Pre-sales consultation auxiliary method and system based on chat robot
By training a large language model and a LoRa model based on Transformer, combined with historical dialogue records and a QA database, we solved the problems that chatbots could not answer effectively, thus improving the customer consultation experience and purchase conversion rate.
Patent Information
- Application Number
- CN202311381966.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-24
- Publication Date
- 2026-05-12
AI Technical Summary
Existing intelligent customer service chatbots are unable to effectively answer customer questions and have a curt tone, resulting in a poor customer consultation experience.
By training a large language model and a LoRa model based on Transformer, and combining the historical dialogue records between customer service representatives and customers, a dialogue model and an evaluation model are generated. The QA database is then used to rank the answers, thereby improving the chatbot's auxiliary effect.
It improved the chatbot's assistance capabilities, met consumer needs, and increased purchase conversion rates, achieving a tone and professional responses similar to customer service.
Smart Images
Figure CN122019689A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of language processing, and in particular to a pre-sales consultation assistance method and system based on a chatbot. Background Technology
[0002] Existing intelligent customer service chatbots focus primarily on answering customer questions based on existing knowledge or rules. If the customer's question is not in the chatbot's knowledge base, it cannot provide an effective answer. Furthermore, most existing intelligent chatbots suffer from a stiff and unengaging tone, resulting in a very poor customer consultation experience. Summary of the Invention
[0003] The purpose of this invention is to provide a pre-sales consultation assistance method and system based on chatbots, which can improve the assistance effect of chatbots.
[0004] To achieve the above objectives, the present invention provides the following solution:
[0005] A pre-sales consultation assistance method based on a chatbot includes:
[0006] Based on the historical dialogue records between all customer service representatives and customers, a large language model based on Transformer is trained to determine the generated dialogue model.
[0007] Based on the historical dialogue records between each customer service representative and the customer, a LoRa model is trained to obtain the LoRa model for each customer service representative.
[0008] Based on the historical dialogue records between all customer service representatives and customers, and the corresponding purchase results, a classification model is trained to obtain an evaluation model; the purchase result refers to whether the customer makes a purchase.
[0009] Based on the original knowledge base, a QA database is determined using a Transformer-based large language model; the QA database is used to store feature vectors of questions and answers.
[0010] Obtain customer questions; and use a generated dialogue model, a LoRa model, and a QA database to determine three corresponding answers.
[0011] Based on the three responses, a generative dialogue model is used to determine multiple responses.
[0012] The evaluation model is used to rank multiple answers, and the ranking results are sent to the corresponding customer service representatives.
[0013] Optionally, the evaluation model takes the dialogue records between customer service and customers as input and the probability of customer purchase as output.
[0014] Optionally, determining the QA database based on the original knowledge base using a Transformer-based large language model specifically includes:
[0015] The original knowledge base was broken down into multiple text segments;
[0016] Using a Transformer-based large language model, generate corresponding questions and answers for each piece of text;
[0017] Extract feature vectors from questions and answers;
[0018] The feature vectors of questions and answers are stored in a vector database to obtain the QA database.
[0019] Optionally, the step of using an evaluation model to rank multiple answers and sending the ranking results to the corresponding customer service representatives specifically includes:
[0020] Customer service staff will determine the reply content based on the sorting results and send it to the customer.
[0021] Optionally, when the customer service representative's response is not in the sorting results, the customer's question and the customer service representative's answer are added to the corresponding customer service representative's and customer's historical dialogue records, and the LoRa model is retrained.
[0022] Optionally, when the customer service representative's response is not in the sorting results, the feature vectors of the customer's question and the customer service representative's answer are extracted and the QA database is updated.
[0023] A pre-sales consultation assistance system based on a chatbot includes:
[0024] The dialogue model generation module is used to train a large language model based on Transformer based on the historical dialogue records between all customer service representatives and customers, and to determine the generated dialogue model.
[0025] The Lora model training module is used to train the Lora model based on the historical dialogue records between each customer service representative and the customer, so as to obtain the Lora model for each customer service representative.
[0026] The evaluation model determination module is used to train a classification model based on the historical dialogue records between all customer service representatives and customers and the corresponding purchase results, thereby obtaining the evaluation model; the purchase result refers to whether the customer makes a purchase.
[0027] The QA database determination module is used to determine the QA database based on the original knowledge base and using a Transformer-based large language model; the QA database is used to store feature vectors of questions and answers.
[0028] The identification module is used to obtain the customer's question; and uses the generated dialogue model, LoRa model, and QA database to determine the corresponding three answers respectively;
[0029] The secondary recognition module is used to determine multiple answers based on the three responses using a generated dialogue model;
[0030] The sorting module is used to rank multiple answers using an evaluation model and send the ranking results to the corresponding customer service representatives.
