contextual semantic understanding of the present invention.

[0065] A typical combination of hardware and software may be a general-purpose computer system with a computer program. When the program is loaded and executed, it controls the computer system so as to execute the method for recommending insurance sales task scripts based on contextual semantic understanding disclosed in the present invention.

[0066] It will be understood by those skilled in the art that the insurance sales task script recommendation device based on contextual semantic understanding can be embodied as a desktop computer, notebook, mobile smart device, etc., but the foregoing is only an example and also includes other smart devices equipped with the software application of the present invention.

[0067] Corresponding to the embodiment of the method of the present invention, according to another aspect of the present invention, an insurance sales task speech recommendation system based on contextual semantic understanding is also provided. The insurance sales task speech recommendation system based on contextual semantic understanding is the application of the insurance sales task speech recommendation method based on contextual semantic understanding of the present invention in computer program improvement.

[0068] Specifically, the insurance sales task script recommendation system based on contextual semantic understanding includes an insurance sales task script recommendation service subsystem and an insurance sales task script recommendation agent subsystem. The insurance sales task script recommendation service subsystem is configured to: obtain the current conversation context information between the insurance agent and the user in real time; input the obtained current conversation context information into a script recommendation algorithm model; the script recommendation algorithm model executes the script recommendation algorithm and processes the input conversation context information; the script recommendation algorithm model outputs sales tips; and provides real-time feedback of the sales tips to the insurance sales task script recommendation agent subsystem. The insurance sales task script recommendation agent subsystem is provided with a human-computer interaction unit to obtain the sales tips feedback from the insurance sales task script recommendation service subsystem in real time. The insurance sales task script recommendation agent subsystem is further configured to send user feedback information to the insurance sales task script recommendation service subsystem.

[0069] The insurance sales task speech recommendation service subsystem also includes a sales task judgment unit, which is configured to: judge whether the current sales task has been achieved; if not, obtain and update the conversation context information, and re-enter the speech recommendation algorithm model until the sales task is achieved.

[0070] More specifically, the insurance sales task speech recommendation service subsystem also includes a speech recommendation algorithm model unit, the speech recommendation algorithm model of the speech recommendation algorithm model unit executes the speech recommendation algorithm, processes the input dialogue context information, and outputs sales prompt terms; the speech recommendation algorithm model unit is also configured to: perform classification on the current dialogue context information, obtain a classification list of all user information in the entire context; perform language template recommendation for the classification list of the dialogue process; perform information filling on the recommended template, obtain the recommendation speech text to be prompted to the salesperson, and output sales prompt terms.

[0071] Furthermore, the speech recommendation algorithm model unit is also configured as follows: input data preparation and input format, wherein the input data includes current context data, text transcribed by ASR of the conversation process voice data, emotion identification extracted from the voice data, sales process steps of insurance products, customer information, auxiliary processes and step information; the input data is manually labeled as labeled classification data before being input, and is sequentially spliced ​​into a character string with a length not exceeding 1024, and training input and recommendation input are executed.

[0072] Furthermore, the speech recommendation algorithm model unit is also configured to: perform classification training on the input text through the VDCNN algorithm, return specific scenario classification results through input, and execute insurance product sales steps or insurance sales auxiliary steps.

[0073] Furthermore, the speech recommendation algorithm model unit is also configured to: obtain user information and classification data of the conversation text context; perform recommendations based on the user's collaborative filtering algorithm and the classification data of the conversation context, and provide real-time feedback of sales tips to the insurance agent of the insurance sales task speech recommendation agent subsystem.

[0074] Furthermore, the insurance sales task speech recommendation service subsystem also includes a language template filling module, which performs the filling of the language template, wherein an entity recognition algorithm is used to extract entities from the user conversation context, and the language template is filled by the user information, policy information, insurance product information, extracted entity data and other auxiliary information in the language template; the language text that needs to be prompted to the insurance salesperson is obtained, and the speech recommendation information under the sales task is fed back to the insurance agent of the insurance sales task speech recommendation agent subsystem.

[0075] It will be appreciated by those skilled in the art that the present invention has been described with reference to the flowcharts and / or block diagrams of the methods, systems, and computer program products according to the present invention. Each block in the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions (which are passed through the processor of the computer or other programmable data processing device) generate means for implementing the functions specified in one or more blocks in the flowcharts and / or block diagrams.

[0076] Those skilled in the art will appreciate that the embodiments of the present invention described above and shown in the accompanying drawings are intended to be illustrative only and are not intended to limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments. Any variations or modifications may be made to the embodiments of the present invention without departing from these principles.

