Voice telephone AI customer extension system and method

Abstract

The invention belongs to an intelligent information technology, and discloses a voice telephone AI (Artificial Intelligence) customer extension system and a voice telephone AI customer extension method. The system comprises a data acquisition and preprocessing module, a voice interaction module, an intelligent decision-making module, a client management module, a system optimization module and the like. A multi-modal fusion speech emotion recognition algorithm, an intelligent dialogue strategy optimization algorithm based on reinforcement learning, a knowledge graph-driven accurate customer portrait construction method, a real-time feedback adaptive verbal skill generation mechanism and a block chain-based customer data security and privacy protection mechanism are adopted. The method completes the whole process of customer expansion based on the system. According to the invention, customer expansion efficiency, accuracy and customer experience can be improved, customer data security is guaranteed, and powerful support is provided for enterprise customer expansion.

Description

Technical Field

[0001] The present invention relates to the intersection of artificial intelligence and marketing, and in particular to a voice telephone AI customer development system and method. Background Art

[0002] Traditional voice call acquisition relies primarily on manual labor, which presents numerous significant limitations. First, inefficiency is a prominent issue. Salespeople are limited in the number of effective calls they can make each day, and they also spend a significant amount of time recording and analyzing information during the call process. This results in low overall efficiency and makes it difficult to quickly reach a large number of potential customers. Second, costs are high. Companies need to recruit, train, and manage a large number of sales personnel, which undoubtedly increases labor costs. Furthermore, manual outbound calls have a high error rate, which can lead to unnecessary losses for the company. Finally, they are significantly affected by human factors. Factors such as the salesperson's mood, experience, and communication skills can all affect customer acquisition results, making it difficult to ensure the accuracy of customer acquisition and achieve standardized and scalable marketing.

[0003] With the development of artificial intelligence technology, a number of AI customer acquisition systems have emerged on the market, but these systems still face some urgent challenges. In terms of voice interaction, existing systems' speech recognition accuracy is insufficient, especially when faced with complex accents, background noise, and other situations, where recognition results are often unsatisfactory. Speech synthesis also lacks naturalness and emotional expressiveness, making human-computer dialogue appear stiff and mechanical, and difficult to provide a good customer experience. When it comes to judging customer intent, most systems rely solely on simple rule matching or single-dimensional data, such as call duration and keyword frequency. These systems are unable to comprehensively and accurately analyze customers' potential needs and purchasing intentions, thus affecting the accuracy and efficiency of customer acquisition. Furthermore, existing systems generally lack real-time optimization mechanisms, making it impossible to adjust conversation strategies and language based on real-time customer feedback, making it difficult to adapt to ever-changing market and customer demands. Summary of the Invention

[0004] Overall System Architecture: This AI-powered voice telephony customer acquisition system primarily consists of a data acquisition and preprocessing module, a voice interaction module, an intelligent decision-making module, a customer management module, and a system optimization module. These modules collaborate to form an integrated whole, automating and intelligently implementing the entire process from customer data collection to targeted customer acquisition.

[0005] System workflow

[0006] The workflow of this voice phone AI customer acquisition system is as follows:

[0007] 1. Data collection and preprocessing: The data collection and preprocessing module collects customer data through various channels and performs preprocessing operations such as cleaning, deduplication, normalization, and labeling to provide high-quality data for subsequent analysis and decision-making.

[0008] 2. Voice Interaction: The voice interaction module conducts voice conversations with customers, recognizing their voice in real time and synthesizing responses. During the conversation, multimodal speech recognition algorithms and emotional speech synthesis technology are used to achieve natural and smooth human-computer interaction and analyze the customer's emotional state.

[0009] 3. Intelligent Decision-Making: The intelligent decision-making module uses customer intention assessment models and dialogue strategy optimization models based on real-time customer feedback and historical data to analyze customer intentions and needs, dynamically adjust dialogue strategies, and select the most appropriate response.

[0010] 4. Customer Management: The customer management module classifies and manages customers based on customer intention assessment results, develops personalized follow-up plans for different categories of customers, and assigns follow-up tasks to sales staff.

[0011] 5. System optimization: The system optimization module collects system operation data in real time, performs performance evaluation and optimization on the algorithm model, and updates and improves the speech template library to improve the performance and effectiveness of the system.

