Processing method and device for session information, equipment and medium

Abstract

The embodiment of the invention discloses a processing method and device for session information, equipment and a medium. The scheme comprises the following steps: acquiring session information introduced by a customer service and a user for a service product, and retrieving a target reference sample similar to the session information from a sample library; the target reference sample comprises sample session information, a sample sales stage corresponding to the sample session information, and sample thinking chain information; the sample thinking chain information is used for describing an analysis process of the sample sales stage obtained by the sample session information; then utilizing a target reference sample and the session information to obtain an enhanced cue word; and providing the enhanced cue word to a large model, and obtaining a sales stage corresponding to the session information and thinking chain information used for describing an analysis process of obtaining the sales stage from the session information by using the large model.

Description

Technical Field

[0001] This application relates to the field of computer technology, and particularly to a method, apparatus, device, and medium for processing session information. Background Art

[0002] With the development of technology, various commodities can be sold online. During the process of commodity sales, in order to better understand user needs and provide better services to users, it is necessary to identify the sales stage where the user or salesperson is located. Based on this, how to identify the sales stage has become a technical problem to be solved urgently. Summary of the Invention

[0003] Embodiments of this specification provide a method, apparatus, device, and medium for processing session information to solve the problem of resource waste existing in the existing session information processing method.

[0004] To solve the above technical problems, the embodiments of this specification are implemented as follows:

[0005] A method for processing session information provided by an embodiment of this specification includes:

[0006] Obtain the session information of the customer service and the user regarding the introduction of business products;

[0007] Retrieve a target reference example similar to the session information from the example library; the target reference example includes example session information, the example sales stage corresponding to the example session information, and example thought chain information; the example thought chain information is used to describe the analysis process of obtaining the example sales stage from the example session information;

[0008] Use the target reference example and the session information to obtain an enhanced prompt word; the enhanced prompt word includes the target reference example and the session information;

[0009] Provide the enhanced prompt word to a large model, and use the large model to obtain the sales stage corresponding to the session information and the thought chain information used to describe the analysis process of obtaining the sales stage from the session information.

[0010] An apparatus for processing session information provided by an embodiment of this specification includes:

[0011] An acquisition module, configured to obtain the session information of the customer service and the user regarding the introduction of business products;

[0012] A retrieval module, configured to retrieve a target reference example similar to the session information from an example library; the target reference example includes example session information, an example sales stage corresponding to the example session information, and example thought chain information; the example thought chain information is used to describe the analysis process of obtaining the example sales stage from the example session information;

[0013] A prompt word generation module, configured to obtain an enhanced prompt word by using the target reference example and the session information; the enhanced prompt word includes the target reference example and the session information;

[0014] A result generation module, provides the enhanced prompt word to a large model, and uses the large model to obtain the sales stage corresponding to the session information and the thought chain information for describing the analysis process of obtaining the sales stage from the session information.

[0015] A processing device for session information provided by an embodiment of this specification includes:

[0016] At least one processor; and,

[0017] A memory communicatively connected to the at least one processor; wherein,

[0018] The memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to:

[0019] Obtain the session information of the customer service and the user regarding the introduction of the business product;

[0020] Retrieve a target reference example similar to the session information from an example library; the target reference example includes example session information, an example sales stage corresponding to the example session information, and example thought chain information; the example thought chain information is used to describe the analysis process of obtaining the example sales stage from the example session information;

[0021] Obtain an enhanced prompt word by using the target reference example and the session information; the enhanced prompt word includes the target reference example and the session information;

[0022] Provide the enhanced prompt word to a large model, and use the large model to obtain the sales stage corresponding to the session information and the thought chain information for describing the analysis process of obtaining the sales stage from the session information.

[0023] A computer-readable medium provided by an embodiment of this specification, on which computer-readable instructions are stored, and the computer-readable instructions can be executed by a processor to implement a method for processing session information.

[0024] One embodiment of this specification achieves the following beneficial effects:

[0025] By obtaining the conversation information between the customer service and the user regarding the business product introduction, and retrieving the target reference example similar to the conversation information from the example library; then using the target reference example and the conversation information to obtain an enhanced prompt; providing the enhanced prompt to the large model, and using the large model to obtain the sales stage corresponding to the conversation information and the thought chain information for describing the analysis process of obtaining the sales stage from the conversation information. In the embodiment of this specification, the target reference example may include the example conversation information, the example sales stage corresponding to the example conversation information, and the example thought chain information, and the example thought chain information can describe the analysis process of obtaining the example sales stage from the example conversation information. In this way, the obtained enhanced prompt can more clearly and accurately guide the large model to analyze the conversation information between the customer service and the user regarding the business product introduction, and identify the sales stage from the conversation information, which can improve the accuracy of identifying the sales stage.

[0026] On the other hand, retrieving the target reference example similar to the conversation information from the example library as the reference example of the large model without the need to pre-train the large model with a large number of samples and then use it. This can avoid manual annotation for a large number of samples and reduce the manual annotation cost.

[0027] On the other hand, in the embodiment of this specification, the thought chain information for the conversation information can be output, which can also improve the interpretability of the obtained sales stage. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in this application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0029] Figure 1 is a schematic diagram of an application scenario of a method for processing conversation information provided by an embodiment of this specification;

[0030] Figure 2 is a flowchart of a method for processing conversation information provided by an embodiment of this specification;

[0031] Figure 3 is a swimlane diagram of a method for processing conversation information provided by an embodiment of this specification;

[0032] Figure 4 is a schematic structural diagram of a device for processing conversation information provided by an embodiment of this specification;

[0033] Figure 5 It is a schematic structural diagram of a processing device for session information provided by an embodiment of this specification. Detailed implementation manners

[0034] To make the objectives, technical solutions, and advantages of one or more embodiments of this specification clearer, the technical solutions of one or more embodiments of this specification will be clearly and completely described below in conjunction with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, rather than all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope protected by one or more embodiments of this specification.

[0035] To facilitate the understanding of the embodiments of this specification, the following are explanations of the terms involved in the embodiments of this specification.

[0036] Sales stage: The sales stage refers to the specific stage of a business product in the sales cycle.

[0037] Sales actions: Sales actions refer to a series of proactive behaviors and strategies taken during the sales process aimed at optimizing service quality. Specifically, it can refer to the product steward performing different sales guides for users in different states, such as opening services, information collection, concept implantation, etc.

[0038] Chain of Thought: Chain of Thought (CoT for short in English) refers to the process of decomposing a problem with relatively complex logic and forming a complete thinking process through a series of logically related thoughts.

[0039] Retrieval-augmented Generation: Retrieval-augmented Generation (RAG for short in English) is a model that combines retrieval and generation technologies. When the model needs to generate text or answer questions, it can retrieve relevant information from an externally introduced knowledge base and then use this retrieved information to guide text generation, with strong interpretability and customization capabilities, improving the quality and accuracy of model prediction.

[0040] Large Language Model: Large Language Model (LLM for short in English), also known as large language model or large model, is a natural language processing technology based on deep learning. They can better understand natural language and generate high-quality text according to the given context. For example, models such as GPT-4, Claude, LLaMA, or Qwen.

[0041] The following will describe in detail the technical solutions provided by each embodiment of this specification in conjunction with the accompanying drawings.

[0042] In today's increasingly competitive market environment, various commodities can be sold online. During the process of commodity sales, it is crucial for enterprises to provide precise services to customer service. In order to better understand user needs and thus provide better services to users, it is usually necessary to identify the sales stage of the commodity.

[0043] Currently, it is usually through manually annotating a large number of cases and using the manually annotated cases to fine-tune or train a large model; then using the fine-tuned or trained large model to identify the sales stage of the product. Manually annotating a large number of cases is inefficient and costly; at the same time, since the sales stage identified by the model lacks a specific analysis process, the interpretability of the sales stage finally output by the model is low.

[0044] To solve the defects in the prior art, the following embodiments are given in this solution.

[0045] Figure 1 It is a schematic diagram of an application scenario of a method for processing session information provided by an embodiment of this specification.

[0046] As Figure 1 shown, this solution may include a server 101, a user terminal 102, and a customer service terminal 103; the user terminal 102 may be a terminal device used by a user; the customer service terminal 103 may be a terminal device used by customer service. The user and the customer service can communicate with each other through the user terminal 102 and the customer service terminal 103 respectively. The user terminal 102 and the customer service terminal 103 may be a personal computer (PC), a tablet computer, a smart phone, a personal digital assistant (PDA), etc.

[0047] In practical applications, the server 101 and the customer service terminal 103 may establish a communication connection in a Wi-Fi (Wireless Fidelity) network, a 2G / 3G / 4G network, or a local area network, and obtain the session information of the user and the customer service about the product introduction from the customer service terminal to determine the sales stage of the product according to the session information.

[0048] In addition, the server 101 and the user terminal 102 may also establish a communication connection in a Wi-Fi (Wireless Fidelity) network, a 2G / 3G / 4G network, or a local area network, and after obtaining the authorization of the user, obtain the session information of the user and the customer service about the product introduction from the user terminal 102 to determine the sales stage of the product according to the session information.

[0049] Next, a method for processing session information provided in the embodiments of the specification will be specifically described in conjunction with the accompanying drawings:

[0050] Figure 2 This is a flowchart of a method for processing session information provided in the embodiments of this specification. From a program perspective, the execution entity of the process can be an application server or an application client.

[0051] As Figure 2 shown, this process may include the following steps:

[0052] Step 202: Obtain the session information of the customer service and the user regarding the introduction of the business product.

[0053] In the embodiments of this specification, the customer service can be a telephone service personnel or an online platform service personnel of the seller of the business product, and is used to assist the user in understanding and purchasing the business product. The customer service can be a robot customer service or a human customer service.

[0054] The session information can be in text form. For example, the session information is the information generated by the user and the customer service communicating in text through a terminal device.

[0055] The session information can also be in audio form. For example, the session information is the information generated by the user and the customer service communicating by voice through a terminal device.

[0056] The session information can also be in a form combining text and audio. For example, the session information is the information generated by the user and the customer service communicating through text and voice on an application terminal. Among them, the application terminal can be an application program developed by the seller of the business product and having functions such as introducing the business product and conducting transactions for the business product.