[0031] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:
[0032] The present invention provides a pre-sales consultation assistance method and system based on chatbots. It utilizes the historical dialogue records between all customer service representatives and customers to determine the generation of a dialogue model, uses the historical dialogue records between each customer service representative and customer to train a LoRa model, and determines the evaluation model and QA database. That is, by combining a professional knowledge base, historical dialogue records, dialogue context, and historical purchase records, it achieves the effect of meeting consumer needs while improving the purchase conversion rate, thereby improving the assistance effect of chatbots. Attached Figure Description
[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0034] Figure 1 This is a schematic diagram of a pre-sales consultation assistance method based on a chatbot provided by the present invention.
[0035] Figure 2 This is a schematic diagram illustrating the principle of a pre-sales consultation assistance method based on a chatbot provided by the present invention. Detailed Implementation
[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0037] The purpose of this invention is to provide a pre-sales consultation assistance method and system based on chatbots, which can improve the assistance effect of chatbots.
[0038] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0039] like Figure 1 and Figure 2 As shown, the pre-sales consultation assistance method based on a chatbot provided by the present invention includes:
[0040] S101: Based on the historical dialogue records between all customer service representatives and customers, train a large language model based on Transformer to determine the generated dialogue model. The generated dialogue model G achieves an effect similar to the vocabulary used by customer service representatives in their daily interactions.
[0041] S102, based on the historical dialogue records between each customer service representative and the customer, train a LoRa model to obtain the LoRa model for each customer service representative. That is, based on the historical dialogue records between customer service representative i and the customer, train a LoRa model to obtain the LoRa model Li for customer service representative i.
[0042] S103, Based on the historical dialogue records between all customer service representatives and customers and the corresponding purchase results, a classification model is trained to obtain an evaluation model F; the purchase result refers to whether the customer makes a purchase; the evaluation model F takes the dialogue records between customer service representatives and customers as input and the probability of the customer making a purchase as output.
[0043] S104. Based on the original knowledge base, a QA database is determined using a Transformer-based large language model; the QA database is used to store feature vectors of questions and answers.
[0044] S104 specifically includes:
[0045] The original knowledge base was broken down into multiple text segments.
[0046] We use a large language model based on Transformer to generate corresponding questions and answers for each piece of text.
[0047] Extract feature vectors from questions and answers.
[0048] The feature vectors of questions and answers are stored in a vector database to obtain the QA database.
[0049] For example, for a popular science article related to acne, first break it down into several paragraphs (each paragraph introduces only one small topic, such as "acne onset time", "acne-affected population", "acne treatment methods", "acne pathological manifestations"). For each paragraph, several questions and answers can be generated. For example, for a paragraph discussing the population affected by acne, the question "What is the age distribution of people with acne?" can be generated, and the answer "Acne mainly occurs in adolescents" can be generated, with the answer content coming from the original article.
[0050] S105, obtain the customer's question; and use the generated dialogue model G, the LoRa model Li, and the QA database to determine the corresponding three answers.
[0051] Based on the customer's answer or question, a possible answer, denoted as A1, is generated using the generative dialogue model G; and based on the current customer service representative i, A2 is generated using Li; and the corresponding answer, denoted as A3, is retrieved using the QA database.
[0052] S106. Based on the three responses, use the generative dialogue model G to determine multiple responses.
[0053] Sort A1, A2, and A3, or concatenate the three, and regenerate several responses using the generative dialogue model G, denoted as Bi.
[0054] S107, use the evaluation model to rank multiple answers; and send the ranking result Ci to the corresponding customer service representative.
[0055] S107 specifically includes:
[0056] Customer service staff will determine the reply content based on the sorting results and send it to the customer.
[0057] Several answers are returned to customer service. Customer service can choose one of the answers Ci or modify the answer in Ci according to their own ideas, and then send it to the customer. Alternatively, if customer service is not satisfied with all the answers, they can edit the reply themselves.
[0058] When the customer service representative's response is not in the sorted results, the customer's question and the customer service representative's answer are added to the corresponding customer service representative's and customer's historical dialogue records, and the LoRa model is retrained.
[0059] When the customer service representative's response is not in the sorted results, extract the feature vectors of the customer's question and the customer service representative's answer, and update the QA database.
[0060] This invention utilizes a professional knowledge base to answer questions while also taking into account the customer service representative's tone of voice, enabling seamless customer service intervention; at the same time, it ranks the model based on historical consumption results, thereby improving pre-sales effectiveness.
