Data processing method and apparatus, medium, electronic device, and program product

Abstract

This disclosure relates to a data processing method, apparatus, medium, electronic device, and program product. The method includes: acquiring first dialogue data between a first user and a second user within a first time period; performing intent recognition and entity extraction on the first dialogue data to obtain first intent information and first entity information; determining first tag information of the second user based on the first intent information and the first entity information; the first tag information being used to represent the second user's business needs and emotional state; and generating communication prompt information based on the first dialogue data and the first tag information. This disclosure, by combining the dialogue data between the first user and the second user and the tag information of the second user, can obtain more accurate communication prompt information, thereby improving communication efficiency and accuracy.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and more specifically, to a data processing method, apparatus, medium, electronic device, and program product. Background Technology

[0002] With the continuous development of artificial intelligence technology, utilizing AI to provide corresponding services (such as transaction services and financial services) is a development trend for most enterprises. Therefore, there is an urgent need for a solution that can efficiently provide communication suggestions to service personnel based on the content of the dialogue between service personnel and customers. Summary of the Invention

[0003] This disclosure provides a data processing method, apparatus, medium, electronic device, and program product to improve the efficiency of communication with a first user.

[0004] In a first aspect, this disclosure provides a data processing method, including: Obtain the first dialogue data between the first user and the second user within the first time period; The first dialogue data is subjected to intent recognition and entity extraction to obtain first intent information and first entity information; The first tag information of the second user is determined based on the first intent information and the first entity information; the first tag information is used to represent the business needs of the second user and the emotional state of the second user. A communication prompt message is generated based on the first dialogue data and the first tag information.

[0005] Secondly, this disclosure provides a data processing apparatus, comprising: The acquisition module is configured to acquire the first dialogue data between the first user and the second user within a first time period. The extraction module is configured to perform intent recognition and entity extraction on the first dialogue data to obtain first intent information and first entity information. The determining module is further configured to determine the first tag information of the second user based on the first intent information and the first entity information; the first tag information is used to represent the business needs of the second user and the emotional state of the second user; The generation module is configured to generate communication prompts based on the first dialogue data and the first tag information.

[0006] Thirdly, this disclosure provides a computer-readable medium having a computer program stored thereon, which, when executed by a processing device, implements the steps of the method described in the first aspect.

[0007] Fourthly, this disclosure provides an electronic device, comprising: A storage device on which computer programs are stored; A processing device for executing the computer program in the storage device to implement the steps of the method in the first aspect.

[0008] Fifthly, this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in the first aspect.

[0009] By employing the technical solution provided in this application, the generation process of communication prompts is more accurate and personalized because it not only relies on the first dialogue data between the first user and the second user but also incorporates first tag information that reflects the second user's business needs and emotional state. Specifically, in the same dialogue intent, different second users may have different business needs and emotional states. The communication prompts generated by this application, combining the first dialogue data and first tag information, enable the first user to effectively communicate with second users who have different business needs and emotional states. For example, it can avoid points that might easily lead to complaints from the second user, provide processing suggestions more relevant to the second user's business needs, and offer dialogues more conducive to achieving the communication objective based on the second user's current emotional state. In other words, by combining the second user's first tag information and the first dialogue data between the second user and the first user, this application makes the generated communication prompts more accurate and effective, allowing the first user to provide more efficient and targeted services to the second user based on the communication prompts.

[0010] Other features and advantages of this disclosure will be described in detail in the following detailed description section. Attached Figure Description

[0011] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale. In the drawings: Figure 1 This is a schematic diagram illustrating an application environment of a data processing method according to an exemplary embodiment of the present disclosure.

[0012] Figure 2 This is an example diagram illustrating the use of the Rasa framework according to an exemplary embodiment of this disclosure.

[0013] Figure 3 This is a flowchart illustrating a data processing method according to an exemplary embodiment of the present disclosure.

[0014] Figure 4 This is an example diagram illustrating the use of customer tag information in a data processing method according to an exemplary embodiment of this disclosure.

[0015] Figure 5 This is a flowchart illustrating a data processing method for secondary development of the Rasa framework according to an exemplary embodiment of this disclosure.

[0016] Figure 6 This is an example diagram illustrating model integration in a data processing method according to an exemplary embodiment of the present disclosure.

[0017] Figure 7 This is an example diagram of an interface for outputting communication prompts in a data processing method according to an exemplary embodiment of the present disclosure.

[0018] Figure 8 This is another example diagram of an interface for outputting communication prompts in a data processing method according to an exemplary embodiment of the present disclosure.

[0019] Figure 9 This is a diagram illustrating an unreasonable dialogue prompt interface according to an exemplary embodiment of the present disclosure.

[0020] Figure 10 This is a detailed process example of a data processing method according to an exemplary embodiment of the present disclosure.

[0021] Figure 11 This is a schematic flowchart illustrating a dialogue flow of an application data processing method according to an exemplary embodiment of the present disclosure.

[0022] Figure 12 This is a structural block diagram of a data processing apparatus according to an exemplary embodiment of the present disclosure.

[0023] Figure 13 This is a structural block diagram of an electronic device according to an exemplary embodiment of the present disclosure. Detailed Implementation

[0024] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0025] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0026] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0027] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0028] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0029] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0030] It is understood that before using the technical solutions disclosed in the various embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.

[0031] First, some terms used in the embodiments of this application will be explained to facilitate understanding by those skilled in the art.

[0032] 1. Multi-turn dialogue framework A multi-turn dialogue framework is a technical architecture used to build and manage dialogue systems that can engage in multi-turn, continuous interaction with users. It can not only handle responses to single user inputs, but also maintain context, track dialogue state, and dynamically adjust response content throughout the dialogue process to achieve natural and smooth human-computer interaction.

[0033] 2. RASA Framework The RASA framework is a machine learning framework that can be used to build, deploy, and manage multi-turn chatbots and conversational applications. For example, RASA can be used to build text- and voice-based chatbots. The RASA framework can be composed of two parts: RASA NLU (Natural Language Understanding) and Rasa Core.

[0034] 2. RASA NLU Rasa NLU is part of the Rasa framework, primarily responsible for natural language understanding. Rasa NLU can include intent recognition and entity extraction, specifically for extracting user intent and key contextual information. Rasa NLU can parse user-input text and transform it into structured data for subsequent processing.

[0035] Specifically, RASA NLU has intent recognition and entity extraction capabilities. Intent recognition determines the user's intent, such as inquiring about the amount owed or the repayment date. Entity extraction extracts key information from the user input, such as the date, repayment method, amount, and contact information.

[0036] 3. Rasa Core Rasa Core is another part of the Rasa framework, primarily responsible for dialogue management and business logic execution. Rasa Core determines how to respond to the user by tracking the dialogue state and context, that is, it selects the optimal response and action based on the dialogue history.

[0037] Specifically, Rasa Core features dialogue state tracking, policy management, and custom actions. Dialogue state tracking records interactions between the user and the system, maintaining the context of the dialogue. Policy management uses machine learning models or rule engines to determine the system's next action. Custom actions execute specific business logic, such as querying outstanding amounts, recording repayment commitments, and processing extension requests; these custom actions can be executed by ActionServer.

[0038] 4. Action Server Action Server is a standalone service that connects chatbots with users and backend services to implement corresponding strategies and control the triggering of the next action. Specifically, Action Server can be used to execute custom actions, which can refer to complex tasks that cannot be handled by simple response templates or rules.

[0039] 5. BERT BERT (Bidirectional Encoder Representation from Transformers) can capture word-level and sentence-level representations through two methods: Masked Language Model (Masked LM) and Next Sentence Prediction (NSP).

[0040] 6. ASR ASR (Automatic Speech Recognition) is used to convert human speech into text or other computer-understandable forms, such as keystrokes, binary codes, or character sequences. ASR technology has been widely used in various applications, including voice assistants, intelligent customer service, voice search, and speech-to-text conversion.

[0041] 7. DM Dialog Management (DM) is a technology in Natural Language Processing (NLP) responsible for managing dialogue flow and decision-making. It can be used to understand user input, maintain dialogue state, generate appropriate responses, and maintain dialogue coherence and contextual understanding across multiple turns. DM is commonly used to build dialogue systems, chatbots, and other applications; that is, DM can determine the system's response based on the current dialogue state and control the context.

[0042] 8. TTS TTS (Text-to-Speech) is a technology that converts text into natural speech. This technology allows computers or devices to read written text and generate human-like speech output, enabling computers to interact with users through voice.