[0012] Advantageous Effects of the Invention

[0013] The voice telephone AI customer acquisition system and method of the present invention has the following significant beneficial effects:

[0014] Improve customer acquisition efficiency: Through automated voice interaction and intelligent decision-making, the system can quickly screen potential customers, improve outbound call efficiency, and reduce manual operation time and workload. Furthermore, real-time adjustments to conversation strategies can better meet customer needs, increase customer response rates and conversion rates, and significantly improve customer acquisition efficiency.

[0015] Improved Accuracy: Utilizing multi-dimensional data analysis and deep learning algorithms, this system can more accurately assess customer intent, providing sales staff with more precise customer information and follow-up recommendations. Multimodal speech recognition algorithms and sentiment analysis technologies can better understand customer intent and emotional state, further improving customer acquisition accuracy.

[0016] Improve customer experience: Natural and smooth voice interaction and personalized communication methods can make customers feel more intimate and authentic service experience, and enhance their goodwill and trust in the company.

[0017] Description of the drawings: Figure 1 This is a flow chart of the working algorithm of the system of the present invention. DETAILED DESCRIPTION

[0018] Example 1

[0019] 1. A voice telephone AI customer development system, including a data acquisition and preprocessing module, a voice interaction module, an intelligent decision-making module, a customer management module and a system optimization module. Each module works together to achieve customer development functions. The voice interaction module adopts a multimodal fusion voice emotion recognition algorithm to perform sentiment analysis based on the acoustic features of the voice and the semantic information of the text. The specific steps include feature extraction, feature fusion and sentiment classification. In the feature fusion process, the acoustic feature vector is {A} = [a_1, a_2, a_n], the semantic feature vector is {S} = [s_1, s_2, s_m], the fused feature vector is {F}, and the acoustic feature weights are Semantic feature weight For sentiment classification, let the classifier weight matrix be {W}, the bias vector be {b}, the sentiment category be C, and the predicted sentiment category probability P = Softmax(WF+b)

[0020] The intelligent decision-making module uses the intelligent dialogue strategy optimization algorithm based on reinforcement learning to define the state, action and reward, and trains the intelligent agent through the deep Q network algorithm to achieve real-time optimization of the dialogue strategy; wherein, the loss function Target value y = r + γmax a′ Q′(s′, a′; θ - ), r is the discount factor, and D is the experience replay buffer. The network parameter θ is updated by minimizing the loss function to optimize the conversational strategy. A knowledge graph-driven approach is used to build precise customer profiles. Multi-source data is collected to construct a knowledge graph, and a graph neural network algorithm is used to generate precise customer profiles. An adaptive speech generation mechanism with real-time feedback, including a speech template library, real-time feedback analysis, and speech adjustment steps, can generate personalized speech based on customer feedback. A blockchain-based customer data security and privacy protection mechanism is employed, including encrypted data storage, access control, and privacy protection measures.

[0021] A voice telephony AI customer acquisition method includes the following steps: a data acquisition and preprocessing module collects and processes customer data; a voice interaction module conducts voice conversations with customers and analyzes their emotional state; an intelligent decision-making module provides intelligent decision support based on customer status and needs; a customer management module manages customer information and follows up on tasks; and a system optimization module optimizes the system based on feedback. During the voice interaction process, a multimodal voice emotion recognition algorithm is executed; during the intelligent decision-making process, a reinforcement learning-based intelligent dialogue strategy optimization algorithm is executed.

[0022] Implementation of data acquisition and preprocessing modules

[0023] In practical applications, the data acquisition and preprocessing module can collect and preprocess data in a variety of ways. For internal enterprise data, it can directly obtain relevant customer information by connecting with the company's existing business systems (such as CRM systems, ERP systems, etc.). For external data, web crawler technology can be used to collect public information from customers on social media, industry forums, and other platforms from the internet. During the data preprocessing stage, data cleaning tools can be used to clean the data to remove duplicate, erroneous, and incomplete data; normalization algorithms can be used to normalize the data to make it comparable; and annotation tools can be used to annotate the data to provide labels for subsequent machine learning model training.

