Dialogue training method and system based on artificial intelligence

Abstract

The invention relates to the technical field of data processing, and discloses a dialogue training method and system based on artificial intelligence, and the method comprises the steps: carrying out the analysis of pre-obtained employee sales data through a preset employee capability evaluation model, and obtaining an employee sales level; analyzing the employee sales data through a preset customer intention recognition model, and constructing a customer type map; generating a plurality of dialogue scenes through a preset dialogue generation model in combination with the employee sales level and the customer type map; by analyzing the sales ability change condition of the employees in each dialogue training process, the difficulty of the next dialogue scene is adjusted by using a preset dialogue optimization model, so that dynamic dialogue training is realized; the dialogue scene is generated according to the employee sales level and the customer type map, so that dialogue practice is more targeted, the difficulty of the dialogue scene is dynamically adjusted, training is more targeted and effective, and the employees are helped to quickly master dialogue skills.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a dialogue training method and system based on artificial intelligence. Background Art

[0002] In the traditional training process for salesperson's conversation skills, manual training is adopted, which takes a lot of time. Moreover, the unified training settings ignore the differences in sales capabilities among different salespersons and cannot provide targeted training for each employee. The existing technologies cannot generate personalized content in combination with the ability levels of employees and different types of customers, and lack a mechanism to dynamically adjust the difficulty of the dialogue scenario according to the changes in employees' capabilities during the dialogue practice process, thus unable to effectively help employees gradually improve their sales capabilities.

[0003] Based on the above problems existing in the prior art, in order to solve at least one of the above problems, the present application proposes a dialogue training method and system based on artificial intelligence. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the main object of the present invention is to provide a dialogue training method and system based on artificial intelligence, which can effectively solve the problems in the background art. The specific technical solutions of the present invention are as follows: A dialogue training method based on artificial intelligence, comprising: Analyzing pre-acquired employee sales data through a preset employee ability evaluation model to obtain an employee sales level; Analyzing the employee sales data through a preset customer intention recognition model to construct a customer type map; Combining the employee sales level and the customer type map, and generating multiple dialogue scenarios through a preset dialogue generation model, wherein the multiple dialogue scenarios are generated corresponding to different customer types; After the end of a dialogue practice process, analyzing the employee sales intention and customer intention in the previous dialogue process through a preset intention analysis model to construct an employee sales intention network and a customer intention network; Performing hierarchical matching on the employee sales intention network and the customer intention network through a preset hierarchical matching model to obtain a matching degree value, wherein the hierarchical matching includes coarse-grained matching, fine-grained matching, and dynamic weight allocation; Adjusting the difficulty of the next dialogue scenario through a preset dialogue optimization model according to the relationship between the matching degree value and a preset matching degree threshold to achieve dynamic dialogue training.

[0005] Specifically, the analyzing pre-acquired employee sales data through a preset employee ability evaluation model to obtain an employee sales level includes: Extract features from the sales conversation text in the pre-acquired employee sales data through a preset text feature extraction model to generate a text feature vector; Analyze the behavior logs in the employee sales data to obtain a time-series behavior feature vector; Analyze the transaction records in the employee sales data to obtain a transaction success rate; Combine the text feature vector, the time-series behavior feature vector, and the transaction success rate, and evaluate the sales ability of employees through a preset employee ability evaluation model to obtain an employee ability matrix; Determine the employee sales level according to the employee ability matrix.

[0006] Specifically, analyzing the employee sales data through a preset customer intention recognition model to construct a customer type map, including: Extract the corresponding customer conversation records from the employee sales data; According to the customer conversation records, analyze the customer type and the intention of the customer in each conversation through a preset customer intention recognition model to construct a customer type map, where the customer type in the customer type map is the root node, the intention is the child node, and the conversion probability between different intentions is the weight of the edge.

[0007] Specifically, according to the customer conversation records, analyze the customer type and the intention of the customer in each conversation through a preset customer intention recognition model to construct a customer type map, including: Analyze the intention of the customer in each conversation through a preset customer intention recognition model according to the customer conversation records to obtain an intention recognition result; Identify the corresponding customer type according to the customer conversation records through a preset customer type recognition model; Calculate the conversion probability between different intentions of the corresponding customer type through the intention recognition result of the customer in each conversation; Use the customer type as the root node, different intentions as the child nodes, and the conversion probability between the corresponding intentions as the weight of the edge to construct a customer type map.

[0008] Specifically, combine the employee sales level and the customer type map, and generate multiple conversation scenarios through a preset conversation generation model, where the multiple conversation scenarios are generated corresponding to different customer types, including: Match in the customer type map according to the employee sales level through a preset mapping rule to obtain a corresponding customer type sub-map; Generate multiple conversation scenarios through a preset conversation generation model according to the customer type sub-map combined with the employee's real-time conversation content.

[0009] Specifically, based on the customer type sub-graph and the real-time conversation content of the employee, multiple conversation scenarios are generated through a preset conversation generation model, including: Initialize and select an initial intention from the customer type sub-graph as the starting point of the conversation; Based on the employee's sales level and the starting point of the conversation, generate the first simulated response of the customer through a preset conversation generation model; Combine the response result of the employee to the first simulated response and analyze the transfer probability of the customer's intention; Select the intention with the highest transfer probability as the next customer intention, and generate the next simulated response of the customer through a preset conversation generation model; Repeat the response generation process until the conversation ends.

[0010] Specifically, through a preset hierarchical matching model, the employee sales intention network and the customer intention network are hierarchically matched to obtain a matching degree value, including: Analyze the coarse-grained matching degree between the employee sales intention network and the customer intention network to obtain a coarse-grained matching degree value; Analyze the fine-grained matching degree between the employee sales intention network and the customer intention network to obtain a fine-grained matching degree value; According to the preset business objectives, weight the coarse-grained matching degree value and the fine-grained matching degree value through a preset hierarchical matching model to obtain corresponding weights; Perform weighted calculation on the coarse-grained matching degree value and the fine-grained matching degree value with the corresponding weights to obtain a matching degree value.

[0011] Specifically, analyzing the coarse-grained matching degree between the employee sales intention network and the customer intention network to obtain a coarse-grained matching degree value includes: Count the occurrence times of various intentions in the employee sales intention network and the customer intention network respectively; Calculate the first matching ratio according to the corresponding relationship between various intentions in the employee sales intention network and the customer intention network and the occurrence times; Assign corresponding weights to various intentions through a preset first dynamic weight distribution model; Perform weighted calculation by combining the first matching ratio and the corresponding weights to obtain a coarse-grained matching degree value.