[0057] The business product can be various products that can be traded online; for example, the business product can be an insurance product, a wealth management product, etc.

[0058] Step 204: Retrieve a target reference example similar to the session information from the example library; the target reference example includes example session information, the example sales stage corresponding to the example session information, and example thinking chain information; the example thinking chain information is used to describe the analysis process of obtaining the example sales stage from the example session information.

[0059] In the embodiments of this specification, multiple reference examples can be stored in the example library; each reference example can include example session information, the example sales stage corresponding to the example session information, and example thinking chain information.

[0060] Among them, the reference examples in the example library can be examples of example business products of the same type as the business product involved in the session information;

[0061] Specifically, being the same in terms of the types of business products involved in the session information may mean being the same in terms of the sub-categories to which the business products involved in the session information belong. For example, the sample business product corresponding to the reference sample and the business products involved in the session information are both endowment insurance products; or, the sample business product corresponding to the reference sample and the business products involved in the session information are both wealth management products of the fixed-income type.

[0062] In addition, being the same in terms of the types of business products involved in the session information may also mean being the same in terms of the major categories to which the business products involved in the session information belong; for example, the sample business product corresponding to the reference sample and the business products involved in the session information are both insurance products, or the sample business product corresponding to the reference sample and the business products involved in the session information are both wealth management products.

[0063] In the embodiments of this specification, the sample business products corresponding to the reference samples in the sample library and the business products involved in the session information generally have the same sales stages. For example, the sales stages of the business products involved in the session information include the first stage, the second stage, and the third stage; the sample sales stage in at least one reference sample in the sample library may be the first stage; the sample sales stage in at least one reference sample in the sample library may be the second sales stage; the sample sales stage in at least one reference sample in the sample library may be the third sales stage.

[0064] The sample sales stage can be determined based on the sample session information. In practical applications, the sample sales stage can be obtained by using a model to annotate the sample session information; or it can be obtained by manually annotating the sample session information.

[0065] In practical applications, there can be multiple sales stages for business products, and different business products may include different sales stages; the same category of business products launched by different companies may include the same or different sales stages. For example, the sales stages corresponding to the endowment insurance products launched by Insurance Company A include stage a and stage b, and the sales stages corresponding to the endowment insurance products launched by Insurance Company B include stage a and stage c. The product provider or service provider can define the specific content of the sales stage according to the actual business needs, and no specific limitation is made here.

[0066] The thought chain can guide the large model to generate answers to questions by simulating a step-by-step reasoning process. In the embodiments of this specification, the sample thought chain information can be the specific analysis content for obtaining the sample sales stage by analyzing the sample session information. The sample thought chain information can improve the understanding ability of the large model, facilitate the large model to give the analysis process content of the sales stage determined from the session information, and improve the interpretability of the results.

[0067] In practical applications, the session information can be matched with the sample session information of the reference samples in the sample library, and the reference samples with higher similarity can be used as the target reference samples, and there can be multiple target reference samples.

[0068] Step 206: Use the target reference sample and the session information to obtain an enhanced prompt; the enhanced prompt includes the target reference sample and the session information.

[0069] In the embodiments of this specification, in prompt engineering, a prompt is a text or sentence that can be used to guide a large model to generate a specific response.

[0070] In the embodiments of this specification, the sample library can represent an external database outside the large model, and the target reference samples retrieved from the external knowledge base can be used to enhance the prompt. Specifically, the target reference samples retrieved from the sample library and the session information can be embedded together into a preset prompt template to obtain an enhanced prompt. The enhanced prompt includes knowledge for identifying the sales stage of business products or products in the same major category as the business products, which can achieve the enhancement processing of the prompt and improve the accuracy of the output of the large model.

[0071] Step 208: Provide the enhanced prompt to the large model, and use the large model to obtain the sales stage corresponding to the session information and the thought chain information for describing the analysis process of obtaining the sales stage from the session information.

[0072] In the embodiments of this specification, the large model can be a large language model. Specifically, the large model can be a model of the GPT series, such as some general large language models like GPT-3.5, GPT-4, and GPT-4o. In addition, the large model can also be a model of other series, such as Tongyi Qianwen model, Ant Baoling large model, etc.

[0073] Inputting the enhanced prompt into the large model can enable the large model to generate a more accurate response with a stronger correlation to the identification of the sales stage of business products. On the other hand, it can avoid pre-training the large model with a large number of samples, thereby avoiding manual annotation of a large number of samples, reducing the manual annotation cost, and improving the efficiency. In addition, the enhanced prompt also includes the thought chain information of the analysis process for obtaining the sales stage based on the session information, enabling the large model to output the thought chain information about the sales stage of business products according to the enhanced prompt, and improving the interpretability of determining the sales stage.

[0074] It should be understood that the order of some steps of the method described in one or more embodiments of this specification can be interchanged according to actual needs, or some of the steps can also be omitted or deleted.

[0075] Figure 2 In the method, by obtaining the conversation information between the customer service and the user regarding the business product introduction, and retrieving the target reference example similar to the conversation information from the example library; then using the target reference example and the conversation information to obtain an enhanced prompt word; providing the enhanced prompt word to the large model, and using the large model to obtain the sales stage corresponding to the conversation information and the thought chain information for describing the analysis process of obtaining the sales stage from the conversation information. In the embodiments of this specification, the target reference example may include example conversation information, the example sales stage corresponding to the example conversation information, and example thought chain information, and the example thought chain information can describe the analysis process of obtaining the example sales stage from the example conversation information. In this way, the obtained enhanced prompt word can more clearly and accurately guide the large model to analyze the conversation information between the customer service and the user regarding the business product introduction, and identify the sales stage from the conversation information, which can improve the accuracy of identifying the sales stage.

[0076] On the other hand, retrieving the target reference example similar to the conversation information from the example library as the reference example of the large model does not require pre-training the large model with a large number of samples and then using it. In this way, it is possible to avoid manual annotation for a large number of samples and reduce the manual annotation cost.

[0077] On the other hand, in the embodiments of this specification, the thought chain information for the conversation information can be output, which can also improve the interpretability of the obtained sales stage.

[0078] Based on Figure 2 the method, the embodiments of this specification also provide some specific implementation schemes of this method, which will be described below.

[0079] It can be understood that the thought chain information is the analysis content of obtaining the example sales stage by analyzing the example conversation information. The content of the thought chain information is usually relatively large. If it is manually compiled, it will take a long time and be inefficient. Based on this, in the embodiments of this specification, a method for automatically generating example thought chain information for the reference example is provided.

[0080] Optionally, the example library may include multiple reference examples in different sales stages.

[0081] Before retrieving the target reference example similar to the conversation information from the example library, it may further include:

[0082] For any one of the multiple reference examples, select a thought chain example from the thought chain database that is in the same sales stage as the reference example; the thought chain database includes at least one thought chain example in different sales stages; one thought chain example includes target example conversation information, the target sales stage corresponding to the target example conversation information, and target example thought chain information representing the analysis process of obtaining the target sales stage from the target example conversation information.

[0083] Provide the thought chain example and the any one reference example to the large model, and use the large model to obtain the example thought chain information of the any one reference example.

[0084] Save the example thought chain information to the example library.

[0085] In the embodiments of this specification, the example library may include multiple reference examples, and for reference examples in different sales stages, there may be multiple or one. In order for the large model to identify various different sales stages, the example library may contain reference examples for each sales stage. For example, the sales stages of insurance product C include the first stage, the second stage, and the third stage. There may be one or more reference examples for the first stage, one or more reference examples for the second stage, and one or more reference examples for the third reference stage.

[0086] In practical applications, the reference example may include example conversation information and the example sales stage corresponding to the example conversation information. Among them, the example sales stage corresponding to the example conversation information may be manually labeled or generated by a model. If the reference example includes example thought chain information, the target reference example can be directly screened from the reference example. If the reference example does not include example thought chain information, the large model can be used to generate example thought chain information based on the thought chain database.

[0087] The thought chain database may include multiple thought chain examples, and there may be one or more thought chain examples for different sales stages. One thought chain example may include target example conversation information, the target sales stage corresponding to the target example conversation information, and target example thought chain information representing the analysis process of obtaining the target sales stage from the target example conversation information. Among them, the target example thought chain information may be written by experts.

[0088] To improve efficiency and reduce the time cost consumed by experts, experts can write the thought chain information corresponding to one example for each sales stage. Of course, experts can also write the thought chain information for multiple examples, and no specific limitation is made here.

[0089] In practical applications, the target example chain-of-thought information can also be obtained by first generating it based on a model and then manually reviewing and correcting it. The chain-of-thought database can be constructed based on the reference examples in the example library, or it can be constructed based on reference examples from other sources. Specifically, one or more reference examples can be selected from the reference examples at each sales stage in the example library as chain-of-thought examples to construct the chain of thought.

[0090] For a reference example in the example library, a chain-of-thought example with the same sales stage as the reference example can be selected from the chain-of-thought example library, and the chain-of-thought example and the reference example can be input into the large model together to instruct the large model to output the example chain-of-thought information of the reference example with reference to the chain-of-thought example.

[0091] In practical applications, the large model for generating example chain-of-thought information can be the same as or different from the large model for generating the sales stage and chain-of-thought information in the previous text, and no specific limitation is made here.

[0092] In practical applications, the example conversation information of the reference example and the example chain-of-thought information of the reference example generated using the large model can be stored together; or a corresponding relationship can be established for the conversation information of the reference example and the example chain-of-thought information of the reference example generated using the large model, so that the example chain-of-thought information corresponding to the example conversation information can be queried based on the example conversation information.

[0093] In the embodiments of this specification, using the large model can quickly obtain the example chain-of-thought information of each reference example, improve the generation efficiency of the example chain-of-thought information, and at the same time reduce the cost of manually annotating the example chain-of-thought information.

[0094] For ease of understanding, a specific method for retrieving the target reference example is also provided in the embodiments of this specification.

[0095] Optionally, before retrieving the target reference example similar to the conversation information from the example library, it may further include:

[0096] Extract the user's user conversation information included in the conversation information.

[0097] Use the intent recognition model to determine the user intent sequence corresponding to the user conversation information.