[0043] 9. Story Stories are sequences of dialogue examples used to train the dialogue management model. Each story describes a series of interactions between the user and the dialogue agent, and can show the specific path from user input to agent response. Stories help the model learn how to respond appropriately in real-world dialogues by specifying the order of user intents and agent actions. Stories can include user intents, entities, dialogue context (slots), and actions performed by the agent. Through these examples, the Rasa framework can learn dialogue patterns and generate corresponding responses in real-world dialogues.

[0044] 10. Domain In the Rasa framework, the domain is a crucial configuration file that defines the various elements required by the dialogue management system, including intents, entities, actions, slots, and templates. The domain file is the core of the dialogue system; it describes the system's scope and functionality, enabling the model to understand and respond to user input.

[0045] In related technologies, agents provide business services by calling customers, such as promoting products, resolving after-sales issues, and following up on payments. During these phone conversations, poor communication quality and a high failure rate often result due to agents' ineffective or mismatched communication styles with customers. For example, in the payment follow-up scenario, agents primarily use phone calls to track payments, which is not only ineffective and unprofessional but also prone to unreasonable payment follow-up.

[0046] In view of this, embodiments of this disclosure provide a data processing method, apparatus, medium, electronic device, and program product, proposing to generate communication prompt information based on the first dialogue data and the first tag information when a first user and a second user are obtained, enabling the first user to effectively communicate with the second user in response to different business needs and emotional situations.

[0047] The application scenarios of the data processing method provided in the embodiments of this disclosure are described below.

[0048] The data processing method provided in this disclosure can generate corresponding communication prompts by combining first dialogue data between a first user and a second user and the first tag information of the second user. These communication prompts may include communication scripts or prompts for those scripts. Therefore, generating these prompts can assist the first user in providing targeted services to the second user, thereby improving the efficiency and rationality of the dialogue. Specifically, this data processing method can be applied to scenarios requiring service provision, and thus can be applied to products in these scenarios, such as payment tracking systems, online dialogue systems, and content recommendation systems.

[0049] The data processing method of this disclosure can be applied to any scenario of communication between agents and customers, such as the scenario of following up on the payment of borrowed resources, the scenario of business consultation, and the scenario of assessing the credit limit of borrowed resources. In the scenario of following up on the payment of borrowed resources, the server acquires the first dialogue data between the agent (first user) and the customer (second user) in real time within a first time period, and determines the customer's first tag information based on the first intent information and the first entity information. The first intent information and the first entity information are obtained by performing intent recognition and entity extraction on the first dialogue data. The first tag information can reflect the customer's business needs and emotional state, etc. Then, communication prompt information is generated based on the customer's first tag information and the real-time acquired first dialogue data. The communication prompt information may include a psychological profile description of the customer, such as a psychological category of being detail-oriented, interpersonal-oriented, or prone to silence. The communication prompt information may also include guidance on communication techniques, such as what kind of emotion the agent should use and what aspects of the content should be emphasized in the communication, etc. The agent communicates with the customer according to the communication prompt information.

[0050] To better understand the data processing method, apparatus, readable medium, electronic device, and program product provided in the embodiments of this disclosure, the application environment applicable to the embodiments of this disclosure is described below.

[0051] Please see Figure 1 , Figure 1 This diagram illustrates an application environment for a data processing method provided in an embodiment of this disclosure. As one implementation, the data processing method provided in this embodiment can be applied to an electronic device. This electronic device can be, for example,... Figure 1 The server 110 shown can be connected to the first electronic device 120 and the second electronic device 130 via a network. The network serves as a medium for providing a communication link between the server 110 and the first electronic device 120 and the second electronic device 130. The network can include various connection types, such as wired communication links, wireless communication links, etc., and this disclosure does not limit this type.

[0052] It should be understood that Figure 1The server 110, network, first electronic device 120, and second electronic device 130 shown are merely illustrative. Depending on the implementation requirements, any number of servers 110, networks, first electronic devices 120, and second electronic devices 130 can be used. For example, server 110 can be a physical server 110, or a server cluster consisting of multiple servers 110, etc. Furthermore, the first electronic device 120 can be an electronic device held by a first user; for example, the first user can be a customer service representative. The second electronic device 130 can be an electronic device held by a second user; for example, the second user can be a customer whose account is being tracked. The first electronic device 120 and the second electronic device 130 can be the same type of device or different types of devices. For example, the first electronic device 120 is a mobile phone, and the second electronic device 130 is also a mobile phone. Or, the first electronic device 120 is a computer, and the second electronic device 130 is a mobile phone.

[0053] The first electronic device 120 is used to provide communication services between a first user and a second user, such as VoIP services, social chat services, etc. The first electronic device 120 can be a landline telephone, personal computer, mobile terminal, tablet computer, etc.

[0054] Taking a personal computer as an example, the first electronic device 120 can be equipped with an application that provides VoIP services. A first user can use this application to make VoIP calls with a second user. The first electronic device 120 can also be equipped with a social networking application for social chat between the first and second users. The first electronic device 120 saves the dialogue data between the first and second users and sends this data to the server 110, which then uses this data to obtain communication prompts.

[0055] The second electronic device 130 is also used to provide communication services between the second user and the first user, such as call services and social chat services. The second electronic device 130 can be a smartphone, landline phone, tablet computer, personal computer, etc. The first electronic device 120 and the server 110 can establish a communication connection via a wired network or a wireless network. The first electronic device 120 and the second electronic device 130 can establish a communication connection via a wireless network.

[0056] As an alternative approach, embodiments of this disclosure can utilize the Rasa framework to build a chatbot and employ it to achieve intelligent question answering. The specific usage of the Rasa framework is as follows: Figure 2As shown, in the process of the first user providing services to the second user, this embodiment of the disclosure can first acquire the dialogue data of both users and perform speech recognition (ASR) on the dialogue data to convert the speech into text. Then, semantic understanding (NLU) is performed on the converted text, that is, semantic understanding is performed using the NLU module to obtain the semantic understanding result. After that, the semantic understanding result is input to the dialogue management (DM) module to manage the context content. Optionally, when the context is filled to a preset level, the Action module can be triggered, and specific operations can be performed based on the Action module to obtain the response text. In addition, this embodiment of the disclosure can perform text-to-speech (TTS) on the response text and return the synthesized speech to the client.

[0057] In some implementations, the Rasa framework setup process includes the following steps: initializing the project; preparing NLU training data; configuring the NLU model; preparing story data, which is used to return context template information; defining a domain, which is used to record entity information; configuring the Rasa Core model; training the model; and testing the robot. While some technologies utilize the Rasa framework for account tracking, the effectiveness and intelligence of intelligent robots in account tracking are poor, and they cannot completely replace human intervention.

[0058] The embodiments of this disclosure will be further explained below with reference to the accompanying drawings.

[0059] Figure 3 This is a flowchart illustrating a data processing method according to an exemplary embodiment of the present disclosure. The method can be applied to electronic devices. (Refer to...) Figure 3 The data processing method may include the following steps: Step S310: Obtain the first dialogue data between the first user and the second user within the first time period.

[0060] As explained above, the first user can be a service provider who provides services to the second user, such as a human agent or an intelligent robot; the second user can be the person being served by the first user, such as a customer whose accounts are being followed up or a customer who needs business consultation.

[0061] In this embodiment of the disclosure, the first dialogue data between the first user and the second user within the first time period may include, but is not limited to, call data, voice data from social applications, text data, etc. Furthermore, the dialogue data may be obtained based on chat content uploaded from an electronic device or based on offline chat content stored on a server.

[0062] In some implementations, obtaining the first dialogue data between the first user and the second user within a first time period may include: obtaining the third tag information of the first user, which is used to represent the service capabilities of the first user; and obtaining the first dialogue data between the first user and the second user within the first time period when the third tag information meets the communication prompt conditions.

[0063] In other words, before acquiring the first dialogue data, this embodiment of the disclosure can first determine whether the third tag information corresponding to the first user meets the communication prompt conditions. If the third tag information of the first user meets the communication prompt conditions, this embodiment of the disclosure can acquire the first dialogue data between the first user and the second user within the first time period. Otherwise, it is not necessary to acquire the first dialogue data between the first user and the second user within the first time period.