[0012] Specifically, analyzing the fine-grained matching degree between the employee sales intention network and the customer intention network to obtain a fine-grained matching degree value includes: Through a preset intention division model, divide each intention in the employee sales intention network and the customer intention network to obtain intention sub-categories; Calculate the second matching ratio based on the occurrence times of each intention sub-category in the employee sales intention network and the customer intention network, and in combination with the corresponding relationship between the intention sub-categories. Assign corresponding weights to each intention sub-category through a preset second dynamic weight assignment model. Perform weighted calculation by combining the second matching ratio and the corresponding weights to obtain the fine-grained matching degree value.

[0013] An artificial intelligence-based dialogue training system for implementing the artificial intelligence-based dialogue training method described above, including: An employee sales level analysis module that analyzes pre-obtained employee sales data through a preset employee ability evaluation model to obtain the employee sales level. A customer type map construction module that analyzes the employee sales data through a preset customer intention recognition model to construct a customer type map. A dialogue generation module that combines the employee sales level and the customer type map to generate multiple dialogue scenarios through a preset dialogue generation model, where the multiple dialogue scenarios are generated corresponding to different customer types. A network construction module that, after the end of a dialogue practice process, analyzes the employee sales intention and the customer intention during the previous dialogue process through a preset intention analysis model to construct an employee sales intention network and a customer intention network. A network matching module that performs hierarchical matching on the employee sales intention network and the customer intention network through a preset hierarchical matching model to obtain a matching degree value, where the hierarchical matching includes coarse-grained matching, fine-grained matching, and dynamic weight assignment. A dialogue optimization module that adjusts the difficulty of the next dialogue scenario through a preset dialogue optimization model according to the relationship between the matching degree value and a preset matching degree threshold to achieve dynamic dialogue training.

[0014] Compared with the prior art, the present invention has the following beneficial effects: This application constructs a customer type map based on multi-source sales data, generates dialogue scenarios according to the employee sales level and the customer type map, making the dialogue practice more targeted. Employees with different ability levels can be matched to dialogue scenarios suitable for their own levels, and at the same time, simulate practice for different types of customers to improve the employees' ability to handle various customers. At the same time, by analyzing the matching degree of the employee sales intention and the customer intention during the dialogue practice process, dynamically adjust the difficulty of the next dialogue scenario to ensure that employees can gradually improve their sales ability, enhance the training effect, make the training more targeted and effective, and thus help employees master dialogue skills faster. Description of the Drawings

[0015] Figure 1It is the workflow diagram of a dialogue training method based on artificial intelligence in Embodiment 1 of the present invention; Figure 2 It is the schematic diagram of the construction of the customer type map in Embodiment 1 of the present invention; Figure 3 It is the schematic diagram of the hierarchical matching process of the employee sales intention network and the customer intention network in Embodiment 1 of the present invention; Figure 4 It is the structural schematic diagram of a dialogue training system based on artificial intelligence in Embodiment 2 of the present invention. Detailed implementation manners

[0016] To make the above objects, features and advantages of the present invention more obvious and understandable, the following will describe the detailed implementation manners of the present invention in conjunction with the drawings of the specification.

[0017] In the following description, many specific details are set forth to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar promotions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0018] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure or characteristic that can be included in at least one implementation manner of the present invention. The "in one embodiment" appearing in different places in this specification does not all refer to the same embodiment, nor is it a separate or alternative embodiment that excludes other embodiments.

[0019] Embodiment 1 This embodiment provides a dialogue training method based on artificial intelligence, as Figure 1 shown, the dialogue training method based on artificial intelligence includes: S101. Analyze the pre-obtained employee sales data through a preset employee ability evaluation model to obtain the employee sales level; S102. Analyze the employee sales data through a preset customer intention recognition model to construct a customer type map; S103. Combine the employee sales level and the customer type map, and generate multiple dialogue scenarios through a preset dialogue generation model, where the multiple dialogue scenarios are generated corresponding to different customer types; S104. After the end of a dialogue practice process, analyze the employee sales intention and the customer intention in the previous dialogue process through a preset intention analysis model to construct an employee sales intention network and a customer intention network; S105. Through a preset hierarchical matching model, perform hierarchical matching on the employee sales intention network and the customer intention network to obtain a matching degree value, where the hierarchical matching includes coarse-grained matching, fine-grained matching, and dynamic weight allocation; S106. According to the relationship between the matching degree value and a preset matching degree threshold, adjust the difficulty of the next dialogue scenario through a preset dialogue optimization model to achieve dynamic dialogue training.

[0020] In this embodiment, corresponding dialogue scenarios are generated according to the sales ability levels of employees and different customer types to assist employees in conducting sales conversations and helping employees improve their sales conversation skills. At the same time, during the training process, according to the improvement of employees' sales abilities, the dialogue difficulty of each sales scenario is dynamically adjusted to adapt to the personal abilities of employees and improve the training effect. Compared with the traditional fixed-mode training scenarios, the dialogue training scenarios intelligently generated by this solution can adapt to the differences of different employees and provide personalized training scenarios for each employee, thereby quickly improving the dialogue skills of employees.

[0021] In this embodiment, first, employee sales data is collected, including multi-source data such as sales conversation texts, behavior logs, and transaction records. The sales conversation texts record the content of the communication between employees and customers; the behavior logs record the behavior information such as the time and frequency of the interaction between employees and customers; the transaction records reflect the actual results of sales, such as the volume of transactions and the turnover. The sales ability of employees can be analyzed and evaluated based on the employee sales data. The employee sales data is analyzed through a preset employee ability evaluation model to obtain the employee sales level, and personalized generation of dialogue scenarios is performed for each employee according to the employee sales level, making the generated dialogue scenarios more in line with the actual sales level of employees and avoiding the limitations of single-dimensional evaluation.

[0022] Specifically, by analyzing different types corresponding to different customers, diverse dialogue scenarios are provided. The customer types are analyzed based on the message reply situations of customers in the employee sales data, and a customer type map is constructed. The customer intentions and types in the employee sales data are identified through a preset customer intention recognition model. According to the customer types and the changes in customer intentions during each dialogue, the conversion probability between different intentions corresponding to the customer types is calculated. Using the customer types as the root nodes, different intentions as the child nodes, and the conversion probability between the corresponding intentions as the edge weights, a customer type map is constructed. By constructing the customer type map, the intention distribution of different customer types and the conversion relationship between intentions can be intuitively displayed, providing rich customer information for generating targeted dialogue scenarios, helping employees better understand customers, and improving the effect of sales communication.

[0023] Furthermore, based on the employee sales level obtained through analysis and corresponding to the customer type that is suitable for the employee, and combined with the characteristics of different customer types and intention conversion relationships in the customer type map, targeted dialogue scenarios are generated using the preset dialogue generation model to provide each employee with a personalized dialogue scenario. The generated dialogue takes into account both the employee's ability level and the customer's actual type and behavior pattern, which can effectively improve the quality and effectiveness of employee dialogue practice and enhance employees' ability to deal with different customers.