[0098] The retrieving of the reference example similar to the conversation information from the example library may specifically include:

[0099] Retrieve the target reference example similar to the user intent sequence from the example library according to the user intent sequence.

[0100] In practical applications, user session information is the session information provided by the user in the session information between the customer and the user. Specifically, if the user and the customer service communicate through an application terminal, the user session information may be the information sent by the user to the customer service through the application terminal. If the user and the customer service communicate through a phone call, the user session information may be the information spoken by the user during the communication.

[0101] In the embodiments of this specification, if the session information is in text form, a text editor can be used to extract the user's user session information from the session information. Among them, the text editor can be at least one of Notepad++ and Sublime Text. In addition, programming languages such as Python and JavaScript can also be used to write scripts to extract the user's user session information from the text information. Of course, the user session information can also be extracted by other software and hardware that can extract the session information of specific objects, which is not specifically limited here.

[0102] If the session information is in audio form, a voice editor can be used to extract the user's user session information from the session information.

[0103] If the session information is in a combined form of text and audio, the audio in the session information can be first converted into text, and then the above-mentioned text editor or the written script can be used to extract the user session information. It is also possible to use the cooperation of a voice editor and a text editor to extract the user session information from the session information.

[0104] In the embodiments of this specification, if the extracted user session information includes audio information, the audio information can be pre-converted into text information, and then the intent recognition model can be used to determine the user intent sequence corresponding to the user session information; or, if the intent recognition model can recognize voice information, the audio information does not need to be converted into text information, and the intent recognition model can be directly used to determine the user intent sequence corresponding to the user session information.

[0105] The intent recognition model is a model that can recognize the intent corresponding to the session through the session information. The intent recognition model in the embodiments of this specification can be an existing intent recognition model or an intent recognition model trained according to product requirements, which is not specifically limited here.

[0106] During the sales process of different business products, the intent of the user session information can be different. For example, during the sales process of insurance products, the intent of the user sending the user session information can include querying insurance policies, asking about insurance amounts, requesting product recommendations, etc.

[0107] In the embodiments of this specification, the user intention sequence can be matched with the example user intention sequence for similarity, and the reference example with a higher similarity between the user intention sequence and the example user intention sequence is selected as the target reference example.

[0108] In practical applications, the user intention sequence can reflect the user's attitude towards business products or towards customer service, thus directly affecting the sales stage of business products. Determining the target reference example based on the user intention sequence can facilitate improving the accuracy of the model in determining the sales stage of business products, and further improve the accuracy of user demand determination, so as to provide better services for users.

[0109] In practical applications, the example user intention sequence can be pre-identified before obtaining the conversation information between the customer and the user. As an implementation manner, in addition to including the example conversation information, the example sales stage corresponding to the example conversation information, and the example thought chain information, the reference example in the example library can also include the example user intention sequence; among them, the example user intention sequence can be determined by using an intention recognition model to recognize the example user conversation information in the example conversation information.

[0110] In the embodiments of this specification, a specific generation process of the user intention sequence is also provided.

[0111] Optionally, the user conversation information includes multiple user conversation sentences.

[0112] Using the intention recognition model to determine the user intention sequence corresponding to the user conversation information may specifically include:

[0113] Using the intention recognition model to determine the conversation intention corresponding to each user conversation sentence.

[0114] Sort the conversation intentions corresponding to each user conversation sentence in the order of each user conversation sentence in the user conversation information to obtain the user intention sequence of the user.

[0115] In the embodiments of this specification, it can be understood that during the communication between the user and the customer service, each sentence spoken by the user or each piece of information sent to the customer service can be used as a single user conversation sentence. The user conversation information usually includes multiple user conversation sentences.

[0116] In the embodiments of this specification, the intention recognition model can be used to perform intention recognition on each user conversation sentence respectively to determine the conversation intention corresponding to each user conversation sentence. Then, sort the conversation intentions corresponding to each conversation sentence in the order of the user conversation sentences to obtain the user intention sequence.

[0117] For example, multiple user conversation sentences included in the user conversation information are respectively "What is the premium?", "What is the insured amount of the insurance?", "Show me my insurance policy", and the conversation intents determined by using the intent recognition model are respectively "Inquire about premium", "Inquire about insured amount", "View insurance policy"; then the user intent sequence is "Inquire about premium, Inquire about insured amount, View insurance policy".

[0118] For another example, multiple user conversation sentences included in the user conversation information are respectively "What is the premium?", "Show me my insurance policy", "What was the insured amount of the insurance again?", and the conversation intents determined by using the intent recognition model are respectively "Inquire about premium", "View insurance policy", "Inquire about insured amount"; then the user intent sequence is "Inquire about premium, View insurance policy, Inquire about insured amount".

[0119] It can be understood that if the user conversation sentence of the user is about the content of other fields, it may cause the intent recognition model to be unable to recognize the conversation intent of the user conversation sentence, then the conversation intent of the user conversation sentence can be represented by a specified character. The conversation intent represented by the specified character can be deleted before or after sorting the conversation intents in the order of the user conversation sentences to obtain the user intent sequence. Of course, the above-mentioned conversation intent represented by the specified character can also not be deleted, so as to obtain the user intent sequence.

[0120] In addition, the specific content for determining the target reference example based on the user intent sequence is also provided in the embodiments of this specification.

[0121] Optionally, the retrieving the target reference example similar to the user intent sequence from the example library according to the user intent sequence may specifically include:

[0122] Calculating a first similarity degree between the user intent sequence and the example user intent sequences of each reference example in the example library; the example user intent sequence is obtained by performing intent recognition on the conversation information of the example user in the reference example by using the intent recognition model.

[0123] Determining the reference example with the first similarity degree greater than or equal to a first preset degree as the target reference example similar to the user intent sequence.

[0124] Or, sorting each reference example according to the first similarity degree, and determining the reference examples located in the first preset number of digits as the target reference examples similar to the user intent sequence.

[0125] For the convenience of application and to improve the retrieval efficiency, the example user intent sequence can also be pre-generated and saved.

[0126] Optionally, before retrieving the target reference example similar to the user intention sequence from the example library according to the above-mentioned user intention sequence, the following steps may also be included:

[0127] For any reference example in the example library, extract the example user session information of the example user included in the any reference example; the example user session information includes multiple example user conversation sentences.

[0128] Use the intention recognition model to determine the example conversation intention corresponding to each example user conversation sentence.

[0129] Sort the example conversation intentions corresponding to each example user conversation sentence according to the order of each example user conversation sentence in the example user session information to obtain the example user intention sequence of the example user.

[0130] Save the example user intention sequence to the example library.

[0131] In practical applications, the example conversation information in the reference example may include example user session information. The example user session information may be the session information provided by the example user in the example conversation information between the example user and the example customer service. The example user session information may include multiple example user conversation sentences. Each message sent by the example user to the example customer service may be an example user conversation sentence.

[0132] The intention recognition model can be used to perform intention recognition on each example user conversation sentence respectively to determine the example conversation intention corresponding to each example user conversation sentence. Then, sort and splice the example conversation intentions corresponding to each example user conversation sentence according to the order of the example user conversation sentences to obtain the example user intention sequence.

[0133] After obtaining the example user intention sequence, the example user intention sequence can be saved in the example library for subsequent applications. Specifically, the example conversation information of the reference example and the example user intention sequence of the reference example can be stored together; or, a corresponding relationship can be established between the example conversation information in the reference example and the example user intention sequence of the reference example, so that the example user intention sequence corresponding to the example conversation information can be queried based on the example conversation information.

[0134] In the embodiments of this specification, the user intention sequence can be matched with the example user intention sequences of each reference example in the example library to determine the target reference example. For example, the target reference example can be determined by calculating the first similarity degree between the user intention sequence and the example user intention sequences of each reference example in the example library.

[0135] The similarity algorithm can be used to calculate the first similarity degree between the user intention sequence and the sample user intention sequences of each reference sample in the sample library. The similarity algorithm can be one or more of the cosine similarity algorithm, Euclidean distance algorithm, edit distance algorithm, and Manhattan distance algorithm.

[0136] Specifically, the similarity degree between the conversation intention at the same position in the user intention sequence and the sample conversation intention in the sample user intention sequence can be calculated first; then, the similarity degrees of the conversation intention and the sample conversation intention at each position are weighted to obtain the first similarity degree.

[0137] In practical applications, the first similarity degree can be represented by a score or a percentage; the reference sample with the first similarity degree between the user intention sequence and the sample user intention sequence greater than the first preset similarity degree can be selected as the target reference sample. It is also possible to sort the first similarity degrees between the user intention sequence and each sample user intention sequence from largest to smallest, and select the reference samples corresponding to the first preset number of sample user intention sequences as the target reference samples. It is also possible to sort the first similarity degrees between the user intention sequence and each sample user intention sequence from smallest to largest, and select the reference samples corresponding to the last first preset number of sample user intention sequences as the target reference samples. Among them, the first preset number can be set according to actual needs and will not be specifically limited here.

[0138] In the embodiments of this specification, in order to further improve the efficiency of retrieving the target reference sample from the sample library based on the user intention sequence, the embodiments of this specification also make further adjustments to the user conversation intention sequence.

[0139] Optionally, the step of sorting the conversation intentions corresponding to each user conversation sentence in the order of each user conversation sentence in the user conversation information to obtain the user intention sequence of the user may specifically include:

[0140] Sort the conversation intentions corresponding to each user conversation sentence in the order of each user conversation sentence in the user conversation information to obtain an initial user intention sequence.

[0141] If the first conversation intention included in the initial user intention sequence is the same as the adjacent second conversation intention, then delete the first conversation intention or the second conversation intention to obtain the user intention sequence.

[0142] After obtaining the conversation intention of each user conversation sentence using the intention recognition model, the conversation intentions can be sorted first in the order of the user conversation sentences in the user conversation information to obtain an initial user intention sequence.

[0143] If the initial user intention sequence contains n consecutive identical session intentions, then n - 1 of the consecutive identical session intentions can be deleted, and only one of them is retained. Here, n is greater than or equal to 2. For example, the obtained initial user intention sequence is "Ask about premium", "Ask about premium", "Ask about premium", "Chat", "Ask about sum insured", "Ask about sum insured", "Ask about sum insured", "Request product recommendation", "Query policy". Then, two of the three consecutive identical "Ask about premium" can be deleted, and only one "Ask about premium" is retained; and two of the three consecutive identical "Ask about sum insured" can be deleted, and only one "Ask about sum insured" is retained.