[0064] The third tag information can be used to represent the service capability of the first user. The communication prompt condition can be that the service capability of the first user is lower than a preset capability level. That is, when it is determined that the service capability of the first user is lower than the preset capability level, this embodiment of the disclosure can obtain the first dialogue data between the first user and the second user within a first time period. This can improve communication efficiency, that is, it can provide targeted prompt information to the first user who needs it, while the first user who does not need prompt information does not need to generate prompt information. This not only ensures communication efficiency, but also avoids the generation of unnecessary prompt information.

[0065] Step S320: Perform intent recognition and entity extraction on the first dialogue data to obtain first intent information and first entity information.

[0066] As an optional approach, after obtaining the dialogue data between the first user and the second user within a first time period, embodiments of this disclosure can perform intent recognition and entity extraction on the first dialogue data to obtain first intent information and first entity information.

[0067] Intent recognition and entity extraction can be implemented based on Rasa NLU within the Rasa framework, specifically for extracting user intent and key contextual information. Using Rasa's intent recognition and entity extraction can increase the development efficiency of dialogue management.

[0068] Step S330: Determine the first tag information of the second user based on the first intent information and the first entity information.

[0069] In this embodiment of the disclosure, the first tag information can be used to represent the business needs and emotional state of the second user. Specifically, the first tag information may include at least one of the second user's complaint tag, business tag, and psychological tag, wherein the complaint tag and business tag are used to characterize the second user's business needs, and the psychological tag is used to characterize the second user's emotional state.

[0070] The complaint point label identifies information that second-party users are likely to complain about. Specifically, it identifies the dissatisfaction expressed by second-party users regarding the service or product and their stated intention to complain during communication. This label helps first-party personnel quickly identify and address second-party complaints, thereby providing appropriate solutions and services. For example, complaint point labels may include mentions of the China Banking Regulatory Commission, mentions of consumer associations, threats of legal action, complaints about service quality, complaints about communication attitude, complaints about processing efficiency, and threats of media exposure.

[0071] This business tag is used to describe a customer's loan behavior and credit status. Based on this tag, financial institutions can assist in assessing a second user's credit risk, loan demand, and repayment ability. For example, the business tag may include loan history, overdue period, loan history, loan amount, loan type, loan term, repayment method, credit score, and debt level.

[0072] This psychological tag helps the first user better understand the second user's psychological state. The tag can be used to describe the second user's psychological state and behavioral characteristics. For example, psychological tags may include irritability, tendency to remain silent, high aggression, mood (whether pleasant or unpleasant), calmness, politeness, anxiety, confidence, patience, and decisiveness. The psychological tag can also be used to describe the second user's emotional state and affective characteristics, allowing for a better understanding of the second user's emotional changes and thus adjusting communication strategies and responses. For example, the psychological tag may include consecutive negative emotional samples and their frequency, the strength of the customer's repayment intention, whether there have been multiple instances of intentional loan delays, whether credit scores have been mentioned, and the number of times credit scores have been mentioned consecutively. Specifically, psychological tags can be obtained by analyzing the second user's basic information or by analyzing the dialogue data between the first and second users in real time.

[0073] In some implementations, the first tag information of the second user may include tags with multiple dimensions. Specifically, in addition to complaint point tags, business tags, and psychological tags, the first tag information may also include customer segmentation tags, which can be used to determine which category the second user belongs to. For example, segmentation tags may include students, teachers, housewives, office workers, freelancers, retirees, business owners, teenagers, high-income groups, and low-income groups, etc.

[0074] Optionally, the first tag information may also include a customer rating tag, which can be obtained through comprehensive analysis of customer characteristics, including customer profile features and emotional characteristics. For example, the rating tag may include high-risk customers, medium-risk customers, low-risk customers, and premium customers.

[0075] It should be noted that, in addition to the multiple tags mentioned above, the first tag information of the second user may also include other tags, as long as they can be used to represent the second user's personality information. The specific tags included depend on the actual situation and will not be elaborated here.

[0076] In some implementations, determining the first tag information of the second user based on the first intent information and the first entity information may include: if the difference between the first intent information and the second intent information satisfies a first difference condition, and the difference between the first entity information and the second entity information satisfies a second difference condition, then determining the second tag information of the second time period as the first tag information of the second user; wherein the second tag information of the second time period is determined based on the intent information and entity information identified from the dialogue data within the second time period; if the difference between the first intent information and the second intent information does not satisfy the first difference condition, and / or the difference between the first entity information and the second entity information does not satisfy the second difference condition, then updating the second tag information based on the first intent information and the first entity information to obtain the first tag information.

[0077] In other words, after obtaining the first intent information and the first entity information within the first time period, and obtaining the second intent information and the second entity information within the second time period, this embodiment of the disclosure can determine whether the difference between the first intent information and the second intent information satisfies the first difference condition, and determine whether the difference between the first entity information and the second entity information satisfies the second difference condition. If both satisfy the difference condition, the second tag information within the second time period can be determined as the first tag information of the second user, that is, the first tag information can be updated using the second tag information.

[0078] Conversely, if at least one of the above two difference conditions is not met, the embodiments of this disclosure may not update the first tag information, that is, update the second tag information according to the first intent information and the first entity information to obtain the first tag information.

[0079] In some implementations, the process of using the first intent information and the first entity information may include: acquiring historically stored second dialogue data and business data related to the second user; and determining the first tag information of the second user based on the second dialogue data and the business data.

[0080] Specifically, this business data may include basic information retained by the second user in the past, such as the second user's ID, name, gender, age, education, occupation, income, family members, hobbies, values, personality, geographical location, whether they have a mortgage, whether they have overdue amounts, and the overdue period. This second dialogue data consists of dialogue data from at least one past interaction between the first user and the second user.

[0081] The second user's second dialogue data and business data can be obtained in advance by the first user before actively initiating a dialogue with the second user. After obtaining the second user's second dialogue data and business data, this embodiment of the disclosure can store the second dialogue data and business data in a designated database. That is, the second dialogue data and business data and the user identifier in the designated database can be stored in a corresponding relationship.

[0082] For example, an agent plans to communicate with customer A. The agent can collect customer A's second conversation data and business data a day in advance and upload this basic information to a server. The server can then store this basic information in a designated database. During the conversation between the agent and customer A, the agent can retrieve customer A's second conversation data and business data from this database and determine customer A's tag information based on this data.

[0083] In this process, embodiments of this disclosure can perform intent recognition and entity extraction on the second dialogue data to obtain corresponding intent information and entity information, and then determine the first tag information of the second user based on the intent information, entity information and business data.

[0084] In some implementations, the second user's first tag information can be obtained through a tag extraction model. This involves inputting the second user's second dialogue data and business data into the tag extraction model, which then classifies and summarizes the second user's second dialogue data and business data, outputting the second user's first tag information. This tag extraction model can be implemented, for example, using a logistic regression model, decision tree, or random forest; this disclosure does not limit its implementation. The tag extraction model can be trained using a large amount of dialogue data, business data, and tag information, and the tags in the training samples can be obtained through manual labeling.

[0085] Furthermore, after initially obtaining the first tag information of the second user based on the second dialogue data and business data, the first tag information can be updated based on the currently acquired first dialogue data between the first user and the second user. In other words, the first tag information can be dynamically adjusted during the dialogue between the first user and the second user.

[0086] For example, the initial psychological characteristic of the second user is identified as irritable. However, if the psychological characteristics of the second user are determined to be calm, confident, and patient by identifying the first dialogue data between the first user and the second user, then the psychological label of the second user can be updated to calm, confident, and patient.

[0087] In one possible implementation, updating the first tag information based on the currently acquired first dialogue data may include: performing intent recognition and entity extraction on the first dialogue data, filling the recognized intent information and extracted entity information into the corresponding slots (fields) in the multi-turn dialogue template, and updating at least one of the complaint tag, business tag, and psychological tag based on the information of the slots already filled in the multi-turn dialogue template.

[0088] As described above, intent recognition and entity extraction can be implemented based on Rasa NLU within the Rasa framework, specifically for extracting user intent and key contextual information. Furthermore, the multi-turn dialogue template can be a configuration file within the Rasa framework. This template includes multiple slots, and depending on the intent and entity filled into the slots, it can trigger different actions. In this embodiment, the actions triggered by the multi-turn dialogue template can include updating at least one of the complaint point tags, business tags, and psychological tags.