[0024] At the same time, after each dialogue practice, according to the complete dialogue practice process record, the pre-set intention analysis model, specifically a natural language processing model based on deep learning, is used to input the sorted dialogue data into the model in turn. The model recognizes the intention of each round of employee and customer speech. For example, the model determines that the employee's intention in a certain round of speech is "to emphasize product advantages" and the customer's response intention is "to raise price objections". According to the identified employee intention and customer intention, the employee sales intention network and the customer intention network are constructed respectively. For the employee sales intention network, the identified employee intention is used as a node. During the dialogue process, the employee transitions from the "product introduction" intention to the "resolving customer questions" intention, and a directed edge is created between the two nodes to indicate the change of intention; the customer intention network is constructed in the same way. By constructing the employee sales intention network and the customer intention network, the matching degree between the employee response and the customer intention can be quickly analyzed, and then the changes in the employee's sales ability can be analyzed.

[0025] Specifically, through hierarchical matching, the matching degree between the employee sales intention network and the customer intention network is analyzed. Hierarchical matching includes three steps: coarse-grained matching, fine-grained matching and dynamic weight allocation. Coarse-grained matching can consider the overall matching between employee and customer intention categories from a macro level, while fine-grained matching performs matching analysis on subcategories within the intention or specific content details. Dynamic weight allocation adjusts the importance of coarse-grained and fine-grained matching results in the final matching degree calculation according to business goals and actual conversation situations. Through this hierarchical matching method, the one-sidedness of single-dimensional matching can be avoided, and the degree of fit between employee sales intentions and customer intentions can be comprehensively and accurately evaluated to obtain a comprehensive matching value, which provides a quantitative basis for adjusting the difficulty of conversation scenarios and helps to achieve more reasonable dynamic conversation training.

[0026] Furthermore, according to the relationship between the calculated matching degree value and the preset matching degree threshold, which is set based on the employee's ability improvement goal and the actual business situation. When the matching degree value is higher than the matching degree threshold, it indicates that the employee performed well in the previous conversation practice and is capable of handling the current difficulty, so the difficulty of the next conversation scenario can be appropriately increased to continuously challenge the employee's ability; when the matching degree is lower than the matching degree threshold, it indicates that the employee has deficiencies at the current difficulty level, and the difficulty of the next conversation scenario needs to be reduced so that the employee can gradually improve their ability. The relevant parameters for generating the conversation scenario are adjusted according to the comparison result through the preset conversation optimization model to achieve dynamic difficulty adjustment. Through the dynamic adaptive adjustment of the conversation practice difficulty, the practice difficulty is flexibly adjusted according to the actual ability performance of the employee, avoiding the employee being frustrated due to excessive difficulty or being unable to effectively improve their ability due to too low difficulty. Continuously provide an effective practice environment for the employee, promote the steady improvement of the employee's sales ability, improve the pertinence and effectiveness of the conversation training, and ultimately improve the overall level of the enterprise sales team.

[0027] This application constructs a customer type map based on multi-source sales data, generates conversation scenarios according to the employee's sales level and the customer type map, making the conversation practice more targeted. Employees with different ability levels can be matched to conversation scenarios suitable for their own levels, and at the same time, simulate practice for different types of customers to improve the employee's ability to handle various customers. At the same time, by analyzing the matching degree between the sales intention and the customer intention of the employee during the conversation practice, the difficulty of the next conversation scenario is dynamically adjusted to ensure that the employee can gradually improve their sales ability, enhance the training effect, make the training more targeted and effective, and thus help the employee master the conversation skills faster.

[0028] Furthermore, analyzing the pre-acquired employee sales data through the preset employee ability evaluation model to obtain the employee's sales level, including: S201. Extract features from the sales conversation text in the pre-acquired employee sales data through the preset text feature extraction model to generate a text feature vector; S202. Analyze the behavior log in the employee sales data to obtain a time-series behavior feature vector; S203. Analyze the transaction record in the employee sales data to obtain the transaction success rate; S204. Combine the text feature vector, the time-series behavior feature vector and the transaction success rate, and evaluate the employee's sales ability through the preset employee ability evaluation model to obtain an employee ability matrix; S205. Determine the employee's sales level according to the employee ability matrix.

[0029] In this embodiment, the sales ability of employees is evaluated in real time. By analyzing employees' sales data from multiple dimensions and considering various factors, an accurate evaluation result of employees' sales ability is obtained. First, through a preset text feature extraction model, the sales dialogue text in employees' sales data is subjected to feature extraction, and the key information in the text is converted into vector form, reflecting the characteristics of employees in aspects such as language use in the dialogue, product introduction methods, and conversation skills in interacting with customers, providing data support in the text dimension for evaluating employees' sales ability.

[0030] Exemplarily, the sales dialogue text is cleaned to remove noise data therein, such as irrelevant special characters, garbled codes, etc. At the same time, the text is segmented, splitting continuous sentences into individual words for easy feature extraction; for example, for the sentence "The performance of our product is very excellent", after segmentation, words such as "we", "this", "product", "of", "performance", "very", "excellent" are obtained; the text feature extraction model is specifically a neural network model based on word vectors, which is trained on a large amount of text data. The model predicts a word based on the context, and in this process, each word is mapped to a low-dimensional vector space, making the distance between words with similar semantics in the vector space relatively close; for example, during the training process, the two words "product" and "commodity" with similar semantics will have relatively close positions of their corresponding word vectors in the space; in this way, each word in the sales dialogue text is converted into a corresponding word vector, and then a weighted aggregation operation is performed on these word vectors to obtain the text feature vector of the entire sales dialogue text; by converting the unstructured sales dialogue text into structured vector data, key ability points such as employees' language style and professional knowledge expression in communication can be captured, providing an important basis for comprehensively evaluating employees' sales ability.

[0031] Specifically, analyze the behavioral logs in the employee sales data. The behavioral logs record various behaviors of employees' interactions with customers, such as the first contact time, subsequent follow-up time, communication methods (phone, email, in-person interview, etc.), and communication frequencies, etc. First, organize these log data to ensure the accuracy and integrity of the data, and sort the data in chronological order. According to the organized behavioral log data, extract the time interval features, communication frequency features, and behavioral pattern features. Obtain the time interval features by calculating the time interval from the first contact between the employee and the customer to the first follow-up, and the time interval between each follow-up. Obtain the communication frequency features by counting the number of communications between the employee and the customer within a certain time period. Obtain the behavioral pattern features by analyzing the changing patterns of the communication methods adopted by the employee at different time points. Combine these features to obtain the time-series behavioral feature vector, which can reflect the employee's customer management ability, sales rhythm grasping ability, etc., and complement each other with indicators such as the text feature vector and the transaction success rate, making the evaluation of the employee's sales ability more comprehensive.