[0144] If the session intentions included in the initial user intention sequence do not have consecutive identities, no deletion is required. That is to say, if the session intentions included in the initial user intention sequence do not have consecutive identities, then the initial user intention sequence is the finally determined user intention sequence. For example, the initial user intention sequence is "Ask about premium", "Ask about sum insured", "Chat", "Ask about premium", "Ask about sum insured", "Ask about premium", "Ask about sum insured", "Request product recommendation", "Query policy". Since there are no consecutive identical session intentions in the initial user intention sequence, the initial user intention sequence is the user intention sequence.

[0145] In practical applications, a similar method as above can be used to obtain the sample user intention sequence.

[0146] Optionally, the sample session intentions corresponding to each sample user conversation sentence can be sorted according to the order of each sample user conversation sentence in the sample user conversation information to obtain the sample user intention sequence of the sample user. Specifically:

[0147] Sort the sample session intentions corresponding to each sample user conversation sentence according to the order of each sample user conversation sentence in the sample user conversation information to obtain the initial sample user intention sequence.

[0148] If the third session intention included in the initial sample user intention sequence is the same as the adjacent fourth session intention, then delete the third session intention or the fourth session intention to obtain the sample user intention sequence.

[0149] In the embodiments of this specification, the method for obtaining the sample user intention sequence is similar to the method for obtaining the user intention sequence, and will not be elaborated here.

[0150] In the embodiments of this specification, consecutive identical session intents in the user intent sequence are merged, and consecutive identical sample session intents in the sample user intent sequence are merged. The merged user intent sequence is used to match the sample user intent sequence, reducing the amount of data for matching and improving the efficiency and success rate of matching.

[0151] In addition, another way to determine the target reference sample from the sample library is provided in the embodiments of this specification.

[0152] Optionally, before retrieving the target reference sample similar to the session information from the sample library, it may further include:

[0153] Extract the customer service session information of the customer service included in the session information.

[0154] Using the sales action recognition model, determine the customer service sales action sequence corresponding to the customer service session information.

[0155] Retrieving the reference sample similar to the session information from the sample library may specifically include:

[0156] According to the customer service sales action sequence, retrieve the target reference sample similar to the customer service sales action sequence from the sample library.

[0157] In the embodiments of this specification, the method for extracting the customer service session information of the customer service is similar to the method for extracting the user session information of the user, so it will not be elaborated here. It can be understood that the customer service session information is the information output by the customer service to the customer service. Specifically, if the customer service communicates with the user through the application terminal, the customer service session information may be the information sent by the user to the user through the application terminal. In the scenario where the customer service communicates with the user through phone voice, the customer service session information may be the information spoken by the customer service.

[0158] In the embodiments of this specification, the customer service sales action sequence may be a sequence composed of multiple customer service sales actions; the sales action recognition model may be a model that can recognize the customer service sales actions of the customer service based on the customer service session information. The sales action recognition model may be a neural network model or a large model. During the sales process of different business products, the sales actions performed by the customer service may be different. For example, during the sales process of insurance products, the sales actions of the customer service may include one or more of opening service, demand mining, solution presentation, facilitation, and order stabilization. During the sales process of financial products, the sales actions of the customer service may include opening service, customer demand assessment, financial plan formulation, product introduction and recommendation, risk disclosure and reminder, contract signing, and fund transfer and transaction confirmation.

[0159] In practical applications, in addition to the example conversation information, the example sales stage corresponding to the example conversation information, and the example thought chain information, the reference examples in the example library may further include an example customer service sales stage sequence; wherein, the example customer service sales stage sequence may also be determined by using a sales action recognition model.

[0160] The customer service sales action sequence can be matched with the example customer service sales action sequence for similarity, and the reference example corresponding to the example customer service sales action sequence with a higher similarity to the customer service sales action sequence is selected as the target reference example.

[0161] In the embodiments of this specification, the customer service sales action sequence can intuitively reflect the sales stage of the customer service for the business product. Determining the target reference example based on the customer service sales action sequence can facilitate improving the accuracy of the model in determining the sales stage of the business product, and further improve the accuracy of determining the user's needs, so as to provide better services for users.

[0162] The embodiments of this specification also provide the specific content of obtaining the customer service sales action sequence.

[0163] Optionally, the customer service conversation information includes multiple customer service conversation sentences.

[0164] Using the sales action recognition model to determine the customer service sales action sequence corresponding to the customer service conversation information may specifically include:

[0165] Using the sales action recognition model to determine the sales actions corresponding to each of the customer service conversation sentences.

[0166] Sort the sales actions corresponding to each of the customer service conversation sentences in the order of each of the customer service conversation sentences in the customer service conversation information to obtain the customer service sales action sequence.

[0167] In the embodiments of this specification, during the communication between the customer service and the user, each sentence spoken by the customer service or each piece of information sent to the user can be used as a single customer service conversation sentence. The customer service conversation information usually includes multiple customer service conversation sentences.

[0168] The customer service sales action recognition model can be used to identify the sales actions for each customer service conversation sentence respectively, and determine the sales actions corresponding to each customer service conversation sentence. Then, sort the sales actions corresponding to each conversation sentence in the order of the customer service conversation sentences to obtain the customer service sales action sequence.

[0169] For example, the multiple customer service conversation sentences included in the customer service conversation information are respectively "Hello, I am the customer service steward connecting with you from Insurance Company A", "Then you can contact me at any time. I can provide you with one-on-one service at any time", "Do you have any questions about relevant insurance policies? For example, those related to claims or policy interpretation. If so, I can give you a comprehensive and detailed explanation", and the sales actions determined by the sales action recognition model are respectively "opening service", "opening service", "demand mining"; then the customer service sales action sequence is "opening service, opening service, demand mining".

[0170] In the embodiments of this specification, the method of determining the customer service sales action sequence by using the sales action recognition model to determine the sales action of each customer service conversation sentence is convenient and fast.

[0171] In addition, the embodiments of this specification also provide the specific content for determining the target reference example based on the customer service sales action sequence.

[0172] Optionally, retrieving the target reference example similar to the customer service sales action sequence from the example library according to the customer service sales action sequence may specifically include:

[0173] Calculating the second similarity degree between the customer service sales action sequence and the example customer service sales action sequences of each reference example in the example library; the example customer service sales action sequence is obtained by performing sales action recognition on the conversation information of the example customer service in the reference example by using the sales action recognition model.

[0174] Determining the reference example with the second similarity degree greater than or equal to the second preset degree as the target reference example similar to the customer service sales action sequence.

[0175] Or, sorting each reference example according to the second similarity degree, and determining the reference examples in the first second preset positions as the target reference examples similar to the customer service sales action sequence.

[0176] In practical applications, the reference examples stored in the example library may also include example customer service sales action sequences; before retrieving the target reference example similar to the customer service sales action sequence from the example library according to the customer service sales action sequence, the example customer service sales action sequence may also be pre-generated, specifically:

[0177] For any reference example in the example library, extracting the example customer service conversation information of the example customer included in the any reference example; the example customer service conversation information includes multiple example customer service conversation sentences.

[0178] Use the sales action recognition model to determine the sample sales actions corresponding to each of the sample customer service conversation sentences.

[0179] Sort the sample sales actions corresponding to each of the sample customer service conversation sentences in the order of each of the sample customer service conversation sentences in the sample customer service conversation information to obtain the sample customer service sales action sequence of the sample customer service.

[0180] Save the sample customer service sales action sequence to the sample library.

[0181] The sample customer service conversation information is the conversation information sent by the sample customer service to the sample user. The sample customer service conversation information may include multiple sample customer service conversation sentences. Each message sent by the sample customer service to the sample user can be used as a sample customer service conversation sentence.

[0182] In practical applications, the sales action recognition model can be used to identify the sales actions for each sample customer service conversation sentence respectively, and determine the sample sales actions corresponding to each sample customer service conversation sentence. Then, sort and splice the sample sales actions corresponding to each sample customer service conversation sentence in the order of the sample customer service conversation sentences to obtain the sample customer service sales action sequence.

[0183] After obtaining the sample customer service sales action sequence, the sample customer service sales action sequence can be saved in the sample library for subsequent applications. Specifically, the sample customer service sales action sequence of the same reference sample can be stored together with the sample conversation information, or a corresponding relationship can be established between the sample customer service sales action sequence and the sample conversation information in the same reference sample, so that the sample customer service sales action sequence corresponding to the sample conversation information can be queried based on the sample conversation information.

[0184] In the embodiments of this specification, the customer service sales action sequence can be matched with the sample customer service sales action sequences of each reference sample in the sample library to determine the target reference sample. Specifically, the target reference sample can be determined by calculating the second similarity degree between the customer service sales action sequence and the sample customer service sales action sequences of each reference sample in the sample library.

[0185] Among them, the method for calculating the second similarity degree between the customer service sales action sequence and the sample customer service sales action sequence of the reference sample is similar to the method for calculating the first similarity degree between the user intention sequence and the sample user intention sequence in the previous text, and will not be elaborated here.

[0186] In practical applications, the second similarity degree can also be expressed in fractions or percentages; the manner of determining the target reference sample according to the second similarity degree is similar to the manner of determining the target reference sample according to the first similarity degree in the previous text, and will not be elaborated here.

[0187] In the embodiments of this specification, since matching each sales action in the customer service sales action sequence with each sample sales action in the sample customer service sales action sequence is inefficient and has a low success rate, therefore, in the embodiments of this specification, further adjustments are made to the customer service sales action sequence.

[0188] Optionally, the sorting the sales actions corresponding to each customer service conversation sentence in the order of each customer service conversation sentence in the customer service conversation information to obtain the customer service sales action sequence may specifically include:

[0189] Sort the sales actions corresponding to each customer service conversation sentence in the order of each customer service conversation sentence in the customer service conversation information to obtain an initial sales action sequence.

[0190] If the first sales action included in the initial sales action sequence is the same as the adjacent second sales action, then delete the first sales action or the second sales action to obtain the customer service sales action sequence.