[0089] For example, the multi-turn dialogue template includes multiple emotion slots. Updating the psychological label based on the information of the already filled slots in the multi-turn dialogue template can include: if, in N consecutive rounds of dialogue, the entity information filled into the corresponding emotion slot does not match the current psychological label, then the psychological label is updated, where N is a preset positive integer greater than or equal to 1. For instance, if the second user's current psychological characteristic is irritability, and the emotional entities extracted in three consecutive rounds of dialogue are all calm, then the multi-turn dialogue template can update the second user's psychological label to calm if it detects that the emotional entities filled into the emotion slots three times consecutively are all calm.

[0090] This embodiment of the disclosure determines the user's segmentation information (tag information) by analyzing the second user's first tag information, and analyzes the user's psychological and behavioral information through various dimensions of data. For example, age, gender, family information, repayment ability, personality traits, and whether the user is prone to anger or complaints. Accurate first tag information of the second user enables subsequent communication prompts to be more targeted.

[0091] Step S340: Generate communication prompt information based on the first dialogue data and the first tag information.

[0092] In this embodiment of the disclosure, the first tag information may include the business needs of the second user. Generating communication prompt information based on the first dialogue data and the first tag information may include: generating communication prompt information based on the first dialogue data and the first tag information when the business needs of the second user belong to a first type of need; and generating communication prompt information based on the first dialogue data when the business needs of the second user belong to a second type of need.

[0093] Since the first tag information may include the second user's business needs, after obtaining the first tag information, this embodiment of the disclosure can determine the category of the business need. Based on this, different data is used to generate communication prompt information according to the different categories. Specifically, when it is determined that the second user's business need belongs to the first type of need, communication prompt information is generated based on the first dialogue data and the first tag information.

[0094] Optionally, when it is determined that the second user's business need belongs to the second type of need, a communication prompt message is generated based on the first dialogue data. For example, the first type of need may be a complaint-related need, and the second type of need may be other needs besides complaint-related needs.

[0095] In some implementations, generating communication prompts based on first dialogue data and first tag information may include: performing semantic analysis on the first dialogue data and determining a set of dialogues that match the semantic analysis results; determining the prohibited dialogues for the second user based on the first tag information, and determining the available dialogues for the second user based on the prohibited dialogues and the set of dialogues; and generating communication prompts based on the available dialogues.

[0096] As can be seen, in the process of generating communication prompts, this embodiment of the disclosure can first perform semantic analysis on the first dialogue data to obtain semantic analysis results. Based on this, a set of dialogue phrases corresponding to the semantic analysis results is determined; different semantic analysis results result in different dialogue phrase sets. Additionally, this embodiment of the disclosure can also determine the prohibited dialogue phrases corresponding to the second user based on the first tag information. Based on these prohibited dialogue phrases, the dialogue phrase set can be filtered to obtain the available dialogue phrases corresponding to the second user. Finally, communication prompts can be generated based on these available dialogue phrases.

[0097] This embodiment of the disclosure can perform semantic recognition and analysis on first dialogue data between a first user and a second user to obtain semantic recognition results. Based on these results, key information in the dialogue data is determined, and this key information is combined with the second user's first tag information to comprehensively determine the corresponding communication prompt information.

[0098] Here, key information, primary tag information, and communication prompt information can be stored in the target database according to a preset correspondence. After obtaining the key information and primary tag information, the communication prompt information can be retrieved by searching the target database.

[0099] In some implementations, the first tag information may include the second user's emotion tag. Generating communication prompt information based on the first dialogue data and the first tag information may include: generating communication prompt information based on the first dialogue data and reassuring language when the emotion tag indicates that the second user's emotion is a negative emotion type; and generating communication prompt information based on the first dialogue data and efficiency-related language when the emotion tag indicates that the second user's emotion is a positive emotion type.

[0100] During a conversation with a second user, if the second user's emotion is detected as negative, the primary approach is to soothe them. This involves generating communication prompts based on the first user's conversation data and soothing phrases to avoid conflict. Conversely, if the second user's emotion is detected as positive, the primary approach is to prioritize efficiency. This involves generating communication prompts based on the first user's conversation data and efficiency-related phrases to ensure service efficiency.

[0101] This disclosure embodiment can utilize a model to generate communication prompts, that is, to perform information generation operations based on first dialogue data and first tag information through the model. Specifically, inputting the first dialogue data between the first user and the second user and the first tag information of the second user into the model can generate communication prompts. This model can be implemented, for example, through random forests, decision trees, neural networks, etc., and this disclosure does not limit it to these. In addition, this model can be referred to as an auxiliary account tracking model.

[0102] After generating communication prompts based on the first dialogue data and the first tag information, this embodiment of the disclosure can output the communication prompts. Based on these communication prompts, the first user can provide more targeted services to the second user.

[0103] As an optional approach, embodiments of this disclosure can output communication prompts to assist the first user in providing services to the second user based on these prompts; that is, the acquired communication prompts can be displayed through a visual interface. Optionally, the server can send the acquired communication prompts to a first electronic device held by the first user, and the first electronic device can then display the communication prompts visually.

[0104] Through the above technical solution, after obtaining the first dialogue data between the first user and the second user and the first tag information of the second user, the embodiments of this disclosure can generate communication prompt information based on the first dialogue data and the first tag information. Since the generation of the communication prompt information combines the first dialogue data between the first user and the second user and the first tag information of the second user, the communication prompt information can specifically assist the first user in providing personalized services to the second user. Moreover, the communication prompt information can be continuously updated as the dialogue progresses, thereby improving the flexibility of the service.

[0105] In some implementations, the data processing method may further include the following steps: Based on at least one of the following features among the profile features, emotional features, and service features of each first user, first users who meet the sample data collection conditions are selected from multiple first users; third dialogue data between the selected first users and second users in history are obtained; training samples for the model are constructed based on the third dialogue data and the second tag information of the second user, so as to train the model based on the training samples; on this basis, the communication prompt information is generated by the model based on the first dialogue data and the first tag information.

[0106] In other words, by using at least one of the agent's profile features, emotional features, and service features, excellent agents are selected, and the training samples of the model are constructed using the third dialogue data of the excellent agent's history with customers and the second label information of the customer, so that the model can learn the dialogue experience of excellent agents.

[0107] The specific screening process is explained below: In this embodiment, the first user ID of each first user can be obtained first. Based on this, at least one of the profile features, sentiment features, and service features corresponding to each first user ID can be obtained. Then, based on at least one of the profile features, sentiment features, and service features of the first user, a first score corresponding to the first user can be determined. When it is determined that the first score exceeds a preset threshold, the first user can be determined as a first user that meets the screening conditions.

[0108] In some implementations, the first user ID is used to represent a specific service personnel within the team to which the first user belongs, serving as a unique identifier for the first user. The profile characteristics of the first user may include the first user's age, gender, years of service, work experience, and accounts receivable follow-up (collection) experience indicators, etc. Based on these profile characteristics, the overall performance level of the first user can be measured.

[0109] In other implementations, the emotional characteristics of the first user can be obtained by extracting the user's call text within a specified time period. These emotional characteristics may include the number of times the first user was complained about within the specified time period, the proportion of negative emotional responses in ASR (Accepted Speech Reflection), the maximum number of consecutive negative emotional responses in ASR, and the number of consecutively occurring negative emotional texts in ASR. The proportion of negative emotional responses in ASR can be the percentage of negative emotions in the first user's voice input. For example, if the first user's voice input contains 100 words, and 20 of them represent negative emotions, then the proportion of negative emotional responses in ASR is 20%.

[0110] The maximum consecutive negative sentiment count in ASR can be calculated by determining how many consecutive words with negative sentiment the first user utters. For example, if the first user's voice input contains 100 words, and a continuous speech contains 5 words expressing negative sentiment, then the maximum consecutive negative sentiment count in ASR is 5.

[0111] The number of consecutive negative sentiment texts in ASR can be the number of consecutive negative sentiment text segments in the first user's voice input. For example, if the first user's voice input contains 3 consecutive sentences, 2 of which express negative sentiment, then the number of consecutive negative sentiment texts in ASR is 2.

[0112] The aforementioned percentage of negative sentiment in ASR, the maximum number of consecutive negative sentiment in ASR, and the number of consecutive negative sentiment texts in ASR can be used to reflect the working status of the first user.