[0032] At the same time, analyze the transaction records in the employee sales data and calculate the transaction success rate. From the transaction records in the employee sales data, screen out the records of successful transactions and total transactions. The successful transaction records can be determined according to the signs of transaction completion (such as payment arrival, contract signing, etc.). Count the number of successful transactions and the number of total transactions, and calculate the transaction success rate. The transaction success rate can reflect the employee's ability to convert sales opportunities into actual results. Incorporate the transaction success rate into the employee ability evaluation system to make the evaluation results more in line with the actual situation.

[0033] Furthermore, conduct a comprehensive analysis of the extracted text feature vector, time-series behavioral feature vector, and transaction success rate. Through a preset employee ability evaluation model, specifically a pre-trained neural network model, comprehensively process these multi-dimensional data. The model learns the relationships between each feature and the employee's sales ability based on a large amount of historical data, thereby comprehensively evaluating the employee's sales ability and outputting the employee ability matrix. Through the fusion and comprehensive analysis of multi-source data, the sales ability of employees can be comprehensively and accurately evaluated. According to the calculated employee ability matrix, determine the employee's sales level through a certain mapping rule. By converting the complex employee ability matrix into a simple and intuitive employee sales level, employees can be classified and managed and receive targeted training. Employees with different sales levels can be matched with different difficulty levels and types of dialogue practice scenarios to improve the efficiency of training and management.

[0034] Exemplarily, according to the actual needs of the enterprise and the characteristics of the sales business, mapping rules between the employee ability matrix and the employee sales level are formulated. The scores of each dimension in the employee ability matrix are weighted and summed according to the allocated weights to obtain a comprehensive score. Then, different grade intervals are divided based on this comprehensive score. For example, a comprehensive score between 80 and 100 is the senior sales level, between 60 and 79 is the intermediate sales level, and below 60 is the junior sales level.

[0035] Further, analyzing the employee sales data through a preset customer intention recognition model to construct a customer type map includes: S301: Extract the corresponding customer conversation records from the employee sales data; S302: According to the customer conversation records, analyze the customer type and the intention of the customer in each conversation through a preset customer intention recognition model to construct a customer type map. Among them, the customer type in the customer type map is the root node, the intention is the sub-node, and the conversion probability between different intentions is the weight of the edge.

[0036] Further, analyzing the customer type and the intention of the customer in each conversation through a preset customer intention recognition model according to the customer conversation records to construct a customer type map includes: S401: According to the customer conversation records, analyze the intention of the customer in each conversation through a preset customer intention recognition model to obtain an intention recognition result; S402: According to the customer conversation records, identify the corresponding customer type through a preset customer type recognition model; S403: Calculate the conversion probability of the corresponding customer type between different intentions through the intention recognition result of the customer in each conversation; S404: Use the customer type as the root node, different intentions as the sub-nodes, and the conversion probability between the corresponding intentions as the weight of the edge to construct a customer type map.

[0037] In this embodiment, customer conversation records are extracted from employee sales data, the customer intentions and behaviors are analyzed, and according to the customer conversation records, the customer intentions are identified through a preset customer intention recognition model. The customer intention recognition model is specifically a convolutional neural network model. The extracted customer conversation records are preprocessed such as data cleaning, noise removal, and word segmentation, and the preprocessed customer conversation records are input into the pre-trained model. The model calculates and outputs the probability distribution of the intention category corresponding to each conversation, and selects the intention category with the highest probability as the recognition result. For example, if the model outputs that the probability of a certain conversation belonging to the "product consultation" intention is 0.7, the probability of belonging to the "purchase intention inquiry" intention is 0.2, and the probability of belonging to other intentions is 0.1, then the intention recognition result of this conversation is determined to be "product consultation"; by analyzing the customer intentions, the customer needs can be deeply understood, and then the customer types can be determined.

[0038] Specifically, according to the customer conversation records, the customer types are identified through a preset customer type recognition model. Different customers show different behavior patterns and language styles in the conversation. The preset customer type recognition model classifies the customers by learning these features. The model is specifically a random forest model, which is trained based on historical customer data to mine the common features of different types of customers to obtain a pre-trained customer type recognition model. Multiple features are extracted from the customer conversation records for customer type recognition, such as the conversation length, the number of professional words used, the types of questions (open-ended questions, closed-ended questions), etc.; the extracted customer type features are input into the customer type recognition model, and the model outputs the corresponding customer type. By classifying the customers, it helps to provide diverse conversation scenarios.

[0039] Furthermore, during the process of customers communicating with employees, their intentions will change. By counting the number of conversions between various intentions for different customer types and combining the total number of conversations, the intention conversion probability is calculated to reflect the behavior patterns and demand change trends of customers at different stages; each record in the customer conversation records is analyzed to determine the customer type, the starting intention, and the target intention, and the corresponding intention conversion probability is calculated. For example, for a conversation record belonging to the "potential customer" type, the starting intention is "product consultation", and the target intention becomes "purchase intention inquiry" after communication. For each customer type, for each pair of starting intention and target intention, the conversion number is divided by the total number of times the starting intention appears to obtain the conversion probability between these two intentions for the corresponding customer type. For example, in the "potential customer" type, the number of conversions from "product consultation" to "purchase intention inquiry" is 20 times, and the total number of times the "product consultation" intention appears is 100 times, then the conversion probability is 20÷100 = 0.2.

[0040] Specifically, such as Figure 2As shown, the analyzed customer types are used as the root nodes, and various intents are used as child nodes and connected to the corresponding customer type root nodes. The conversion probabilities between intents are used as the weights of the connecting edges to form a directed weighted graph, constructing a customer type map. The customer behavior and intent information are visually displayed through the customer type map, and corresponding dialogue scenarios can be quickly generated based on the customer type map.

[0041] Furthermore, by combining the employee's sales level and the customer type map, multiple dialogue scenarios are generated through a preset dialogue generation model. Among them, the multiple dialogue scenarios are generated corresponding to different customer types, including: S501. According to the employee's sales level, match in the customer type map through a preset mapping rule to obtain the corresponding customer type sub-graph; S502. According to the customer type sub-graph and the employee's real-time dialogue content, generate multiple dialogue scenarios through a preset dialogue generation model.

[0042] In this embodiment, according to the sales level of each employee, the corresponding customer types suitable for each employee are matched in the constructed customer type map. Employees with different ability levels are suitable for dealing with different types of customers. According to the customers they are good at dealing with, customers they are not good at dealing with are matched for them to improve the ability of each employee to adapt to different types of customers. According to this rule, the customer types and their related intent relationships that are adapted to the employee's ability are screened out in the customer type map through a preset mapping rule, forming a customer type sub-graph, providing customer simulation scenarios with appropriate difficulty and type for employees at different ability levels, avoiding employees being unable to effectively improve their ability due to facing overly simple or complex customer simulation scenarios, improving the pertinence and effectiveness of dialogue practice, and promoting the gradual improvement of employees' sales ability.