[0191] In practical applications, a sample customer service sales action sequence can be obtained based on a similar method as above.

[0192] Optionally, the sorting the sample sales actions corresponding to each sample customer service conversation sentence in the order of each sample customer service conversation sentence in the sample customer service conversation information to obtain the sample customer service sales action sequence of the sample customer service may specifically be:

[0193] Sort the sample sales actions corresponding to each sample customer service conversation sentence in the order of each sample customer service conversation sentence in the sample customer service conversation information to obtain an initial sample sales action sequence.

[0194] If the third sales action included in the initial sample customer service sales action sequence is the same as the adjacent fourth sales action, then delete the third sales action or the fourth sales action to obtain the sample customer service sales action sequence.

[0195] In the embodiments of this specification, after obtaining the sales action of each customer service conversation sentence by using the sales action recognition model, the sales actions can first be sorted in the order of the customer service conversation sentences in the customer service conversation information to obtain an initial sales action sequence.

[0196] If the initial sales action sequence contains m consecutive identical sales actions, then m - 1 of the consecutive identical sales actions can be deleted, and only one of them is retained. Here, m is greater than or equal to 2. For example, the obtained initial sales action sequence is "Opening Service", "Needs Mining", "Solution Presentation", "Solution Presentation", "Solution Presentation", "Closing". Then, two of the three consecutive identical "Solution Presentation" can be deleted, and only one "Solution Presentation" is retained. The resulting customer service sales action sequence is "Opening Service, Needs Mining, Solution Presentation, Closing".

[0197] If the sales actions contained in the initial sales action sequence are not consecutively identical, no deletion is required. For example, the initial sales sequence is "Opening Service", "Needs Mining", "Solution Presentation", "Needs Mining", "Solution Presentation", "Closing". Since there are no consecutively identical sales actions in the initial sales action sequence, the initial sales action sequence is the customer service sales action sequence.

[0198] In the embodiments of this specification, the method for obtaining the sample customer service sales action sequence is similar to the method for obtaining the customer service sales action sequence described above, and will not be elaborated here.

[0199] In the embodiments of this specification, the consecutively identical sales actions in the customer service sales action sequence are merged, and the consecutively identical sample sales actions in the sample customer service sales action sequence are also merged. The merged customer service sales action sequence is used to match the sample customer service sales action sequence, reducing the amount of data for matching and improving the efficiency and success rate of matching.

[0200] In addition, another method for retrieving the target reference sample from the sample library is provided in the embodiments of this specification.

[0201] Optionally, retrieving the target reference sample similar to the session information from the sample library may specifically include:

[0202] Selecting a first set of reference samples similar to the user intention sequence from the sample library according to the user intention sequence corresponding to the session information; the user intention sequence is obtained by performing intention recognition on the user's session information included in the session information using an intention recognition model.

[0203] Retrieving the target reference sample similar to the customer service sales action sequence from the first set of reference samples according to the customer service sales action sequence corresponding to the session information.

[0204] In the embodiments of this specification, the user intention sequence corresponding to the session information may be the user intention sequence corresponding to the user session information. The user intention sequence may be recognized based on the user session information by using an intention recognition model. The user session information may be the session information of the user extracted from the session information. It can be understood that the user session information is the information output by the user to the customer service. Different extraction methods may be adopted for different forms of session information. Regarding the specific content of extracting the user session information from the session information, since it has been described in detail above, it will not be elaborated here.

[0205] Among them, the intention recognition model may be a model that can recognize the intention related to the business product in the user session information sent by the user; the intention recognition model may be a neural network model or a large model. The method of using the intention recognition model to recognize the user intention sequence can be referred to the previous text and will not be elaborated here.

[0206] In the embodiments of this specification, the sample user intention sequence is included in the reference samples in the sample library, and the sample user intention sequence may be recognized by using the intention recognition model for the sample user session information in the reference samples.

[0207] The first reference sample set may be filtered according to the similarity degree between the user intention sequence and the sample user intention sequence in the reference samples in the sample library. Specifically, the similarity degree between the user intention sequence and the sample user intention sequence in the reference samples in the sample library may be calculated by using a similarity algorithm, and the first reference sample set may be determined based on the similarity degree between the user intention sequence and the sample user intention sequence. The similarity algorithm may be one or more of the cosine similarity algorithm, Euclidean distance algorithm, edit distance algorithm, and Manhattan distance algorithm.

[0208] In the embodiments of this specification, the content of filtering the first reference sample set according to the similarity degree between the user intention sequence and the sample user intention sequence is similar to the content of filtering the target reference sample according to the first similarity degree in the previous text and will not be elaborated here.

[0209] The customer service sales action sequence corresponding to the session information may be the customer service sales action sequence corresponding to the customer service session information; the customer service sales action sequence may be recognized by the sales action recognition model based on the customer service session information. The customer service session information may be the session information of the customer service extracted from the session information. It can be understood that the customer service session information is the information output by the customer service to the user. The method of extracting the customer service session information is similar to the method of extracting the user session information and will not be elaborated here.

[0210] The sales action recognition model can be a model capable of recognizing the sales actions of a customer service agent based on the conversation information of the customer service agent. The sales action recognition model can be a neural network model or a large model. For the content of obtaining the customer service sales action sequence using the sales action recognition model, reference can be made to the foregoing, which will not be elaborated herein.

[0211] For the convenience of application and understanding, the reference examples in the example library include example customer service sales action sequences, and the example customer service sales action sequences can be obtained by using the sales action recognition model to recognize the example customer service conversation information in the reference examples.

[0212] In the embodiments of this specification, a similarity algorithm can be used to calculate the similarity between the customer service sales action sequence and the example customer service sales action sequence of the reference example in the first reference example set, so as to determine the target reference example according to the similarity between the customer service sales action sequence and the example customer service sales action sequence of the reference example in the first reference example set.

[0213] The content of determining the target reference example according to the similarity between the customer service sales action sequence and the example customer service sales action sequence of the reference example in the first reference example set is similar to the content of determining the target reference example according to the second similarity degree in the foregoing, which will not be elaborated herein.

[0214] In the embodiments of this specification, determining the target reference example based on the user intention sequence and the customer service sales action sequence makes the scenario of the target reference example very close to the scenario of the conversation information, improves the similarity between the selected target reference example and the conversation information, and thus improves the accuracy of the sales stage and the thought chain information of the conversation information output by the large model based on the target reference example.

[0215] Optionally, the business product introduction may include the introduction of selling insurance products.

[0216] Optionally, the example sales stage and the sales stage may include at least one of opening service, demand mining, product recommendation, solution presentation, facilitation, and closing.

[0217] Opening service: In the opening service stage, the customer service agent is usually a product planner. The product planner can attract the user's attention through effective topics, establish trust and a friendly relationship, so as to guide the user to speak and take the first step smoothly for the entire sales process.

[0218] In the demand mining stage, the product planner can deeply understand the user's needs and pain points by asking questions and listening. In this stage, the product planner needs to pay attention to aspects such as the user's concerns, risk tolerance, and budget, and implant concepts in combination with the user's own situation to enhance the user's understanding of insurance and the determination to purchase insurance.

[0219] In the product recommendation stage, product planners can recommend the most suitable insurance products to users based on their needs and risk tolerance. At this stage, product planners need to understand the characteristics and advantages of various products in order to explain the value and coverage of the products to users.

[0220] During the plan presentation stage, product planners can customize insurance plans for users based on the product's characteristics and advantages and in combination with user needs.

[0221] In the promotion and transaction stage, product planners can help users buy insurance products through effective services. At this stage, product planners need to understand users’ purchasing intentions and decision-making processes in order to provide users with the most appropriate purchase advice and support.

[0222] Accurately identifying the user's current sales stage can help product planners provide more accurate services to users.

[0223] Optionally, the sales action may include at least one of opening service, demand exploration, product introduction, solution presentation, promotion, and order retention.

[0224] Figure 3 A swim lane diagram of a method for processing session information provided in an embodiment of this specification. Figure 3 As shown, the process of the session information may involve execution entities such as terminal devices and servers, and the process may include a session information acquisition phase and a session information processing phase.

[0225] In the session information acquisition stage, the execution subject may include a terminal device, and specifically may include the following steps:

[0226] Step 302: Acquire conversation information between customer service and the user regarding business product introduction.

[0227] In the embodiments of this specification, the terminal for acquiring the conversation information between the customer service and the user regarding the introduction of the business product may generally be a customer service terminal device.

[0228] The session information may be information generated by telephone voice communication between the user and the customer service using the terminal device; it may also be information generated by communication between the user and the customer service using the application or application terminal installed in the terminal device.

[0229] In practical applications, the session can be initiated by the customer service to the user, or by the user to the customer service. Specifically, the user can initiate a session with the customer service by clicking on the customer service icon in the user application terminal. After the user clicks on the customer service icon, the user can enter the session interface, and the user can input relevant content in the session interface. The relevant content input by the user can be a question related to the business product. After the customer service application terminal receives the question sent by the user application terminal, the server or the customer service application terminal can display prompt information for explaining or answering the user's question based on the question sent by the user. The customer service can send the prompt information to the user or reply to the user on its own according to the prompt information.

[0230] The customer service terminal can obtain session information at a predetermined frequency during the session, or obtain session information after the session ends, and transmit the session information to the server.

[0231] In the processing stage of the session information, the execution entity can include the server, and the server can be the server of the seller of the business product. In the processing stage of the session information, it can specifically include the following steps:

[0232] Step 304: Obtain the session information of the customer service and the user regarding the introduction of the business product sent by the terminal.

[0233] In the embodiments of this specification, the session information can be in text form, audio form, or a combination of audio and text form.

[0234] Step 306: Extract the user session information of the user included in the session information.

[0235] In the embodiments of this specification, the user session information is the information output by the user to the customer service.

[0236] In practical applications, different extraction methods can be adopted for different forms of session information. If the session information is in text form, a text editor can be used to extract the user session information; if the session information is in audio form, a voice editor can be used to extract the user session information from the session information; if the session information is in a combination of text and audio form, the audio in the session information can be first converted into text, and then the above-mentioned text editor can be used to extract the user session information from the session information. It is also possible to extract the user session information of the user from the session information through the cooperation of the voice editor and the text editor.