Claims

1. A method for recommending insurance sales task scripts based on contextual semantic understanding, characterized in that: The method for recommending speech techniques for insurance sales tasks based on contextual semantic understanding includes the following steps: Provides a service for recommending scripts for insurance sales tasks, and obtains real-time conversation context information between insurance agents and users. Input the acquired current conversation context information into the speech recommendation algorithm model; The speech recommendation algorithm model executes the speech recommendation algorithm and processes the input conversation context information; The sales pitch recommendation algorithm model outputs sales tips; Provide insurance sales task script recommendation agent side, and provide real-time feedback of sales prompts to the insurance agents on the insurance sales task script recommendation agent side. The insurance agents will then describe the content of the insurance sales task according to the prompts. The insurance sales task script recommendation server obtains user feedback information; and The insurance sales task script recommendation server determines whether the current sales task has been completed. If not, it obtains and updates the conversation context information and re-inputs the script recommendation algorithm model until the sales task is completed. Classify the current conversation context information to obtain a categorized list of all user information in the entire context; recommend a language template based on the categorized list of the conversation process; and fill in information on the recommended template to obtain a text of recommended words to prompt the salesperson and output sales prompts; The method for recommending speech scripts for insurance sales tasks based on contextual semantic understanding further includes the steps of training a speech script recommendation algorithm model: preparing input data and input format, wherein the input data includes current context data, text transcribed by ASR of speech data during the conversation, emotional identifiers extracted from speech data, sales process steps of insurance products, customer information, auxiliary processes and step information; the input data is manually annotated as labeled classification data before being input, and is sequentially concatenated into a string of length not exceeding 1024, and training input and recommendation input are executed; and classification training is performed on the input text through the VDCNN algorithm, and specific scenario classification results are returned through the input to execute insurance product sales steps or insurance sales auxiliary steps; Obtain user information and categorized data of the conversation text context; use a user-based collaborative filtering algorithm to perform recommendations based on the categorized data of the conversation context, and provide real-time feedback of sales prompts to the insurance agent who recommends the sales task script. Execute the filling of language templates, wherein, the entity recognition algorithm is used to extract entities from the user conversation context, and the language template is filled by the user information, policy information, insurance product information, extracted entity data and other auxiliary information in the language template; obtain the language text that needs to be prompted to the insurance salesperson, and feedback the recommended speech information under the sales task to the insurance agent on the insurance sales task speech recommendation agent side.

2. A system for recommending insurance sales task scripts based on contextual semantic understanding, which is applied to the method for recommending insurance sales task scripts based on contextual semantic understanding according to claim 1, characterized in that: The insurance sales task speech recommendation system based on contextual semantic understanding includes an insurance sales task speech recommendation service subsystem and an insurance sales task speech recommendation agent subsystem; the insurance sales task speech recommendation service subsystem is configured from the insurance sales task speech recommendation agent subsystem to: obtain the conversation context information between the current insurance agent and the user in real time; input the obtained current conversation context information into the speech recommendation algorithm model; the speech recommendation algorithm model executes the speech recommendation algorithm and processes the input conversation context information; the speech recommendation algorithm model outputs sales prompt terms; and feeds back the sales prompt terms to the insurance sales task speech recommendation agent subsystem in real time; wherein the insurance sales task speech recommendation agent subsystem is provided with a human-computer interaction unit, which obtains the sales prompt terms fed back by the insurance sales task speech recommendation service subsystem in real time; wherein the insurance sales task speech recommendation agent subsystem is further configured to: send user feedback information to the insurance sales task speech recommendation service subsystem; wherein the insurance sales task speech recommendation service subsystem also includes a sales task judgment unit, and the sales task judgment unit is configured to: judge whether the current sales task is achieved, if If it is not completed, the conversation context information is obtained and updated, and the speech recommendation algorithm model is re-entered until the sales task is achieved; wherein the insurance sales task speech recommendation service subsystem also includes a speech recommendation algorithm model unit, and the speech recommendation algorithm model of the speech recommendation algorithm model unit executes the speech recommendation algorithm, processes the input conversation context information, and outputs sales prompts; the speech recommendation algorithm model unit is also configured to: perform classification on the current conversation context information, obtain a classification list of all user information in the entire context; perform language template recommendation for the classification list of the conversation process; perform information filling on the recommended template, obtain the recommended speech text to the salesperson, and output sales prompts; wherein the speech recommendation algorithm model unit is also configured to: prepare input data and input format, wherein the input data includes the current context data, the text of the ASR transcription of the conversation process voice data, the emotional identifier extracted from the voice data, the sales process steps of the insurance product, customer information, auxiliary process and step information, the input data is manually marked as labeled classification data before being input, and is spliced ​​into a string of no more than 1024 in length in sequence, and performs training input and recommendation input; wherein The speech recommendation algorithm model unit is also configured to: perform classification training on the input text through the VDCNN algorithm, return the specific scenario classification results through the input, and execute the insurance product sales steps or insurance sales auxiliary steps; wherein, the speech recommendation algorithm model unit is also configured to: obtain user information and classification data of the conversation text context; based on the user's collaborative filtering algorithm, perform recommendations according to the classification data of the conversation context, and provide real-time feedback of sales prompt terms to the insurance agent of the insurance sales task speech recommendation agent subsystem; wherein the insurance sales task speech recommendation service subsystem also includes a language template filling module, and the language template filling module performs the filling of the language template, wherein, the entity recognition algorithm is used to extract entities from the user conversation context, and the language template is filled by the user information, policy information, insurance product information, extracted entity data and other auxiliary information in the language template; obtain the language text that needs to be prompted to the insurance salesperson, and feedback the speech recommendation information under the sales task to the insurance agent of the insurance sales task speech recommendation agent subsystem.

3. A device for recommending insurance sales task scripts based on contextual semantic understanding, characterized in that: include: Memory, for storing software applications, A processor is used to execute the software application, and each program of the software application correspondingly executes the steps in the insurance sales task speech recommendation method based on contextual semantic understanding as described in claim 1.

Images