[0043] Specifically, according to the customer types and their intent relationship information in the matched customer type sub-graph, combined with the employee's real-time dialogue content, the preset dialogue generation model is used to simulate the responses of customers under different intents, and the dialogue is gradually advanced according to the dialogue logic to generate multiple complete dialogue scenarios close to the actual sales scenario, enabling employees to be fully exercised in the simulation environment; the generated dialogue scenarios are both based on the real behavior patterns of customers and can be dynamically adjusted according to the real-time performance of employees, highly simulating the real sales scenario, enabling employees to better master dialogue skills in practice, improving the ability to deal with different customers and various dialogue situations, and enhancing the practicality and effect of sales training.

[0044] Furthermore, the generating of multiple dialogue scenarios by combining the customer type sub-graph and the employee's real-time dialogue content through a preset dialogue generation model includes: S601. Initialize and select an initial intent from the customer type sub-graph as the starting point of the dialogue; S602. Combine the starting point of the conversation with the employee's sales level, and generate the first simulated response of the customer through a preset conversation generation model; S603. Analyze the transfer probability of the customer's intention in combination with the employee's response to the first simulated response; S604. Select the intention with the highest transfer probability as the next customer intention, and generate the next simulated response of the customer through a preset conversation generation model; S605. Repeat the response generation process until the conversation ends.

[0045] In this embodiment, an initial intention is initially selected in the customer type sub-graph to obtain the starting point of the conversation. The occurrence frequencies of each intention node in the customer type sub-graph are counted. For example, the customer type sub-graph is traversed using a graph algorithm, and the number of times each intention node is connected is recorded. In a specific customer type sub-graph, it is found through statistics that the "product consultation" intention node has the most connections, indicating that in the interaction between such customers and employees, it is more common to start the conversation from product consultation. Therefore, the "product consultation" intention is preferentially selected as the initial conversation starting point; by reasonably selecting the initial intention, it is possible to simulate the common opening methods of customers in real sales scenarios, enabling employees to quickly enter the simulated conversation scenario, and the diverse selection strategies help improve the employees' ability to handle different customer openings.

[0046] Specifically, according to the employee's sales level and the selected starting point of the conversation, the first simulated response result is generated through a preset conversation generation model. The employee's sales level corresponds to different ability levels. When communicating with customers, the response methods and content expectations of customers are different for employees at different levels. For employees with a lower level, the customer response result is relatively gentle; for employees with a higher level, the customer response result can be more incisive, increasing the difficulty of the conversation practice for employees with a higher level. The conversation generation model is trained based on a large amount of historical sales conversation data and learns the language patterns under different employee sales levels and customer intentions; inputting the employee's sales level and the starting point of the conversation into the model, the model can generate the first simulated response of the customer that conforms to the scenario based on this information; the generated simulated response takes into account the actual ability level of the employee, making the conversation practice more targeted. When facing a customer response that matches their own ability, employees can better adapt to and improve their communication skills. At the same time, the conversation generation model generates simulated responses based on historical data, ensuring the authenticity and reasonableness of the responses and enhancing the credibility of the simulated scenario.

[0047] Exemplarily, for employees at the primary sales level, the starting point of the conversation is the "product consultation" intention. After concatenating the vector representing the primary sales level and the "product consultation" intention vector, they are input into the trained dialogue generation model. After internal calculations, the first simulated reply from the model for the customer is "I am more concerned about the price range of your products. Can you introduce it first?"

[0048] Specifically, after the employee responds to the scenario, the employee's response to the customer's simulated reply will affect the customer's intention direction. By analyzing the employee's reply content and combining the existing intention conversion probability information in the customer type subgraph, infer the customer's next intention and its probability distribution, and select the intention with the highest transfer probability as the customer's next intention. According to the next intention, continue to generate the next customer simulated reply result through the preset dialogue generation model, and dynamically analyze the customer's intention transfer based on the employee's real-time reply, making the dialogue scenario more interactive and realistic. Employees can feel the impact of their replies on the customer's intention in the simulation, so as to learn how to guide the customer's intention and improve the initiative and effectiveness of sales communication.

[0049] At the same time, continuously repeat the process of generating customer replies, analyzing intention transfer, and generating customer replies again to simulate the multi-round interaction in a real sales conversation. Set the dialogue end condition. When the end condition is met, it is considered that a complete dialogue scenario construction is completed. By constructing different dialogue scenarios multiple times, comprehensively exercise the employee's sales dialogue ability; through multi-round dialogues to simulate real sales scenarios, comprehensively exercise the employee's dialogue ability in the entire sales process, including links such as opening, communication promotion, handling objections, and closing deals; the diverse dialogue scenario generation methods can enable employees to come into contact with the reactions and dialogue directions of more different types of customers, and improve the comprehensive ability of employees to handle complex sales scenarios.

[0050] Furthermore, the hierarchical matching of the employee sales intention network and the customer intention network through the preset hierarchical matching model to obtain the matching degree value includes: S701. Analyze the coarse-grained matching degree between the employee sales intention network and the customer intention network to obtain the coarse-grained matching degree value; S702. Analyze the fine-grained matching degree between the employee sales intention network and the customer intention network to obtain the fine-grained matching degree value; S703. According to the preset business objectives, allocate weights to the coarse-grained matching degree value and the fine-grained matching degree value through the preset hierarchical matching model to obtain the corresponding weights; S704. Perform weighted calculation on the coarse-grained matching degree value and the fine-grained matching degree value with the corresponding weights to obtain the matching degree value.

[0051] Such as Figure 3, in this embodiment, the matching degree between the employee sales intention network and the customer intention network is calculated. Through hierarchical matching, the coarse-grained matching degree value and the fine-grained matching degree value are calculated respectively, and the weight values of the coarse-grained matching degree value and the fine-grained matching degree value are dynamically allocated. First, analyze the coarse-grained matching degree between the employee sales intention network and the customer intention network. Coarse-grained matching examines the matching situation between the employee sales intention network and the customer intention network from a macroscopic perspective, paying attention to the overall categories of intentions. By counting the occurrence times of various intentions in the two networks and calculating the matching ratio based on their corresponding relationships, and then combining the preset weights, the coarse-grained matching degree value is finally obtained; by calculating the coarse-grained matching value, the matching situation between the employee and the customer intention can be quickly evaluated from the overall level, providing a macroscopic matching degree index.