[0024] Implementation of voice interaction module

[0025] The speech recognition function of the speech interaction module can adopt an open source speech recognition engine (such as Baidu speech recognition, iFlytek speech recognition, etc.), and be optimized in combination with the multimodal speech recognition algorithm of the present invention. In actual applications, appropriate speech recognition engines and algorithm parameters can be selected according to different business scenarios and customer needs. The speech synthesis function can use speech synthesis technology (such as the Tacotron model based on deep learning) and combine it with emotional speech synthesis technology to generate voice responses with natural emotional expression. During the implementation process, a large amount of speech data can be used for training to continuously optimize the effect of speech synthesis.

[0026] Implementation of intelligent decision-making modules

[0027] The customer intention assessment model in the intelligent decision-making module can be built using deep learning algorithms (such as neural networks and decision trees). During training, a large amount of historical customer data can be used to continuously adjust model parameters and improve model accuracy and stability. The dialogue strategy optimization model can be trained using reinforcement learning algorithms (such as deep Q networks and policy gradient algorithms). In practical applications, the model's strategy can be continuously updated through real-time interaction with customers, achieving real-time optimization of the dialogue strategy.

[0028] Implementation of customer management module

[0029] The customer management module can use a database management system (such as MySQL, Oracle, etc.) to store and manage customer information. Customer classification can be based on factors such as customer intention assessment results, purchasing power, and purchase intention. Follow-up plans can be developed based on individual customer categories, creating personalized follow-up strategies and plans. Furthermore, customer relationship management software (such as Salesforce, Zoho CRM, etc.) can be used to manage customer information and assign follow-up tasks.

[0030] Implementation of system optimization module

[0031] The system optimization module can use data analysis tools (such as Python's Pandas and NumPy libraries) to analyze and compile statistics on system operation data. For algorithm model optimization, machine learning frameworks (such as TensorFlow and PyTorch) can be used to train and optimize the models. For updating the speech template library, customer feedback can be collected and analyzed to continuously improve and update speech templates, enhancing their relevance and effectiveness.

[0032] Specific functions and innovations of each module

[0033] Data acquisition and preprocessing module

[0034] This module is responsible for collecting customer-related data from both within and outside the company, including but not limited to basic customer information (such as name, age, gender, contact information, etc.), historical communication records (such as call recordings and chat logs), consumer behavior data (such as purchase amount, purchase frequency, and purchase preferences), and social media data (such as comments and interaction records on social platforms). The collected data is pre-processed through cleaning, deduplication, normalization, and annotation to improve data quality and usability, providing a reliable data foundation for subsequent analysis and decision-making.

[0035] Voice interaction module

[0036] This module integrates speech recognition and speech synthesis technologies, using a multimodal speech recognition algorithm based on an attention mechanism and emotional speech synthesis technology to achieve natural and smooth human-computer dialogue. The multimodal speech recognition algorithm is one of the innovations of this invention. It comprehensively considers the acoustic characteristics of the speech, the semantic information of the text, and other relevant contextual information to more accurately recognize the speech content. The specific steps are as follows:

[0037] Feature extraction: Extract acoustic features from speech signals, such as pitch, duration, timbre, and speaking rate. Simultaneously, use natural language processing technology to extract semantic features from speech-to-text results, such as keywords, parts of speech, and grammatical structure.

[0038] Feature Fusion: Utilizing the attention mechanism in deep learning, we assign weights to different features, achieving effective feature fusion. This approach fully leverages the strengths of different features and improves speech recognition accuracy.

[0039] Sentiment Classification: The fused feature vectors are fed into a pre-trained sentiment classification model to output the customer's emotional state (e.g., positive, negative, neutral, etc.). Emotional speech synthesis technology generates natural, expressive voice responses based on the conversation context and the customer's emotional state, providing a more intimate and authentic communication experience.

[0040] Intelligent decision-making module

[0041] This module builds a customer intention assessment model and a dialogue strategy optimization model based on deep learning and reinforcement learning algorithms, dynamically adjusting dialogue strategies based on real-time customer feedback. The intelligent dialogue strategy optimization algorithm based on reinforcement learning is another innovative feature of this invention. The specific implementation is as follows:

[0042] Define states, actions, and rewards: States include the customer's historical responses, the current conversation turn, and their emotional state. Actions are optional responses that the system can choose. Rewards are set based on customer feedback and conversation goals. For example, positive rewards are given if the customer shows interest, while negative rewards are given if the customer refuses.