[0113] In addition to the features mentioned above, sentiment features may also include the number of times prohibited or sensitive words are triggered, i.e., the number of times customers are insulted or verbally abused, and whether the customer is appeased. This appeasing feature can be used to indicate the first user's work status and emotional state, thus better preventing customer complaints. This embodiment of the disclosure can extract the first user's historical dialogue data over multiple consecutive time periods as training ASR text, and then calculate the relevant sentiment features.

[0114] It should be noted that emotional characteristics may also include positive emotional expression, negative emotional expression, speech rate, tone variation, emotional stability, emotional change trends, emotional recognition ability, emotional regulation ability, and emotional fatigue level.

[0115] In other implementations, the service characteristic of the first user can be a performance characteristic, which can be represented by the first user's performance and the collection rate of different accounts receivable follow-up queues. That is, the service characteristic can include accounts receivable follow-up queues, collection rates, etc. The accounts receivable follow-up queue can be a dedicated queue for customers with overdue payments. The accounts receivable follow-up queue is usually classified and sorted according to factors such as the degree of delinquency, the amount owed, and the customer type, including customer information, delinquency status, accounts receivable follow-up records, and accounts receivable follow-up plans. Therefore, using the accounts receivable follow-up queue as one of the service characteristics can better measure the performance of the first user.

[0116] In addition, the service characteristics of the first user may also include the amount of payment received, the success rate of payment follow-up, the efficiency of payment follow-up, the call connection rate, the promised repayment rate, the number of follow-ups, the average follow-up time, customer satisfaction, the default rate, and the problem-solving rate.

[0117] This embodiment of the disclosure can select an excellent first user from multiple first users by combining at least one of the first user's profile features, emotional features, and service features, and use the historical dialogue data of the excellent first user to determine the training samples for the training model. This can ensure the accuracy of the communication script suggestions to a certain extent.

[0118] In some implementations, determining a first user's first score based on at least one of the first user's profile features, emotional features, and service features includes: The profile score of the first user is determined based on the profile characteristics of the first user; The service emotion score for the first user is determined based on the first user's emotional characteristics. The service indicator score of the first user is determined based on the service characteristics of the first user. The first score is determined based on the profile score, service emotion score, and service indicator score.

[0119] As an optional approach, after extracting at least one feature from the first user's profile features, emotional features, and service features, this embodiment of the disclosure can combine these features to comprehensively determine the first user's first score. Specifically, this embodiment of the disclosure can obtain the score corresponding to each feature, and then perform a weighted summation of these scores to obtain the first score.

[0120] For example, in this embodiment of the disclosure, a profile score of a first user can be determined based on the profile features of the first user, and the formula for calculating the profile score can be: ; Among them, A i This refers to the i-th feature of the profile to be screened; a iThis refers to the weight corresponding to the i-th feature; A represents the profile score of the first user, which represents the service level score of the first user; n refers to the number of profile features of the first user.

[0121] Optionally, embodiments of this disclosure can determine the service emotion score of the first user based on the first user's emotional characteristics. This service emotion score can be the work emotion score of the agent, and its calculation formula can be: ; Among them, B i This refers to the i-th feature of the first user's emotional characteristics; b i B refers to the weight corresponding to the i-th feature; B represents the service emotion score, which represents the service attitude score of the first user; m refers to the number of emotional features of the first user.

[0122] Optionally, embodiments of this disclosure may determine the service indicator score of the first user based on the service characteristics of the first user. This service indicator score may be a performance indicator score, and its calculation formula may be: ; Among them, C i This refers to the i-th feature of the service characteristics of the first user; c i This refers to the weight corresponding to the i-th feature; C represents the service indicator score; t refers to the number of service features for the first user.

[0123] As an alternative approach, embodiments of this disclosure may calculate the first score based on at least one of the first user's profile score, service sentiment score, and service indicator score.

[0124] As an example, one of the following scores—the first user's profile score, the service sentiment score, and the service metric score—can be used as the first score. For instance, the first user's profile score can be used as the first score. Alternatively, the service metric score can be used as the first score.

[0125] As another example, embodiments of this disclosure may calculate the first score based on at least two of the first user's profile score, service sentiment score, and service indicator score. For example, the first score may be calculated based on the first user's profile score and service sentiment score. Alternatively, the first score may be calculated based on at least two of the first user's profile score and service indicator score.

[0126] Preferably, in this embodiment of the disclosure, the first score can be calculated based on the first user's profile score, service sentiment score, and service indicator score. In this case, the formula for calculating the first score can be: Seat ID score =A+B+C; Among them, seat_ID score This is the total score for the first user. A, B, and C represent the profile score, service sentiment score, and service indicator score for the first user, respectively.

[0127] Here, after obtaining the first score, this embodiment of the disclosure can determine whether the first score exceeds a preset threshold. If the first score exceeds the preset threshold, the first user's statement can be used as the first communication statement. That is, the first user whose first score exceeds the preset threshold is determined as the first user who meets the screening criteria.

[0128] The preset threshold can be pre-set. In this embodiment, the first user sample can be divided into first user levels based on the affiliation level of the first score. For example, the first user level can be divided into gold, silver, and bronze agents. The preset threshold can be set according to specific business scenarios to reasonably divide the first user sample.

[0129] In other words, the preset threshold in this embodiment may include multiple sub-preset thresholds. These multiple sub-preset thresholds can be used to classify the level of the first user, thereby obtaining the first user who meets the screening criteria. Based on this, this embodiment can collect the service ASR text of the first user within a specified time period and use the ASR text to train a model for generating communication prompts.

[0130] It should be noted that the screening of the first user in this embodiment can be achieved through a script screening model. The input data of the script screening model can be the basic information of multiple first users, and the output can be the first user who meets the screening criteria. Here, the first user who meets the screening criteria can be referred to as an excellent agent.

[0131] Through the above technical solution, a first score can be calculated for each first user, and the first user can be divided based on the first score. Based on the division result, the embodiments of this disclosure can obtain more accurate and appropriate sample data.

[0132] Furthermore, after selecting the first user who meets the criteria and obtaining the third dialogue data between the first user and the second user, the training samples for the model can be constructed based on the third dialogue data and the fourth tag information of the second user. Specifically, this can include: performing intent recognition and entity extraction on the third dialogue data to obtain intent information and entity information; determining the communication script used by the first user in response to the intent information and entity information from the third dialogue data; using the intent information, entity information, and the fourth tag information as data samples; and using the communication script or the prompt information of the communication script as the label of the data samples to obtain model training samples.

[0133] In this embodiment, the fourth tag information is similar to the first tag information, and it may also include sub-tag information of multiple dimensions. Under the same intent information (e.g., the intent to inquire about how to handle the same business), different sub-tag information can correspond to different communication scripts. For example, for a second user with a high complaint rate, no house or car loan, and who is detail-oriented, the corresponding excellent communication script could be: "Based on your situation, I suggest the following solutions: First, [Solution 1], second, [Solution 2], these measures should effectively solve your problem."

[0134] In some implementations, the fourth tag information may include a customer level tag, which may be determined by: determining a second score for the second user based on at least one of the second user's profile features and pressure-manipulation features; classifying the second user into customer levels based on the second score; and determining the second user's level tag based on the level classification result.

[0135] Specifically, in this embodiment of the disclosure, the customer ID of each second user can be obtained first. Based on this, at least one of the profile features and pressure-appeal features corresponding to each customer ID can be obtained. Then, a second score corresponding to the second user can be obtained based on at least one of the profile features and pressure-appeal features. The second user can be classified into customer levels according to the second score, and the result of the level classification can be used as the tag information of the second user.

[0136] In some implementations, a customer ID is used to represent a specific customer among multiple customers, uniquely identifying the second user. The second user's profile features may include their age, income, occupation, gender, family members, whether they have a mortgage, overdue amount, and overdue period. Based on these profile features, customer positioning of the second user is possible.

[0137] In other implementations, the pressure-appeal characteristic of the second user can be obtained by extracting the second user's ASR text over a specified time period. This pressure-appeal characteristic can be the customer's emotional characteristics. The pressure-appeal characteristic may include the number of consecutive negative samples from the second user, the strength of the customer's willingness to repay, and whether the customer has repeatedly and deliberately delayed the loan application.

[0138] Furthermore, the pressure-appeal level feature can be used to reflect a customer's willingness to repay. Specifically, in this embodiment, the second user's historical ASR text can be used to determine whether the second user mentions credit reporting and the number of times they mention it consecutively. In other words, the pressure-appeal level feature can also include whether credit reporting is mentioned and the number of times it is mentioned consecutively; these two features can indicate the degree to which the second user cares about credit reporting. Based on these features, this embodiment can score the second user's level of psychological aversion and then calculate the customer's pressure-appeal level based on the score.