[0052] Furthermore, analyze the fine-grained matching degree between the employee sales intention network and the customer intention network. Fine-grained matching delves into the interior of the intention, further divides each intention into sub-categories, calculates a more detailed matching ratio through counting the occurrence times of sub-categories and the corresponding relationships between sub-categories, and then combines their respective weights to obtain the fine-grained matching degree value; through fine-grained matching, the matching situation between the employee and the customer in terms of intention details can be deeply explored, and the ability performance of the employee in dealing with the specific demand details of the customer can be discovered, making the matching degree analysis more comprehensive and accurate.

[0053] At the same time, according to the business objectives, weight allocation is carried out on the calculated coarse-grained matching degree value and fine-grained matching degree value. Different business objectives have different emphases on coarse-grained matching and fine-grained matching in the sales process; for example, when the business focus is on quickly facilitating transactions, the matching degree of transaction-related intentions in coarse-grained matching is more critical; while when the business emphasizes the in-depth exploration and satisfaction of customer needs, the importance of the fine-grained matching degree is higher; the preset hierarchical matching model flexibly allocates weights to the coarse-grained matching degree value and the fine-grained matching degree value according to these business objectives. Through weight allocation, the matching degree calculation can be closely tailored to the current business needs of the enterprise, highlighting the key requirements for different aspects of the employee's sales ability in different business stages, and helping the enterprise to more effectively evaluate the employee's performance and optimize the sales strategy according to the actual business situation.

[0054] Exemplarily, when the business objective focuses on quickly facilitating transactions, the weight of the coarse-grained matching degree value can be set to 0.7, and the weight of the fine-grained matching degree value can be set to 0.3; when the business objective is to improve customer satisfaction and deeply explore customer needs, the weight of the coarse-grained matching degree value can be set to 0.4, and the weight of the fine-grained matching degree value can be set to 0.6.

[0055] Specifically, the calculated coarse-grained matching value and fine-grained matching value are weighted according to the assigned weights, and the intention matching at the macro and micro levels is comprehensively considered to obtain a matching value that fully reflects the degree of fit between the employee's sales intention and the customer's intention; avoiding the limitations of single-dimensional matching analysis. It can more accurately reflect the overall fit between the employee and the customer's intention in the sales conversation, and provide a more scientific and comprehensive basis for the difficulty adjustment of the conversation scene and the evaluation of employee capabilities.

[0056] Furthermore, the coarse-grained matching degree of the employee sales intention network and the customer intention network is analyzed to obtain a coarse-grained matching degree value, including: S801, counting the number of occurrences of each type of intention in the employee sales intention network and the customer intention network respectively; S802, calculating a first matching ratio according to the correspondence between the various intentions in the employee sales intention network and the customer intention network and the number of occurrences; S803, assigning corresponding weights to various types of intentions through a preset first dynamic weight assignment model; S804: Perform weighted calculation based on the first matching ratio and the corresponding weight to obtain a coarse-grained matching degree value.

[0057] In this embodiment, first, the number of occurrences of each type of intent in the employee sales intention network and the customer intention network is counted respectively, and the employee sales intention network and the customer intention network are traversed by a depth-first search (DFS) algorithm. During the traversal process, each time an intent node is accessed, it is checked whether the intent category already exists in the counting record. If not, a new record item is created and the count is initialized to 1; if it already exists, the corresponding count is increased by 1; based on the correspondence between the intents in the employee sales intention network and the customer intention network and the counted number of occurrences, the intent matching ratio is calculated to obtain a first matching ratio; by comparing the matching ratios of different corresponding intent groups, it can be found out which intent matches the employees perform better in and which ones need improvement, thereby providing direction for targeted training and strategy adjustments.

[0058] Specifically, the corresponding weights are assigned to various types of intentions through a preset first dynamic weight assignment model. Different intentions have different importance levels in the sales process. The first dynamic weight assignment model dynamically assigns weights to various types of intentions according to the degree of influence of the intention on the sales result. When calculating the coarse-grained matching degree value, the matching situation of important intentions can be highlighted to affect the overall matching degree; the matching ratios of various corresponding intentions calculated are weighted with their corresponding weights, and the matching situations of all corresponding intention groups are integrated to obtain a coarse-grained matching degree value that can comprehensively reflect the matching degree between the employee's sales intention and the customer's intention at the macro level; through the coarse-grained matching degree value, the matching situations of different intention groups and their importance can be comprehensively considered, and compared with the simple calculation of the matching ratio, it can more accurately evaluate the employee's ability to grasp the customer's main intention in the overall sales conversation.

[0059] Exemplarily, through the graph traversal algorithm, the employee sales intention network and the customer intention network are traversed respectively, and the occurrence times of each intention category are recorded during the traversal process; based on the logic and experience of the sales business, the corresponding relationship between intention categories is determined. For example, "product introduction" corresponds to "product consultation", "promotion and promotion" corresponds to "preference for discounts", etc.; then, for each group of corresponding intentions, its matching ratio is calculated; the calculation formula is: matching ratio = min(employee intention occurrence times, customer intention occurrence times) / max(employee intention occurrence times, customer intention occurrence times). For example, for "product introduction" (appearing 15 times on the employee side) and "product consultation" (appearing 12 times on the customer side), the matching ratio = 12 / 15 = 0.8; using the preset first dynamic weight assignment model, weights are assigned to various types of intentions according to the importance of different intentions to the sales result; for example, the "closing a deal" intention is crucial to the sales result, and the weight can be set to 0.7; the "product introduction" intention weight is set to 0.3. For each group of corresponding intentions, its matching ratio is multiplied by the corresponding weight, and then all the products are accumulated to obtain the coarse-grained matching degree value.

[0060] Furthermore, analyzing the fine-grained matching degree between the employee sales intention network and the customer intention network to obtain the fine-grained matching degree value includes: S901. Through a preset intention division model, each intention in the employee sales intention network and the customer intention network is divided to obtain intention sub-categories; S902. According to the occurrence times of each intention sub-category in the employee sales intention network and the customer intention network, combined with the corresponding relationship between intention sub-categories, calculate the second matching ratio; S903. Assign corresponding weights to each intention sub-category through a preset second dynamic weight assignment model; S904. Combine the second matching ratio and the corresponding weights for weighted calculation to obtain the fine-grained matching degree value.

[0061] In this embodiment, the fine-grained matching degree between the employee sales intention network and the customer intention network is analyzed, and the fine-grained matching degree value is calculated. First, the intentions in the employee sales intention network and the customer intention network are divided through a preset intention division model to obtain intention sub-categories. The preset intention division model is obtained based on an in-depth understanding of the sales business and a large amount of historical data analysis. By subdividing the intentions in the network, considering the complexity and diversity of the intentions in the sales conversation, and the problem that a single intention classification cannot fully display its details, the specific intention details expressed by the employee and the customer in the conversation can be more accurately analyzed through intention subdivision.