[0237] Step 308: Use the intent recognition model to determine the user intent sequence corresponding to the user session information.

[0238] The intent recognition model can be a model that can recognize the intent related to the business product in the user session information sent by the user; the intent recognition model can be a neural network model or a large model.

[0239] In the embodiments of this specification, the user session information may include multiple user session sentences; an intent recognition model can be used to perform intent recognition on each user session sentence respectively to determine the session intent corresponding to each user session sentence. Then, sort the session intents corresponding to each session sentence in the order of the user session sentences, and merge the consecutive identical session intents to obtain the user intent sequence.

[0240] Step 310: According to the user intent sequence, retrieve a first reference example set similar to the user intent sequence from the example library; the reference examples in the example library include example session information, the example sales stage corresponding to the example session information, and example thought chain information; the example thought chain information is used to describe the analysis process of obtaining the example sales stage from the example session information.

[0241] In the embodiments of this specification, the example library can be an existing external knowledge base related to business products, or an external knowledge base constructed according to business products. Multiple reference examples can be stored in the example library; each reference example can include example session information, the example sales stage corresponding to the example session information, example thought chain information, and an example user intent sequence. Among them, the example user intent sequence can be obtained by using an intent recognition model based on the example session information. It can be understood that the example user intent sequence can be a sequence obtained by merging consecutive identical example session intents.

[0242] A similarity algorithm can be used to calculate the first similarity degree between the user intent sequence and the example user intent sequences of each reference example in the example library. Then, determine the reference examples with the first similarity degree greater than or equal to the first preset degree as the reference examples in the first reference example set to obtain the first reference example.

[0243] Alternatively, each reference example can also be sorted according to the first similarity degree, and the reference examples in the first preset number of positions can be determined as the reference examples in the first reference example set to obtain the first reference example.

[0244] The example thought chain information can be the specific analysis content of analyzing the example session information to obtain the example sales stage. The example thought chain information can improve the interpretability of the example sales stage and facilitate the user to clearly understand the specific reason for the finally determined example sales stage.

[0245] Specifically, the specific content of the example thought chain information can be as follows:

[0246] "Farthest Sales Status of User": "Facilitated" "Result Analysis": "According to the analysis of the conversation content, in sentences

[13] -

[25] , the butler recommended Product X and explained the features of the product. The user asked in

[16] whether additional insurance could be added at any time, indicating that the user was interested in the product. The customer service further explained the features of the product and information related to additional insurance in

[17] -

[22] . However, the user did not ask further or indicate that they would purchase, so the deal was not closed. In the subsequent conversation, sentences

[29] -

[73] , the customer service further explained and recommended Product Y, and the user also had positive interactions. However, the user did not finally show a definite intention to purchase or make a purchase. Therefore, the farthest sales status completed by the user is facilitated. During the entire conversation process, the customer service made product recommendations, designed solutions, and tried to facilitate the sale, but the user did not finally complete the deal."

[0247] Among them, the content in 【】 in the specific content of the above sample thought chain information indicates the sentence number in the text information.

[0248] Step 312: Extract the customer service session information of the customer service included in the session information.

[0249] The customer service session information is the information output by the customer service to the customer service. In the embodiments of this specification, the method for extracting the customer service session information of the customer service is similar to the method for extracting the user session information of the user, so it will not be elaborated here.

[0250] In the embodiments of this specification, the steps of extracting the customer service session information and the user session information can be carried out synchronously.

[0251] Step 314: Use the sales action recognition model to determine the customer service sales action sequence corresponding to the customer service session information.

[0252] The sales action recognition model can be a model that can recognize the sales actions of the customer service based on the customer service session information. The sales action recognition model can be a neural network model or a large model.

[0253] The customer service session information usually includes multiple customer service conversation sentences. The sales action recognition model can be used to recognize the sales actions for each customer service conversation sentence respectively, and determine the sales action corresponding to each customer service conversation sentence. Then, sort the sales actions corresponding to each conversation sentence in the order of the customer service conversation sentences, and merge the consecutive same conversation intents to obtain the customer service sales action sequence.

[0254] Step 316: Retrieve the target reference sample similar to the customer service sales action sequence from the first reference sample set.

[0255] In the embodiments of this specification, the reference examples may further include example customer service sales action sequences. Among them, the example customer service sales action sequences are obtained by the sales action recognition model based on the example customer service conversation information.

[0256] The second similarity degree between the customer service sales action sequence and the example customer service sales action sequences of each reference example in the first reference example set can be calculated using a similarity algorithm. Then, the reference examples with the second similarity degree greater than or equal to the second preset degree are determined as the target reference examples.

[0257] Alternatively, each reference example is sorted according to the second similarity degree, and the reference examples in the first second preset positions are determined as the target reference examples.

[0258] Step 318: Use the target reference example and the conversation information to obtain an enhanced prompt; the enhanced prompt includes the target reference example and the conversation information.

[0259] The target reference example retrieved from the external knowledge base and the conversation information can be embedded into a preset prompt template together to obtain an enhanced prompt. Since the enhanced prompt includes knowledge for identifying the sales stage of the business product or products in the same major category as the business product, the enhancement of the prompt can be achieved.

[0260] Step 320: Provide the enhanced prompt to the large model, and use the large model to obtain the sales stage corresponding to the conversation information and the chain-of-thought information for describing the analysis process of obtaining the sales stage from the conversation information.

[0261] Since the enhanced prompt includes knowledge for identifying the sales stage of the business product or products in the same major category as the business product, inputting the enhanced prompt into the large model can enable the large model to generate a more accurate response with a stronger correlation to the identification of the sales stage of the business product, while avoiding pre-training the large model with a large number of samples, thereby avoiding manual annotation of a large number of samples, reducing the manual annotation cost, and improving the efficiency. And since the enhanced prompt also includes the chain-of-thought information for the analysis process of obtaining the sales stage based on the conversation information, the large model can output the chain-of-thought information about the sales stage of the business product according to the enhanced prompt, improving the interpretability of determining the sales stage.

[0262] For ease of understanding, an example of an enhanced prompt is provided in the embodiments of this specification, which is specifically as follows:

[0263] ## Persona

[0264] Suppose you are a professional insurance sales customer service, and you need to complete the task according to the task description, materials provided, and requirements. ## Conversation between the user and the customer service:

[0265] 【0】Customer service: Hello, esteemed user. I am your exclusive insurance customer service. Two high-quality savings insurance products have been launched on Abao: 1. Product B (with dividends): Suitable for long-term savings plans, withdrawals can be made midway, and the estimated average annual rate can reach 3.4%+; 2. Product C: Suitable for medium- to long-term savings plans, with 100% principal protection and guaranteed returns according to the contract, and an average annual increase of about 3.44% at maturity; You can directly reply with a number to obtain exclusive product deduction benefits~

[0266] 【1】Customer service: Hello, ma'am. I'm Zhu. I'm calling because you previously inquired about Product C. You thought the time was too long at that time. Now we have a new D life insurance product called D. The good thing about this product is that if it is paid in a lump sum, it only takes 3 years.

[0267] 【2】User: Hmm.

[0268] 【3】Customer service: The principal is recovered. For example, if you pay 10,000 yuan in a lump sum, after 3 years, its cash value will exceed 10,000 yuan. Some customers may hold it for a long time, but not that long. This product may have a more suitable return for them.

[0269] 【4】User: Well, what will it be like after signing? Will there be a cashback or something?

[0270] 【5】Customer service: Oh, this is also D life insurance. There are three ways to receive: either surrender the policy to receive the cash value, apply for a partial reduction in insurance to receive a part, or take out a policy loan.

[0271] 【6】User: Send it to me and let me see.

[0272] 【7】Customer service: Ah, okay, okay. I'll send it to you. Take a look.

[0273] 【8】Customer service:

Tool

[0274] 【9】Customer service: A*6xAAABBBCCCDDDEEEFFFF

[0275]

[10] Customer service: Hello, you previously inquired about D life insurance. This product D life insurance is underwritten by M Bank Life Insurance Co., Ltd. The highlight of the product is that the principal is recovered in 3 years with a lump sum payment, which is suitable for customers who pay a lump sum and hold it for a medium- to long-term. If you consider applying for insurance, I can do a return calculation for you.

[0276] ##User Completes Sales Status Definition

[0277] 1. Opening Service: The user takes the initiative to speak or the butler starts the conversation, attracting the user to speak based on the user's characteristics.

[0278] 2. Requirement Mining: The user takes the initiative to state their needs or answers the butler's questions to state their specific needs. If it is just a response to the butler's need questions, it does not count as completing the requirement mining status.

[0279] 3. Product Recommendation: The user shows interest in the insurance products recommended by the butler, including but not limited to Stable Profit, Dividend-Increasing Whole Life Insurance, etc., and responds by asking for details about the relevant products.

[0280] 4. Solution Design: The user shows interest in the personalized design solutions provided by the butler, including various designs such as the insured amount and the insurance method, and gives corresponding responses.

[0281] 5. Facilitation: The user agrees to the facilitation actions such as the benefits and market cycles actively sent by the butler.

[0282] 6. Transaction: The user successfully purchases insurance.

[0283] ##The Furthest Sales Status Completed by the Current Dialogue User

[0284] Product Recommendation

[0285] ##Task Description

[0286] 1. Based on the content of the current conversation between the user and the butler, and the furthest sales status completed by the current dialogue user, gradually analyze the conversation content and give the corresponding result analysis. The furthest sales status is defined as the later status in the above sales status sequence. For example, if the user has completed three sales statuses: opening service, product recommendation, and solution design, then the furthest sales status he can reach is solution design.

[0287] 2. First, identify the sales status completed by the butler in the conversation process, and based on this, judge whether the user has a clear response. If so, it is considered that the user has completed this sales status; if not, it means that the user has not completed it.

[0288] 3. Secondly, according to the status completed by the user, combined with the above status sequence, obtain the furthest sales status that the user can complete.

[0289] 4. Finally, based on the furthest sales status reached by the user, add the original text paragraph description.