[0043] Training the agent: The Deep Q-Network (DQN) algorithm is used to train the agent, which learns the optimal dialogue strategy through continuous interaction with customers.

[0044] Real-time adjustment: In actual conversations, the agent selects actions based on its current state and updates its rewards and strategies based on customer feedback, enabling real-time optimization of the conversation strategy.

[0045] Customer management module

[0046] Establish a customer information management system to manage customers throughout their lifecycle. Categorize customers based on their interest assessment results, such as high-intent, medium-intent, and low-intent, and develop personalized follow-up plans for each category. Also, record customer communication history, changes in interest, and other information to provide sales staff with a comprehensive customer view, facilitating targeted marketing and service delivery.

[0047] System optimization module

[0048] Real-time collection of system operation data allows for performance evaluation and optimization of algorithm models. Through statistical analysis of system operation data, we calculate evaluation metrics such as speech recognition accuracy, customer intent prediction accuracy, and conversation conversion rate. Based on these evaluation results, we adjust and optimize the algorithm model. Furthermore, we update and refine our script template library, providing more appropriate script suggestions based on different customer groups and conversation scenarios to improve conversation effectiveness and customer conversion rates.

Claims

1. A voice phone AI customer acquisition system, characterized by: It includes data collection and preprocessing module, voice interaction module, intelligent decision-making module, customer management module and system optimization module. Each module works together to achieve customer development function.

2. The voice telephone AI customer acquisition system according to claim 1, characterized in that: The speech interaction module adopts a multimodal fusion speech emotion recognition algorithm to perform emotion analysis based on the integrated speech acoustic features and text semantic information. The specific steps include feature extraction, feature fusion and emotion classification. In the feature fusion process, the acoustic feature vector is {A}=[a_1,a_2,,a_n], the semantic feature vector is {S}=[s_1,s_2,,s_m], the fused feature vector is {F}, and the acoustic feature weight is Semantic feature weight Attn(x)=σ(w T x+b), In sentiment classification, let the classifier weight matrix be {W}, the bias vector be {b}, the sentiment category be C, and the predicted sentiment category probability distribution P = Softmax(WF+b).

3. The voice telephone AI customer acquisition system according to claim 1, characterized in that: The intelligent decision-making module uses an intelligent dialogue strategy optimization algorithm based on reinforcement learning to define states, actions, and rewards, and trains the agent through a deep Q-network algorithm to achieve real-time optimization of the dialogue strategy. Target value y = r + γmax a′ Q′(s′,a′;θ - ), r is the discount factor, D is the experience replay buffer, and the network parameters θ are updated by minimizing the loss function to optimize the dialogue strategy.

4. The voice telephone AI customer acquisition system according to claim 1, characterized in that: A knowledge graph-driven method for building accurate customer portraits is adopted. Multi-source data is collected to build a knowledge graph, and graph neural network algorithms are used to generate accurate customer portraits.

5. The voice telephone AI customer acquisition system according to claim 1, characterized in that: An adaptive speech generation mechanism with real-time feedback, including a speech template library, real-time feedback analysis and speech adjustment steps, can generate personalized speech based on customer feedback.

6. The voice telephone AI customer acquisition system according to claim 1, characterized in that: Adopt a blockchain-based customer data security and privacy protection mechanism, including data encryption storage, access control and privacy protection measures.

7. A voice phone AI customer acquisition method, characterized in that: The application of the voice telephone AI customer acquisition system according to any one of claims 1 to 6 comprises the following steps: 1). The data acquisition and preprocessing module collects and processes customer data; 2) The voice interaction module conducts voice conversations with customers and analyzes their emotional state; 3). The intelligent decision-making module provides intelligent decision support based on customer status and needs; 4). The customer management module manages customer information and follows up tasks; 5). The system optimization module optimizes the system based on the feedback.

8. The AI customer acquisition method using voice calls according to claim 7, characterized in that: During the voice interaction process, a multimodal fusion voice emotion recognition algorithm process is executed; during the intelligent decision-making process, an intelligent dialogue strategy optimization algorithm process based on reinforcement learning is executed.

Images