[0062] Exemplarily, the intention division model is specifically a clustering model based on machine learning. First, a large amount of text data containing intention descriptions is clustered and trained to learn the feature differences between different intention sub-categories, and then new intentions are divided. For example, a large amount of text data about product introduction is clustered into several different clusters, and each cluster corresponds to an intention sub-category. For example, function-related text is clustered into the "function introduction" cluster, and performance-related text is clustered into the "performance introduction" cluster; the information of each intention node in the employee sales intention network and the customer intention network is input into the intention division model, and the model outputs the corresponding intention sub-category; for example, the "product introduction" intention node in the employee sales intention network is divided into multiple intention sub-category nodes such as "function introduction" and "performance introduction" after being processed by the model.

[0063] Specifically, similar to the coarse-grained matching, the fine-grained matching calculates the matching ratio by comparing the occurrence times of the employee and customer intention sub-categories and combining the corresponding relationship between the intention sub-categories to obtain the second matching ratio; according to the different importance of different intention sub-categories in the sales process, a corresponding weight is assigned to each intention sub-category through a preset second dynamic weight allocation model. The second dynamic weight allocation model is specifically a decision tree model. The decision tree model is trained through a large amount of historical data to obtain a pre-trained second dynamic weight allocation model. Each type of intention sub-category is input into the model, and the model outputs the weight value corresponding to each intention sub-category. The calculated second matching ratios of various corresponding intention sub-categories are weighted with their corresponding weights to obtain a fine-grained matching degree value that can comprehensively reflect the matching degree between the employee sales intention and the customer intention at the micro level; by calculating the fine-grained matching degree, compared with the simple calculation of the matching ratio, it can more accurately evaluate the employee's ability to grasp the specific needs details of the customer in the overall sales conversation.

[0064] Exemplarily, through the intention division model, each intention in the employee sales intention network and the customer intention network is subdivided. The "product introduction" intention is subdivided into sub-categories such as "function introduction", "performance introduction", "material introduction", etc.; the "product consultation" intention is subdivided into sub-categories such as "function inquiry", "performance inquiry", "material inquiry", etc.; the occurrence times of each intention sub-category are respectively counted, and then, for each group of corresponding intention sub-categories, the matching ratio is calculated. For example, the "function introduction" sub-category of the employee appears 10 times, and the "function inquiry" sub-category of the customer appears 8 times, and the matching ratio = 8 / 10 = 0.8; through the second dynamic weight allocation model, weights are assigned to each intention sub-category according to the influence degree of the sub-category on the sales process. For some products, the "performance introduction" and "performance inquiry" sub-categories have a greater impact on the sales decision, and the weight is set to 0.4; the "material introduction" and "material inquiry" sub-categories have a weight of 0.2, etc.; multiply the matching ratio of each sub-category by the corresponding weight, and accumulate all the products to obtain the fine-grained matching degree value.

[0065] Embodiment 2

[0066] In this embodiment, as Figure 4 , a dialogue training system based on artificial intelligence is provided for implementing the above-mentioned dialogue training method based on artificial intelligence, including: An employee sales level analysis module analyzes pre-obtained employee sales data through a preset employee ability evaluation model to obtain the employee sales level; A customer type graph construction module analyzes the employee sales data through a preset customer intention recognition model to construct a customer type graph; A dialogue generation module combines the employee sales level and the customer type graph, and through a preset dialogue generation model, generates multiple dialogue scenarios, where the multiple dialogue scenarios are generated corresponding to different customer types; A network construction module analyzes the employee sales intention and the customer intention in the previous dialogue process through a preset intention analysis model after the end of a dialogue practice process to construct an employee sales intention network and a customer intention network; A network matching module performs hierarchical matching on the employee sales intention network and the customer intention network through a preset hierarchical matching model to obtain a matching degree value, where the hierarchical matching includes coarse-grained matching, fine-grained matching, and dynamic weight allocation; A dialogue optimization module adjusts the difficulty of the next dialogue scenario through a preset dialogue optimization model according to the relationship between the matching degree value and a preset matching degree threshold to achieve dynamic dialogue training.

[0067] In this embodiment, the employee sales level analysis module includes a data collection unit, a text feature extraction unit, a behavior log analysis unit, a transaction record analysis unit, and an employee ability evaluation unit. This module comprehensively evaluates the sales ability of employees, quantifies it into an employee sales level, enables the system to provide appropriate dialogue practice scenarios according to the actual level of employees, and promotes the improvement of employees' sales ability; the customer type graph construction module includes a dialogue record extraction unit, a customer intention recognition unit, an intention conversion probability calculation unit, and a graph construction unit. This module deeply analyzes the intentions and behavior patterns of customers in sales conversations, constructs a customer type graph, provides rich customer information for the dialogue generation module, thereby generating more practical and targeted dialogue scenarios, and also helps enterprises better understand customers and formulate precise sales strategies.

[0068] Specifically, the dialogue generation module includes a customer type sub-graph matching unit, a dialogue starting point selection unit, a simulated reply generation unit, an intention transfer analysis unit, and a dialogue scenario construction unit. This module combines the employee sales level and the customer type graph to generate multiple dialogue scenarios corresponding to different customer types, simulates real sales scenarios through the dialogue scenarios, provides diverse dialogue practice materials for employees, and helps employees improve their communication ability with different types of customers; the network construction module includes a data collection unit, an intention analysis model, and a network construction unit. This module constructs an employee sales intention network and a customer intention network through the analysis of each dialogue record after each dialogue practice, providing a basis for analyzing the matching degree between the two; the network matching module includes a coarse-grained matching unit, a fine-grained matching unit, a dynamic weight allocation unit, and a matching degree calculation unit. This module analyzes the matching degree between the employee sales intention network and the customer intention network through a preset hierarchical matching model, comprehensively considers coarse-grained matching, fine-grained matching, and dynamic weight allocation, and obtains an accurate matching degree value, providing a basis for adjusting the difficulty of dialogue scenarios; the dialogue optimization module includes a threshold comparison unit, a difficulty adjustment decision unit, and a dialogue optimization unit. This module adjusts the difficulty of the next dialogue scenario using a dialogue optimization model according to the relationship between the matching degree value and a preset matching degree threshold, realizing dynamic dialogue training and helping employees gradually improve their sales ability.