[0139] It should be noted that the characteristics of the degree of pressure exerted may also include willingness to repay, ability to repay, level of communication and cooperation, commitment to repayment, repayment plan formulation, cooperative attitude, threatening behavior, emotional expression, stability of attitude, history of cooperation, and problem-solving ability.

[0140] In some implementations, obtaining a second score for the second user based on at least one of the second user's profile features and pressure tolerance features includes: Determine the customer positioning score of the second user based on the profile characteristics of the second user; The pressureability score of the second user is determined based on the pressureability characteristics of the second user. The second user's second score is calculated by combining the customer positioning score and the pressure level score.

[0141] As an optional approach, after extracting at least one of the profile features and pressure-appeal level features of the second user, embodiments of this disclosure can combine these features to comprehensively determine the second user's second score. Specifically, embodiments of this disclosure can obtain the score corresponding to each feature, and then perform a weighted summation of these scores to obtain the second score.

[0142] For example, in this embodiment of the disclosure, the customer positioning score of the second user can be determined based on the profile features of the second user. The formula for calculating the customer positioning score can be: ; Among them, M i This refers to the i-th feature of the customer's psychological profile; m i This refers to the weight corresponding to the i-th feature; M represents the customer positioning score, which represents the score of the customer's basic information and psychological characteristics; k refers to the number of customer psychological profile features.

[0143] Optionally, embodiments of this disclosure can determine the pressure tolerance score of the second user based on the pressure tolerance characteristics of the second user, and the formula for calculating the pressure tolerance score can be: ; Where, N iThis refers to the i-th characteristic of the customer's ability to exert pressure; n i 1 refers to the weight corresponding to the i-th feature; N represents the score of the customer's ability to exert pressure; d refers to the number of features related to the customer's ability to exert pressure.

[0144] As an alternative approach, embodiments of this disclosure may calculate a second score based on at least one of a customer positioning score and a pressure tolerance score.

[0145] As an example, either the customer positioning score or the pressure tolerance score can be used as the second score. For example, the customer positioning score can be used as the second score. Or, the pressure tolerance score can be used as the second score.

[0146] Preferably, in this embodiment of the disclosure, the second score can be calculated based on a combination of the customer positioning score and the pressure exertion score. In this case, the formula for calculating the second score can be: Customer_ID score =M+N; Among them, customer_ID score This is the total score for the second user, where M and N represent the customer positioning score and the pressure level score, respectively.

[0147] Here, after obtaining the second score, this embodiment of the disclosure can classify customers into customer levels based on the second score. During the classification process, the second score can be compared with multiple level ranges. When the second score belongs to a certain level range, the second user can be determined to belong to that level. For example, if the second score is determined to belong to the first level range, then the customer is determined to be a premium customer.

[0148] After obtaining the second score, this embodiment of the disclosure can classify the second user into customer levels based on the second score, and use the classification result as the level label of the second user.

[0149] The above technical solution allows for the calculation of the total score for each second user, and based on this score, a level label can be assigned to the second user. To better illustrate the role of the level label, embodiments of this disclosure provide the following... Figure 4 The example diagram shown is obtained by... Figure 4 Knowing that after automatically dialing customer numbers, this embodiment of the disclosure can use an I-model to calculate customer information, that is, to obtain tag information using the customer's basic information. Here, the I-model can be the tag extraction model mentioned above. Based on this, customer information is initialized to optimize the obtained tag information. Then, this embodiment of the disclosure can classify questions through intent recognition and select corresponding script responses based on intent and customer score (level).

[0150] Furthermore, the communication prompts output by the model can include communication scripts, or prompts for communication scripts (i.e., communication suggestions), to prompt the first user on which script to use. For example, prompting the first user to use a question to guide the customer to say the result of not repaying, or prompting the first user to avoid getting entangled with the customer on the reasons for not repaying.

[0151] In this embodiment, there is a corresponding relationship between the communication scripts and the customer's tag information; different customer tag information corresponds to different first communication scripts. For example, if the customer is determined to be in group A based on their tag information, this embodiment can train the model using the first type of communication script. Similarly, if the customer is determined to be in group B based on their tag information, this embodiment can train the model using the second type of communication script. This allows the model to learn effective communication scripts for different groups, thereby not only reducing customer complaint rates but also improving service efficiency for the first user.

[0152] The model trained above for generating communication prompts can be integrated into a secondary development target dialogue framework, which may include the Rasa framework. Based on this target dialogue framework, intent recognition and entity extraction of the dialogue data can be achieved. As described above regarding the multi-turn dialogue template, based on slot filling, the multi-turn dialogue template can trigger updates to complaint point tags, business tags, and psychological tags. It can also trigger calls to the trained model. For example, the model can be called by taking the first tag information of the second user, along with the intent and entity information filled into the slots of the multi-turn dialogue template, as input. This allows the model to generate communication prompts based on the first tag information and the intent and entity information filled into the slots of the multi-turn dialogue template.

[0153] As described above, the trained model can be integrated into the Rasa framework through secondary development. Please refer to [link / reference]. Figure 5 In the process of secondary development of the Rasa framework, the embodiments disclosed herein can first perform session management on the client. Specifically, client authentication, timeout settings, and dialogue settings are performed. For example, when a client makes a call for the first time, the client's context is managed based on the Rasa framework; when the client makes a second call, if it is determined that the previous context is still in the Rasa session, the session timeout setting and dialogue reset can be performed on the client's context, thus avoiding inaccurate algorithm results.

[0154] Furthermore, embodiments of this disclosure can encapsulate the text content obtained by ASR semantic transfer into... Figure 5In the userMessage shown, the next step is to enter the Rasa strategy module (dialogue strategy design). This decision module can integrate pre-trained models. The Rasa strategy module can design actions that can be triggered by multi-turn dialogue templates, thereby triggering the pre-integrated model to perform actions to generate communication prompts based on the slot filling status of multi-turn dialogue templates.

[0155] Specifically, after acquiring the dialogue data of each round of conversation between the second user and the first user, this embodiment of the disclosure can perform a speech recognition operation to convert the dialogue data into text, i.e., obtain the dialogue text. Then, it performs intent recognition and entity extraction on the dialogue text and the tag information of the second user, and fills the recognized intent and extracted entities into the slots of the multi-round dialogue template (also known as the script flow template stries.md). This process enables the storage of multi-round dialogue information.

[0156] Furthermore, during the process of populating the intent information and entity information into the multi-turn dialogue template, this embodiment of the disclosure can detect whether the multi-turn dialogue template has been triggered, that is, determine whether the template population is complete, and submit the entire policy triggering action after the template population is complete. The action triggered by the above-mentioned multi-turn dialogue can be based on a pre-integrated model to obtain communication prompt information.

[0157] After performing secondary development on the Rasa framework (using Rasa for dialogue management) and obtaining the model, this embodiment of the disclosure can integrate the model into the secondary-developed Rasa framework. The specific process is as follows: Figure 6 As shown. After the model is trained, it can be integrated into Rasa offline. Based on this, online deployment, as well as operations such as speech recommendation and psychographic description, can be performed. Figure 6 The high-quality collection scripts mentioned above are an example of the communication scripts involved in the model training process in the above embodiments. For example, the classification of user characteristics based on psychology can be referred to the description of customer level tags above.

[0158] This disclosure implements dialogue management based on Rasa. Through Rasa's slot filling, entity recognition, and entity extraction, the development efficiency of dialogue management can be increased. Furthermore, by combining Rasa with an auxiliary collection model and applying it to financial debt collection follow-up scenarios, customer psychological constructs can be completed through configuration and context.

[0159] In some implementations, after generating the communication prompt information, the present disclosure embodiments can visualize the communication prompt information and the first tag information output, such as displaying psychological tags, business tags, and appeal point tags to the first user, which can help the first user judge the accuracy of the communication script suggestions.

[0160] Specifically, communication prompts and first tag information can be output to the first electronic device where the first user is located. The display interface of the first electronic device can be as follows: Figure 7 As shown. Figure 7 The mention of "complaint points" refers to examples of the aforementioned complaint point tags; "loan tags" refer to examples of the aforementioned business tags; "psychological characteristics" refer to examples of the aforementioned psychological tags; and "communication suggestions" refer to examples of the aforementioned communication prompts. Based on Figure 7 It is clear that the risk of a complaint from a second user is extremely high when discussing laws and regulations. In this case, the first user can adjust their communication strategy in a timely manner based on this point to avoid a complaint.