[0290] 5. Please note that the sales status completed by the user and the sales status completed by the butler are not exactly the same. Do not return the sales status completed by the butler.

[0291] 6. The returned result format is strictly in JSON, including two fields: the user's farthest sales status and result analysis.

[0292] ## Reference Case

[0293] 【0】Customer Service: Hey, hello, can you hear me? Well, I'm the insurance butler who connects with you on Platform A.

[0294] 【1】User: Go ahead.

[0295] 【2】Customer Service: Ah, hello. Well, the thing is, I'm calling today to sync two things with you. I also noticed that you purchased Insurance E and health services. Specifically, the F insurance in Insurance E, right?

[0296] 【3】Customer Service: In the future, I'll be the insurance butler connecting with you on Platform A. If you have any policy questions, you can contact me at any time. I can provide one-on-one service.

[0297] 【4】Customer Service: Second, I'd like to know if you've had any policy issues during the recent policy management process. I can give you a comprehensive and detailed explanation.

[0298] 【5】User: Haven't looked at it after buying.

[0299] 【6】Customer Service: Ah, haven't looked at it. It doesn't matter...

[0300] 【7】Customer Service:

Tool

[0301] 【8】Customer Service: https: / / aaa.bbb.com / cc_insbutler / d / e / A*f

[0302] 【9】Customer Service:

Product Recommendation

[0303] 1. Product G

[0304] 2. Product H

[0305] (There's also a benefit of using peace beans to offset the insurance premium if you apply now. For details, ask me and I'll show you how to check.)

[0306] Based on the above content, the result returned by the large model is:

[0307] {"User's furthest sales status": "Opening service",

[0308] "Result analysis": "From the conversation between the current user and the butler, we gradually analyzed and found that: in sentences 【0】-【1】, the butler performed the opening service sales action and the user responded, so the user completed the opening service sales status; in sentences 【2】-【5】, the butler conducted need exploration for the user, and the user replied that they haven't looked after buying, indicating that the user did not express specific needs, so it is considered that the user did not complete the need exploration sales status; in sentences 【7】-【9】, the butler made a product recommendation and the user did not give any reply, so it is considered that the user did not complete the product recommendation sales status. Therefore, the furthest sales status completed by the user is #Opening service#"}

[0309] In the embodiments of this specification, the "

Tool

Tool

Tool

Tool

Tool

[0310] Similarly, in the embodiments of this specification, the "[Tool] Product Recommendation Card: Product G|Product H (with dividends)" and "https: / / aaa.bbb.com / cc_insbutler / d / e / A*f" in the conversation information between the user and the customer service can be prompt messages sent by the server to the customer service terminal device based on the existing conversation information between the user and the customer service. Among them, the "[Tool] Product Recommendation Card: Product G|Product H (with dividends)" can be a pre-created link to a detailed introduction of Product G and Product H. When the customer service clicks on the "[Tool] Product Recommendation Card: Product G|Product H (with dividends)", it can jump to a page containing the detailed introduction of Product G and Product H. "https: / / aaa.bbb.com / cc_insbutler / d / e / A*f" can be a pre-created link to the statistical information of the endowment insurance for a recent period; by clicking on the "https: / / aaa.bbb.com / cc_insbutler / d / e / A*f" link, the customer service can jump to a page with the statistical information of the endowment insurance for a recent period. The customer service can directly send the "[Tool] Product Recommendation Card: Product G|Product H (with dividends)" and "https: / / aaa.bbb.com / cc_insbutler / d / e / A*f" to the user, or can introduce products or perform benefit calculations for the user based on the content in the jump pages of the "[Tool] Product Recommendation Card: Product G|Product H (with dividends)" and "https: / / aaa.bbb.com / cc_insbutler / d / e / A*f".

[0311] Based on the same idea, the embodiments of this specification also provide a device corresponding to the above method. Figure 4 It is a schematic structural diagram of a processing device for conversation information provided by the embodiments of this specification. As Figure 4 shown, the device may include:

[0312] An acquisition module 402, configured to acquire the conversation information between the customer service and the user regarding the introduction of business products.

[0313] A retrieval module 404, configured to retrieve a target reference example similar to the conversation information from a sample library; the target reference example includes sample conversation information, the corresponding sample sales stage of the sample conversation information, and sample thinking chain information; the sample thinking chain information is used to describe the analysis process of obtaining the sample sales stage from the sample conversation information.

[0314] A prompt word module 406, configured to obtain an enhanced prompt word by using the target reference example and the conversation information; the enhanced prompt word includes the target reference example and the conversation information.

[0315] A result generation module 408 is configured to provide the enhanced prompt to a large model, and use the large model to obtain the sales stage corresponding to the session information and the thought chain information for describing the analysis process of obtaining the sales stage from the session information.

[0316] Based on Figure 4 For the device of, embodiments of this specification also provide some specific implementation manners of this method, which will be described below.

[0317] The sample library includes a plurality of reference samples in different sales stages.

[0318] Optionally, Figure 4 The device in may further include:

[0319] A selection module is configured to, for any one of the plurality of reference samples, select a thought chain sample with the same sales stage as the reference sample from the thought chain database; the thought chain database includes at least one thought chain sample in different sales stages; one thought chain sample includes target sample session information, the target sales stage corresponding to the target sample session information, and target sample thought chain information for representing the analysis process of obtaining the target sales stage from the target sample session information;

[0320] A thought chain generation module is configured to provide the thought chain sample and the any one reference sample to the large model, and use the large model to obtain the sample thought chain information of the any one reference sample;

[0321] A first storage module is configured to store the sample thought chain information into the sample library.

[0322] Optionally, Figure 4 The device in may further include:

[0323] A first extraction module is configured to extract the user session information of the user included in the session information.

[0324] A first determination module is configured to use an intent recognition model to determine the user intent sequence corresponding to the user session information.

[0325] The retrieval module 404 may specifically include:

[0326] A first retrieval unit is configured to retrieve a target reference sample similar to the user intent sequence from the sample library according to the user intent sequence.

[0327] Optionally, the user session information includes multiple user session sentences.

[0328] The first determination module may specifically include:

[0329] A first determination unit, configured to use the intent recognition model to determine the conversation intent corresponding to each user conversation sentence.

[0330] A user intent sequence acquisition unit, configured to sort the conversation intents corresponding to each user conversation sentence according to the order of each user conversation sentence in the user conversation information, so as to obtain the user intent sequence of the user.

[0331] Optionally, the first retrieval unit may specifically be configured to:

[0332] Calculate a first similarity degree between the user intent sequence and the sample user intent sequences of each reference sample in the sample library; the sample user intent sequence is obtained by performing intent recognition on the conversation information of the sample user in the reference sample by using the intent recognition model.

[0333] Determine the reference samples with the first similarity degree greater than or equal to a first preset degree as target reference samples similar to the user intent sequence.

[0334] Alternatively, sort the respective reference samples according to the first similarity degree, and determine the reference samples located in the first preset number of positions as target reference samples similar to the user intent sequence.

[0335] Optionally, the user intent sequence acquisition unit may specifically be configured to:

[0336] Sort the conversation intents corresponding to each user conversation sentence according to the order of each user conversation sentence in the user conversation information, so as to obtain an initial user intent sequence.

[0337] If the first conversation intent included in the initial user intent sequence is the same as the adjacent second conversation intent, then delete the first conversation intent or the second conversation intent to obtain the user intent sequence.

[0338] Optionally, optionally, Figure 4 The device in may further include:

[0339] A second extraction module, configured to extract the customer service conversation information of the customer service included in the conversation information.

[0340] A second determination module, configured to use a sales action recognition model to determine the customer service sales action sequence corresponding to the customer service conversation information.

[0341] The retrieval module 404 may specifically include:

[0342] A second retrieval unit, configured to retrieve a target reference example similar to the customer service sales action sequence from the example library according to the customer service sales action sequence.

[0343] The customer service session information includes multiple customer service conversation sentences.

[0344] Specifically, the second determination module may include:

[0345] A second determination unit, configured to use the sales action recognition model to determine the sales actions corresponding to each of the customer service conversation sentences.

[0346] A customer service sales action sequence acquisition unit, configured to sort the sales actions corresponding to each of the customer service conversation sentences in the order of the customer service conversation sentences in the customer service session information, to obtain the customer service sales action sequence.

[0347] Optionally, the second retrieval unit may specifically be configured to:

[0348] Calculate a second similarity degree between the customer service sales action sequence and the example customer service sales action sequences of each reference example in the example library; the example customer service sales action sequences are obtained by performing sales action recognition on the session information of the example customer service in the reference example using the sales action recognition model.

[0349] Determine the reference examples with the second similarity degree greater than or equal to a second preset degree as the target reference examples similar to the customer service sales action sequence.

[0350] Alternatively, sort the reference examples according to the second similarity degree, and determine the reference examples in the first second preset positions as the target reference examples similar to the customer service sales action sequence.

[0351] Optionally, the retrieval module 404 may specifically be configured to:

[0352] Select a first set of reference examples similar to the user intention sequence from the example library according to the user intention sequence corresponding to the session information; the user intention sequence is obtained by performing intention recognition on the session information of the user included in the session information using an intention recognition model.

[0353] Retrieve the target reference examples similar to the customer service sales action sequence from the first set of reference examples according to the customer service sales action sequence corresponding to the session information.

[0354] Optionally, the business product introduction includes an introduction to selling insurance products.

[0355] Based on the same idea, the embodiments of this specification also provide a device corresponding to the above method.

[0356] Figure 5 It is a schematic structural diagram of a processing device for session information provided by the embodiments of this specification. As Figure 5 shown, the device 500 may include:

[0357] At least one processor 510; and,

[0358] A memory 530 communicatively connected to the at least one processor; wherein,

[0359] The memory 530 stores instructions 520 executable by the at least one processor 510, and when the instructions are executed by the at least one processor 510, the at least one processor 510 can:

[0360] Obtain the session information of the customer service and the user for the business product introduction.

[0361] Retrieve a target reference example similar to the session information from the example library; the target reference example includes example session information, the corresponding example sales stage of the example session information, and example thought chain information; the example thought chain information is used to describe the analysis process of obtaining the example sales stage from the example session information.

[0362] Use the target reference example and the session information to obtain an enhanced prompt word; the enhanced prompt word includes the target reference example and the session information.