[0069] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A dialogue training method based on artificial intelligence, characterized in that: include: Analyze the pre-acquired employee sales data through the preset employee capability evaluation model to obtain the employee sales level; Analyze the employee sales data through a preset customer intent recognition model to construct a customer type map; In combination with the employee sales level and the customer type map, a plurality of dialogue scenarios are generated through a preset dialogue generation model, wherein the plurality of dialogue scenarios are generated corresponding to different customer types; After a conversation practice process is completed, the employee's sales intention and customer's intention in the previous conversation process are analyzed through the preset intention analysis model to build the employee's sales intention network and the customer's intention network; By using a preset hierarchical matching model, the employee sales intention network and the customer intention network are hierarchically matched to obtain a matching value, wherein the hierarchical matching includes coarse-grained matching, fine-grained matching and dynamic weight allocation; According to the relationship between the matching value and the preset matching threshold, the difficulty of the next dialogue scene is adjusted through the preset dialogue optimization model to achieve dynamic dialogue training.

2. The artificial intelligence-based dialogue training method according to claim 1, characterized in that: The pre-acquired employee sales data is analyzed by a preset employee capability evaluation model to obtain the employee sales level, including: According to the sales conversation text in the pre-acquired employee sales data, feature extraction is performed on the sales conversation text by using a preset text feature extraction model to generate a text feature vector; Analyze the behavior log in the employee sales data to obtain a time series behavior feature vector; Analyze the transaction records in the employee sales data to obtain a transaction success rate; Combining the text feature vector, the time series behavior feature vector and the transaction success rate, the sales ability of the employee is evaluated through a preset employee ability evaluation model to obtain an employee ability matrix; Determine the employee sales level based on the employee capability matrix.

3. The artificial intelligence-based dialogue training method according to claim 1, characterized in that: The analyzing the employee sales data by using a preset customer intention recognition model to construct a customer type map includes: Extract corresponding customer conversation records from employee sales data; According to the customer conversation records, the customer type and the customer's intention in each conversation are analyzed through a preset customer intention recognition model to construct a customer type map, wherein the customer type in the customer type map is the root node, the intention is the child node, and the conversion probability between different intentions is the weight of the edge.

4. The artificial intelligence-based dialogue training method according to claim 3, characterized in that: The method of analyzing the customer type and the customer's intention in each conversation by using a preset customer intention recognition model based on the customer conversation record to construct a customer type map includes: According to the customer conversation records, the customer's intention in each conversation is analyzed by a preset customer intention recognition model to obtain an intention recognition result; According to the customer conversation record, the corresponding customer type is identified by using a preset customer type identification model; Based on the customer intention recognition results in each conversation, the conversion probability between different intentions of the corresponding customer type is calculated; The customer type is taken as the root node, different intentions are taken as child nodes, and the conversion probability between corresponding intentions is taken as the weight of the edge to construct a customer type map.

5. The artificial intelligence-based dialogue training method according to claim 1, characterized in that: The combining of the employee sales level and the customer type map and generating a plurality of dialogue scenarios through a preset dialogue generation model, wherein the plurality of dialogue scenarios are generated corresponding to different customer types, including: According to the sales level of employees, the customer type graph is matched through the preset mapping rules to obtain the corresponding customer type sub-graph; According to the customer type subgraph and the real-time conversation content of the employees, a plurality of conversation scenarios are generated through a preset conversation generation model.

6. The artificial intelligence-based dialogue training method according to claim 4, characterized in that: The method generates multiple dialogue scenarios based on the customer type subgraph and the real-time dialogue content of the employees through a preset dialogue generation model, including: Initialize and select an initial intent from the customer type subgraph as the starting point of the conversation; According to the employee sales level and the starting point of the conversation, a first simulated response from the customer is generated through a preset conversation generation model; Combined with the employee's response results to the first simulated response, the transfer probability of the customer's intention is analyzed; Select the intention with the highest transfer probability as the next customer intention, and generate the customer's next simulated response through a preset dialogue generation model; The response generation process is repeated until the conversation ends.

7. The artificial intelligence-based dialogue training method according to claim 1, characterized in that: The employee sales intention network and the customer intention network are hierarchically matched by a preset hierarchical matching model to obtain a matching value, including: Analyze the coarse-grained matching degree of the employee sales intention network and the customer intention network to obtain the coarse-grained matching degree value; Analyze the fine-grained matching degree of the employee sales intention network and the customer intention network to obtain the fine-grained matching degree value; According to the preset business objectives, the coarse-grained matching value and the fine-grained matching value are weighted by using the preset hierarchical matching model to obtain the corresponding weights; The coarse-grained matching value and the fine-grained matching value are weightedly calculated with corresponding weights to obtain a matching value.

8. The artificial intelligence-based dialogue training method according to claim 7, characterized in that: The coarse-grained matching degree of the employee sales intention network and the customer intention network is analyzed to obtain a coarse-grained matching degree value, including: Count the number of occurrences of each type of intention in the employee sales intention network and the customer intention network respectively; Calculate a first matching ratio based on the correspondence between the various intentions in the employee sales intention network and the customer intention network and the number of occurrences; Assign corresponding weights to each type of intention through a preset first dynamic weight allocation model; A weighted calculation is performed in combination with the first matching ratio and the corresponding weight to obtain a coarse-grained matching degree value.

9. The artificial intelligence-based dialogue training method according to claim 7, characterized in that: The fine-grained matching degree of the employee sales intention network and the customer intention network is analyzed to obtain a fine-grained matching degree value, including: Through the preset intention segmentation model, each intention in the employee sales intention network and the customer intention network is divided to obtain intention subcategories; Calculate the second matching ratio according to the number of occurrences of each intention subcategory in the employee sales intention network and the customer intention network and the corresponding relationship between the intention subcategories; Assigning a corresponding weight to each intent subcategory through a preset second dynamic weight assignment model; A weighted calculation is performed in combination with the second matching ratio and the corresponding weight to obtain a fine-grained matching degree value.

10. A dialogue training system based on artificial intelligence, characterized in that: A method for implementing an artificial intelligence-based dialogue training method as claimed in any one of claims 1 to 9, comprising: The employee sales grade analysis module analyzes the pre-acquired employee sales data through the preset employee capability evaluation model to obtain the employee sales grade; A customer type map building module analyzes the employee sales data through a preset customer intent recognition model to build a customer type map; A dialogue generation module, combining the employee sales level and the customer type map, generates a plurality of dialogue scenarios through a preset dialogue generation model, wherein the plurality of dialogue scenarios are generated corresponding to different customer types; The network construction module, after a conversation practice process is completed, analyzes the employee's sales intention and customer's intention in the previous conversation process through a preset intention analysis model to build an employee sales intention network and a customer intention network; A network matching module, which performs hierarchical matching on the employee sales intention network and the customer intention network through a preset hierarchical matching model to obtain a matching value, wherein the hierarchical matching includes coarse-grained matching, fine-grained matching and dynamic weight allocation; The dialogue optimization module adjusts the difficulty of the next dialogue scene according to the relationship between the matching value and the preset matching threshold through a preset dialogue optimization model to achieve dynamic dialogue training.

Images