[0161] Furthermore, through business tags, the first user can clearly understand the second user's financial situation, thereby assessing the ease of payment follow-up and developing a corresponding payment follow-up strategy. Psychological characteristics allow for a preliminary understanding of the second user's personality and interests, enabling tailored payment follow-up. Moreover, communication suggestions enable the first user to communicate more accurately and easily with the second user, ultimately facilitating payment follow-up.

[0162] In this embodiment, if the second user is different, the corresponding communication suggestions, psychological characteristics, business tags, and points of contention may also be different. For example, when the second user is customer B, the corresponding communication suggestions, psychological characteristics, business tags, and points of contention can be as follows: Figure 8 As shown, Figure 8 and Figure 7 By comparison, we can see that if the second user is different, the corresponding content displayed will also be different, thus enabling more targeted tracking of accounts.

[0163] It should be noted that the communication prompts in this embodiment can be adjusted in real time according to the content of the dialogue data, that is, the content displayed on the interface for each round of dialogue can be different.

[0164] Alternatively, embodiments of this disclosure can also monitor the tracking of violent debts, that is, monitor keywords related to violent debt tracking in the dialogue data, and notify the first user without using a prompt marker, as detailed below. Figure 9 As shown, this approach can reduce the complaint rate while assisting with payment follow-up. Keywords related to aggressive payment follow-up may include insulting or verbally abusing customers.

[0165] Through the above technical solution, this embodiment of the disclosure can recommend excellent scripts, communication script suggestions, user psychological characteristics and user profiles to the first user in the context of the dialogue between the second user and the first user in the scenario of payment follow-up, so as to assist the first user in payment follow-up, and can appropriately remind the customer of legal provisions and other clauses to put pressure on the customer, which can greatly improve the efficiency of payment follow-up.

[0166] This disclosure discloses a multi-turn dialogue system built on Rasa. The efficient and fast ensemble algorithm reduces development costs while enabling intent recognition and slot filling of extracted entities for context management. Furthermore, by extracting features from basic customer information for psychological classification based on customer profiles, and training the algorithm model with selected high-quality dialogue, an auxiliary communication model is obtained. When the first user provides services to the second user, recommended dialogue, user profiles, loan status, user psychological state analysis, and complaint status analysis are provided in real time, which can increase the first user's repayment rate to customers to a certain extent.

[0167] Furthermore, by using Rasa, artificial intelligence (AI), and psychological assistance in accounts receivable tracking, the embodiments of this disclosure can greatly improve the efficiency of accounts receivable tracking personnel, increase the collection rate, and effectively avoid complaints. Based on the sales techniques of excellent agents and customer psychology analysis from different dimensions, the embodiments of this disclosure can have a positive effect on accounts receivable tracking, collection, and complaint handling.

[0168] In one specific implementation, the data processing method may include the following steps: Step 1: Obtain training samples.

[0169] Step 2: Train a model for generating communication prompts based on training samples.

[0170] After training the model, this embodiment of the present disclosure can integrate the model into the Rasa framework. The Rasa framework can be a secondary development framework, where secondary development can be achieved through Rasa dialogue management. Based on this, this embodiment of the present disclosure can deploy the Rasa framework online and utilize it for dialogue recommendation. The detailed process is as follows... Figure 10 As shown. Here, the model can be an AI algorithm. This embodiment of the disclosure can achieve dialogue management by utilizing AI algorithms and Rasa DM. When following up on the first user's bill, the integrated algorithm recommends scripts to assist in the follow-up, enabling different bill follow-up strategies to be adopted for different user groups with different characteristics.

[0171] It should be noted that in this embodiment of the disclosure, Rasa dialogue management can be performed after the model training is completed, as detailed below. Figure 6As shown, Rasa dialogue management can also be performed after semantic recognition (intent recognition), as detailed below. Figure 10 As shown. There are no explicit restrictions on when Rasa conversation management should be performed; it can be chosen based on the actual situation.

[0172] Step 3: Obtain the dialogue data between the first user and the second user.

[0173] Step 4: Determine the tag information for the second user.

[0174] Step 5: Perform intent recognition and entity extraction on the dialogue data and tag information.

[0175] Step 6: Fill the slots in the multi-turn dialogue template with the identified intent information and extracted entity information.

[0176] Step 7: Based on the slot filling status of the multi-turn dialogue template, call the model to obtain the communication prompt information output by the model.

[0177] Step 8: Output communication prompts so that the first user can provide services to the second user based on the communication prompts.

[0178] This disclosed embodiment analyzes the second user's characteristics across multiple dimensions, such as age, education level, and mortgage information, to develop a payment follow-up plan. Furthermore, it uses psychological methods to label the customer's psychology, providing the first user with information such as whether the customer is irritable, prone to silence, or in a good mood. This allows the first user to gain an initial understanding of the second user, facilitating the smooth execution of the payment follow-up plan during the conversation and achieving business goals.

[0179] Furthermore, this embodiment of the disclosure can categorize information that customers are prone to complaining about using historical data, such as mentions of the China Banking Regulatory Commission or data indicating a potential complaint, for the first user to use, thus preventing customer complaints. Simultaneously, based on the communication skills of experienced agents, this embodiment of the disclosure can provide the first user with communication suggestions, such as using keywords like legal provisions to advise the first user to warn the second user from a legal perspective, thereby achieving the goal of tracking payments.

[0180] By combining the Rasa framework for dialogue management, this embodiment of the disclosure can integrate intent recognition and triggering strategies into a multi-turn dialogue system, and trigger different strategies through Action control. These strategies include calling the model to generate communication prompts and real-time updates of the second user's complaint tags, business tags, and psychological tags, thereby enabling the first user to give a reasonable response to the second user's dialogue in a timely manner, so as to assist in the tracking of accounts receivable.

[0181] The following describes a dialogue flow of the data processing method according to the embodiments of this disclosure, using a specific application scenario.

[0182] In some implementations, taking the aforementioned accounts receivable tracking strategy development scenario as an example, combined with... Figure 11 A dialogue flow provided by an embodiment of this disclosure will be described.

[0183] like Figure 11 As shown, after obtaining multiple seats to be screened, this embodiment of the disclosure can calculate the total seat score of these seats by extracting their seat features, and use the total seat score to screen out the excellent seats from the seats to obtain the excellent seat scripts (seat scripts).

[0184] Similarly, after obtaining multiple sample customers, this embodiment of the disclosure can extract the features of these sample customers and calculate the total customer score based on these features. Based on this, each sample customer is tagged according to the total customer score to obtain customer tag information (such as the aforementioned level tags).

[0185] Based on this, a model is trained using agent dialogue and customer tagging information, and then integrated into the Rasa framework. After acquiring the dialogue data between the target customer and the agent, speech recognition, intent recognition, and slot filling can be performed on the dialogue data. After slot filling is completed, the slot information and the integrated model are used to obtain payment follow-up information, which is then displayed on the device held by the agent (the first electronic device).

[0186] As a specific implementation method, this embodiment of the disclosure can extract historical customer calls and organize them into three dimensions: call data, customer basic information, and complaint information. Based on this, the extraction of excellent agent scripts, acquisition of customer basic information, and analysis of complaint characteristics are achieved through the analysis of these data features.

[0187] Among them, basic customer information can be obtained through tag classification, such as age, gender, education, whether there is a mortgage or car loan, etc.; user complaint feature analysis can include negative words such as mentioning the China Banking Regulatory Commission, verbal abuse during the process, and violent debt collection.

[0188] After obtaining the aforementioned features, embodiments of this disclosure can summarize them into several labels through psychological analysis to extract data, and then train the model. Optionally, embodiments of this disclosure can label the data features to assist agents in summarizing and generalizing afterward, improving agent work efficiency and enhancing agent experience.

[0189] Furthermore, embodiments of this disclosure can be used to further develop the Rasa framework and integrate a real-time recommendation algorithm (auxiliary account tracking model) into the Rasa framework. The integrated Rasa framework can then be used for intent recognition and for DM dialogue management. Here, intent recognition can employ a classification model, specifically the BRD model; that is, intent recognition can be performed using the BRD model.