[0363] Provide the enhanced prompt word to the large model, and use the large model to obtain the sales stage corresponding to the session information and the thought chain information for describing the analysis process of obtaining the sales stage from the session information.

[0364] Based on the same idea, the embodiments of this specification also provide a computer-readable medium corresponding to the above method. A computer-readable instruction is stored on the computer-readable medium, and the computer-readable instruction can be executed by a processor to implement the above method for processing session information:

[0365] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for Figure 3 the device shown, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0366] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data that have been authorized by the user or fully authorized by all parties. Moreover, the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of the relevant regions, and corresponding operation entrances are provided for users to choose to authorize or refuse.

[0367] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to circuit structures such as diodes, transistors, switches, etc.) or software improvements (improvements to method flows). However, with the development of technology, many method flow improvements today can be regarded as direct improvements to hardware circuit structures. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement to a method flow cannot be implemented using a hardware entity module. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is an integrated circuit whose logical function is determined by the user programming the device. Designers can program themselves to "integrate" a digital system onto a single PLD, without having to ask a chip manufacturer to design and fabricate a dedicated integrated circuit chip. Moreover, today, instead of manually fabricating integrated circuit chips, this programming is mostly implemented using "logic compiler" software, which is similar to the software compilers used in program development and writing. The original code before compilation also has to be written in a specific programming language, which is called a Hardware Description Language (HDL), and there is not just one type of HDL, but many, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones currently are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also be aware that by simply performing a little logical programming on the method flow using the above-mentioned several hardware description languages and programming it into an integrated circuit, it is easy to obtain the hardware circuit that implements the logical method flow.

[0368] The controller can be implemented in any suitable manner. For example, the controller can take the form of, for example, a microprocessor or a processor and a computer-readable medium storing computer-readable program code (such as software or firmware) executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of the controller include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in the form of pure computer-readable program code, it is entirely possible to logically program the method steps to enable the controller to be implemented in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. to achieve the same function. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be regarded as the structures within the hardware component. Or even, the devices for implementing various functions can be regarded as either software modules for implementing the method or structures within the hardware component.

[0369] The systems, devices, modules, or units illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0370] For the convenience of description, when describing the above devices, they are described separately as various units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0371] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program code.

[0372] The present invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each flow and / or block of the flowchart illustrations and / or block diagrams, and combinations of flows and / or blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions executed by the processor of the computer or other programmable data processing apparatus create means for implementing the functions specified in the flowchart Figure 1 one flow or more flows and / or blocks Figure 1 one block or more blocks.

[0373] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means that implement the functions specified in the flowchart Figure 1 one flow or more flows and / or blocks Figure 1 one block or more blocks.

[0374] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart Figure 1 one flow or more flows and / or blocks Figure 1 one block or more blocks.

[0375] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0376] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of computer-readable media.

[0377] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0378] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0379] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0380] The present application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0381] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various modifications and variations can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. A method for processing session information, comprising: Obtain conversation information between customer service and users regarding business product introductions; Retrieving target reference samples similar to the session information from a sample library; The target reference sample includes sample session information, a sample sales stage corresponding to the sample session information, and sample thought chain information; the sample thought chain information is used to describe the analysis process of obtaining the sample sales stage from the sample session information; Using the target reference sample and the conversation information, an enhanced prompt word is obtained; The enhanced prompt word includes the target reference sample and the conversation information; The enhanced prompt words are provided to the big model, and the big model is used to obtain the sales stage corresponding to the session information and the thought chain information used to describe the analysis process of obtaining the sales stage from the session information.

2. The method according to claim 1, wherein the sample library includes a plurality of reference samples at different sales stages; Before retrieving target reference samples similar to the session information from the sample library, the method further includes: For any reference example among the multiple reference examples, a thinking chain example of the same sales stage as the reference example is selected from a thinking chain database; the thinking chain database includes at least one thinking chain example in different sales stages; one thinking chain example includes target sample session information, a target sales stage corresponding to the target sample session information, and target sample thinking chain information for indicating an analysis process of obtaining the target sales stage from the target sample session information; Providing the thought chain sample and any reference sample to the big model, and using the big model to obtain the sample thought chain information of any reference sample; The sample thought chain information is saved in the sample library.

3. The method according to claim 1, before retrieving a target reference sample similar to the session information from a sample library, further comprising: extracting user session information of the user contained in the session information; Determine the user intent sequence corresponding to the user session information by using an intent recognition model; The retrieving a reference sample similar to the session information from the sample library specifically includes: According to the user intention sequence, target reference samples similar to the user intention sequence are retrieved from the sample library.

4. The method according to claim 3, wherein the user session information includes a plurality of user session statements; The determining the user intention sequence corresponding to the user session information by using the intention recognition model specifically includes: Determine the conversation intention corresponding to each of the user conversation statements by using the intention recognition model; The conversation intentions corresponding to the user conversation statements are sorted according to the order of the user conversation statements in the user conversation information to obtain a user intention sequence of the user.

5. The method according to claim 3, wherein the step of retrieving target reference examples similar to the user intention sequence from the example library according to the user intention sequence comprises: Calculating a first similarity between the user intention sequence and the sample user intention sequence of each reference sample in the sample library; The sample user intention sequence is obtained by performing intention recognition on the session information of the sample user in the reference sample using the intention recognition model; Determining a reference sample whose first similarity degree is greater than or equal to a first preset degree as a target reference sample similar to the user intention sequence; Alternatively, the reference samples are sorted according to the first similarity level, and the reference samples located in the front of a first preset number of bits are determined as target reference samples similar to the user intention sequence.

6. The method according to claim 4, wherein the session intentions corresponding to the user session statements are sorted according to the order of the user session statements in the user session information to obtain the user intention sequence of the user, specifically comprising: Sorting the conversation intentions corresponding to the user conversation statements according to the order of the user conversation statements in the user conversation information to obtain an initial user intention sequence; If a first session intention included in the initial user intention sequence is the same as an adjacent second session intention, the first session intention or the second session intention is deleted to obtain the user intention sequence.

7. The method according to claim 1, before retrieving a target reference sample similar to the session information from a sample library, further comprising: Extracting customer service session information of the customer service contained in the session information; Determine the customer service sales action sequence corresponding to the customer service conversation information by using a sales action recognition model; The retrieving a target reference sample similar to the session information from the sample library specifically includes: According to the customer service sales action sequence, target reference examples similar to the customer service sales action sequence are retrieved from the example library.

8. The method of claim 7, wherein the customer service conversation information includes a plurality of customer service conversation statements; The using the sales action recognition model to determine the customer service sales action sequence corresponding to the customer service conversation information specifically includes: Determine the sales action corresponding to each of the customer service conversation sentences using the sales action recognition model; The sales actions corresponding to the customer service conversation sentences are sorted according to the order of the customer service conversation sentences in the customer service conversation information to obtain the customer service sales action sequence.

9. The method according to claim 7, wherein the step of retrieving target reference examples similar to the customer service sales action sequence from the example library according to the customer service sales action sequence comprises: Calculating a second similarity between the sales action sequence of the customer service and the sample customer service sales action sequence of each reference sample in the sample library; The sample customer service sales action sequence is obtained by using the sales action recognition model to perform sales action recognition on the conversation information of the sample customer service in the reference sample; Determine the reference sample whose second similarity degree is greater than or equal to the second preset degree as a target reference sample similar to the sales action sequence of the customer service; Alternatively, the reference samples are sorted according to the second similarity level, and the reference samples located in the first second preset position are determined as target reference samples that are similar to the sales action sequence of the customer service.

10. The method of claim 8, wherein the sales actions corresponding to the customer service conversation statements are sorted according to the order of the customer service conversation statements in the customer service conversation information to obtain the customer service sales action sequence, specifically comprising: Sorting the sales actions corresponding to the customer service conversation statements according to the order of the customer service conversation statements in the customer service conversation information to obtain an initial sales action sequence; If the first sales action included in the initial sales action sequence is the same as the adjacent second sales action, the first sales action or the second sales action is deleted, and the customer service sales action sequence is entered.

11. The method according to claim 1, wherein the step of retrieving target reference samples similar to the session information from a sample library comprises: According to the user intention sequence corresponding to the session information, selecting a first reference sample set similar to the user intention sequence from the sample library; The user intention sequence is obtained by performing intention recognition on the session information of the user contained in the session information using an intention recognition model; According to the customer service sales action sequence corresponding to the session information, the target reference examples similar to the customer service sales action sequence are retrieved from the first reference example set.

12. The method of claim 1, wherein the business product introduction comprises an introduction to sell insurance products.

13. A device for processing session information, comprising: The acquisition module is used to obtain the conversation information between the customer service and the user regarding the introduction of business products; A retrieval module is used to retrieve a target reference sample similar to the session information from a sample library; the target reference sample includes sample session information, a sample sales stage corresponding to the sample session information, and sample thought chain information; the sample thought chain information is used to describe the analysis process of obtaining the sample sales stage from the sample session information; A prompt word generation module, used to obtain enhanced prompt words by using the target reference sample and the conversation information; The enhanced prompt word includes the target reference sample and the conversation information; A result generation module is used to provide the enhanced prompt words to the big model, and use the big model to obtain the sales stage corresponding to the session information and the thought chain information used to describe the analysis process of obtaining the sales stage from the session information.

14. A session information processing device, comprising: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: Obtain conversation information between customer service and users regarding business product introductions; Retrieving a target reference sample similar to the session information from a sample library; the target reference sample includes sample session information, a sample sales stage corresponding to the sample session information, and sample thought chain information; the sample thought chain information is used to describe the analysis process of obtaining the sample sales stage from the sample session information; Using the target reference sample and the conversation information, an enhanced prompt word is obtained; the enhanced prompt word includes the target reference sample and the conversation information; The enhanced prompt words are provided to the big model, and the big model is used to obtain the sales stage corresponding to the session information and the thought chain information used to describe the analysis process of obtaining the sales stage from the session information.

15. A computer-readable medium having computer-readable instructions stored thereon, wherein the computer-readable instructions can be executed by a processor to implement the method for processing session information according to any one of claims 1 to 12.

Images