[0061] Corresponding to the methods in the above embodiments, the pre-sales consultation assistance system based on a chatbot provided by the present invention includes:
[0062] The dialogue model generation module is used to train a large language model based on Transformer based on the historical dialogue records between all customer service representatives and customers, and to determine the generated dialogue model.
[0063] The LoRa model training module is used to train a LoRa model based on the historical dialogue records between each customer service representative and the customer, thus obtaining a LoRa model for each customer service representative.
[0064] The evaluation model determination module is used to train a classification model based on the historical dialogue records between all customer service representatives and customers, as well as the corresponding purchase results, to obtain the evaluation model; the purchase result refers to whether the customer makes a purchase.
[0065] The QA database determination module is used to determine the QA database based on the original knowledge base and using a Transformer-based large language model; the QA database is used to store feature vectors of questions and answers.
[0066] The identification module is used to obtain customer questions and determine three corresponding answers using a generated dialogue model, a LoRa model, and a QA database.
[0067] The secondary recognition module is used to determine multiple responses based on the three responses using a generated dialogue model.
[0068] The sorting module is used to rank multiple answers using an evaluation model and send the ranking results to the corresponding customer service representatives.
[0069] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.
[0070] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A pre-sales consultation assistance method based on a chatbot, characterized in that, include: Based on the historical dialogue records between all customer service representatives and customers, a large language model based on Transformer is trained to determine the generated dialogue model. Based on the historical dialogue records between each customer service representative and the customer, a LoRa model is trained to obtain the LoRa model for each customer service representative. Based on the historical dialogue records between all customer service representatives and customers, and the corresponding purchase results, a classification model is trained to obtain an evaluation model; the purchase result refers to whether the customer makes a purchase. Based on the original knowledge base, a QA database is determined using a Transformer-based large language model; the QA database is used to store feature vectors of questions and answers. Obtain customer questions; and use a generated dialogue model, a LoRa model, and a QA database to determine three corresponding answers. Based on the three responses, a generative dialogue model is used to determine multiple responses. The evaluation model was used to rank the multiple responses. The sorting results will then be sent to the corresponding customer service representative.
2. The pre-sales consultation assistance method based on a chatbot according to claim 1, characterized in that, The evaluation model takes the dialogue records between customer service representatives and customers as input and the probability of a customer making a purchase as output.
3. The pre-sales consultation assistance method based on a chatbot according to claim 1, characterized in that, The process of determining the QA database based on the original knowledge base and using a large language model based on Transformer specifically includes: The original knowledge base was broken down into multiple text segments; Using a Transformer-based large language model, generate corresponding questions and answers for each piece of text; Extract feature vectors from questions and answers; The feature vectors of questions and answers are stored in a vector database to obtain the QA database.
4. The pre-sales consultation assistance method based on a chatbot according to claim 1, characterized in that, The evaluation model is used to rank multiple answers; The sorting results will then be sent to the corresponding customer service representatives, including: Customer service staff will determine the reply content based on the sorting results and send it to the customer.
5. A pre-sales consultation assistance method based on a chatbot according to claim 4, characterized in that, When the customer service representative's response is not in the sorted results, the customer's question and the customer service representative's answer are added to the corresponding customer service representative's and customer's historical dialogue records, and the LoRa model is retrained.
6. The pre-sales consultation assistance method based on a chatbot according to claim 4, characterized in that, When the customer service representative's response is not in the sorted results, extract the feature vectors of the customer's question and the customer service representative's answer, and update the QA database.
7. A pre-sales consultation assistance system based on a chatbot, characterized in that, include: The dialogue model generation module is used to train a large language model based on Transformer based on the historical dialogue records between all customer service representatives and customers, and to determine the generated dialogue model. The Lora model training module is used to train the Lora model based on the historical dialogue records between each customer service representative and the customer, so as to obtain the Lora model for each customer service representative. The evaluation model determination module is used to train a classification model based on the historical dialogue records between all customer service representatives and customers and the corresponding purchase results, thereby obtaining the evaluation model; the purchase result refers to whether the customer makes a purchase. The QA database determination module is used to determine the QA database based on the original knowledge base and using a Transformer-based large language model; the QA database is used to store feature vectors of questions and answers. The identification module is used to obtain the customer's question; and uses the generated dialogue model, LoRa model, and QA database to determine the corresponding three answers respectively; The secondary recognition module is used to determine multiple answers based on the three responses using a generated dialogue model; The sorting module is used to rank multiple answers using an evaluation model and then send the ranking results to the corresponding customer service representatives.