[0190] After completing the above operations, this embodiment of the disclosure can deploy Rasa and AI (Assisted Accounts Receivable Tracking Model). When an agent tracks accounts payable through the call center, this embodiment of the disclosure can convert speech into ASR text and transmit the ASR text to the Rasa framework via HTTP for intent recognition. After the intent is recognized, the entity information is controlled through Rasa's domain to perform context management through Rasa.

[0191] Here, after recognizing the intent, this embodiment can fill slots in Rasa until an Action is triggered. This Action can be inputting slot information into an auxiliary accounts receivable tracking model to obtain accounts receivable tracking information. Optionally, before triggering the Action, this embodiment can first configure the stories.md template in the Rasa framework and guide the agent to fill in the slot information through the Rasa framework.

[0192] As described above, once the slots are filled, the algorithm model can be triggered via an Action to return the algorithm-recommended information to the agent, thereby assisting the agent in following up on accounts. Here, the algorithm model can involve user profiles, communication suggestions, loan tag information, psychological characteristics, and appeal points. These information can correspond to different algorithm models, and different algorithm models can trigger different scripts, slot-filling information, and templates.

[0193] Based on the same concept, embodiments of this disclosure also provide a data processing apparatus, which can be part or all of an electronic device through software, hardware, or a combination of software and hardware. For example... Figure 12 As shown, the data processing device 1200 may include an acquisition module 1210, an extraction module 1220, a determination module 1230, and a generation module 1240: The acquisition module is configured to acquire the first dialogue data between the first user and the second user within a first time period. The extraction module is configured to perform intent recognition and entity extraction on the first dialogue data to obtain first intent information and first entity information. The determining module is further configured to determine the first tag information of the second user based on the first intent information and the first entity information; the first tag information is used to represent the business needs of the second user and the emotional state of the second user; The generation module is configured to generate communication prompts based on the first dialogue data and the first tag information.

[0194] In some implementations, the determining module 1220 is specifically configured as follows: If the difference between the first intent information and the second intent information satisfies the first difference condition, and the difference between the first entity information and the second entity information satisfies the second difference condition, the second tag information of the second time period is determined as the first tag information of the second user; wherein, the second tag information of the second time period is determined based on the intent information and entity information identified from the dialogue data within the second time period; If the difference between the first intent information and the second intent information does not meet the first difference condition, and / or the difference between the first entity information and the second entity information does not meet the second difference condition, the second tag information is updated according to the first intent information and the first entity information to obtain the first tag information.

[0195] In some implementations, the first tag information includes the business needs of the second user; the generation module 1240 is further configured to generate communication prompt information based on the first dialogue data and the first tag information when the business needs of the second user belong to a first type of need; and to generate communication prompt information based on the first dialogue data when the business needs of the second user belong to a second type of need.

[0196] In some implementations, the generation module 1240 is further configured to perform semantic analysis on the first dialogue data and determine a set of dialogues that match the semantic analysis results; determine the banned dialogues corresponding to the second user based on the first tag information, and determine the available dialogues corresponding to the second user based on the banned dialogues and the set of dialogues; and generate communication prompt information based on the available dialogues.

[0197] In some implementations, the first tag information includes the second user's emotion tag, and the generation module 1240 is further configured to generate communication prompt information based on the first dialogue data and reassuring phrases when the emotion tag indicates that the second user's emotion is a negative emotion type; and to generate communication prompt information based on the first dialogue data and efficiency phrases when the emotion tag indicates that the second user's emotion is a positive emotion type.

[0198] In some implementations, the acquisition module 1210 is further configured to acquire third tag information of the first user, the third tag information being used to represent the service capabilities of the first user; and, if the third tag information meets the communication prompt conditions, acquire first dialogue data between the first user and the second user within a first time period.

[0199] Based on the same concept, embodiments of this disclosure also provide an electronic device, including: A storage device on which computer programs are stored; A processing device for executing the computer program in the storage device, in accordance with the steps of any of the above data processing methods.

[0200] Figure 13 This illustrates an electronic device suitable for implementing embodiments of the present disclosure (e.g., Figure 1 A structural diagram of the electronic device or server in the 1300.

[0201] Electronic device 1300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 1301, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1302 or a program loaded from storage device 1308 into random access memory (RAM) 1303. RAM 1303 also stores various programs and data required for the operation of electronic device 1300. Processing device 1301, ROM 1302, and RAM 1303 are interconnected via bus 1304. Input / output (I / O) interface 1305 is also connected to bus 1304.

[0202] Typically, the following devices can be connected to I / O interface 1305: input devices 1306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 1307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1309. Communication device 1309 allows electronic device 1300 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 13 An electronic device 1300 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0203] Based on the same concept, embodiments of this disclosure also provide a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above data processing methods.

[0204] Based on the same concept, embodiments of this disclosure also provide a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above data processing methods.

[0205] It should be understood that the above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0206] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0207] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative forms of implementing the claims. Regarding the apparatus in the above embodiments, the specific manner in which the various modules perform their operations has been described in detail in the embodiments relating to the method, and will not be elaborated upon here.

Claims

1. A data processing method, characterized by, include: Obtain the first dialogue data between the first user and the second user within the first time period; The first dialogue data is subjected to intent recognition and entity extraction to obtain first intent information and first entity information; The first tag information of the second user is determined based on the first intent information and the first entity information; The first tag information is used to represent the second user's business needs and the second user's emotional state; A communication prompt message is generated based on the first dialogue data and the first tag information.

2. The method according to claim 1, characterized in that, Determining the first tag information of the second user based on the first intent information and the first entity information includes: If the difference between the first intent information and the second intent information satisfies the first difference condition, and the difference between the first entity information and the second entity information satisfies the second difference condition, the second tag information of the second time period is determined as the first tag information of the second user; wherein, the second tag information of the second time period is determined based on the intent information and entity information identified from the dialogue data within the second time period; If the difference between the first intent information and the second intent information does not meet the first difference condition, and / or the difference between the first entity information and the second entity information does not meet the second difference condition, the second tag information is updated according to the first intent information and the first entity information to obtain the first tag information.

3. The method according to claim 1, characterized in that, The first tag information includes the second user's business needs; the step of generating communication prompt information based on the first dialogue data and the first tag information includes: If the second user's business needs belong to the first type of needs, a communication prompt message is generated based on the first dialogue data and the first tag information; If the second user's business needs fall under the second type of needs, a communication prompt message is generated based on the first dialogue data.

4. The method according to any one of claims 1 to 3, characterized in that, The step of generating communication prompt information based on the first dialogue data and the first tag information includes: Perform semantic analysis on the first dialogue data and determine the set of utterances that match the semantic analysis results; Based on the first tag information, the prohibited dialogue words corresponding to the second user are determined, and based on the prohibited dialogue words and the dialogue word set, the available dialogue words corresponding to the second user are determined; Communication prompts are generated based on the available dialogue options.

5. The method according to any one of claims 1-3, characterized in that, The first tag information includes the second user's sentiment tag, and the step of generating communication prompt information based on the first dialogue data and the first tag information includes: When the emotion tag indicates that the second user's emotion is a negative emotion type, communication prompt information is generated based on the first dialogue data and reassuring language. When the emotion tag indicates that the second user's emotion is a positive emotion type, communication prompts are generated based on the first dialogue data and efficiency-related phrases.

6. The method according to any one of claims 1-3, characterized in that, The step of obtaining the first dialogue data between the first user and the second user within the first time period includes: Obtain the third tag information of the first user, the third tag information being used to represent the service capabilities of the first user; If the third tag information meets the communication prompt conditions, the first dialogue data between the first user and the second user within the first time period is obtained.

7. A data processing apparatus, characterized in that, include: The acquisition module is configured to acquire the first dialogue data between the first user and the second user within a first time period. The extraction module is configured to perform intent recognition and entity extraction on the first dialogue data to obtain first intent information and first entity information. The determining module is also configured to determine the first tag information of the second user based on the first intent information and the first entity information; The first tag information is used to represent the second user's business needs and the second user's emotional state; The generation module is configured to generate communication prompts based on the first dialogue data and the first tag information.

8. A computer-readable medium having a computer program stored thereon, characterized in that, When executed by a processing device, the computer program performs the steps of the method according to any one of claims 1-6.

9. An electronic device, characterized in that, include: A storage device on which computer programs are stored; A processing device for executing the computer program in the storage device to implement the steps of the method according to any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-6.

Images