Intelligent voice calling method and system

Abstract

The invention discloses an intelligent voice calling method and system, and the method comprises the steps: extracting client interaction data of a to-be-called client from a client database, the client interaction data comprising client basic data, historical transaction records, call records and feedback information; analyzing the client interaction data, identifying potential demands of clients, and generating client demand features; according to the customer demand characteristics, utilizing a pre-trained customer demand prediction model to predict the interest degree of the customer for a specific product or service, and outputting a customer demand probability; and based on the customer interaction data and the customer demand probability, setting a call strategy with a to-be-called customer, and performing call operation according to the call strategy, the call strategy including call time, call content and call style. By using the embodiment of the invention, accurate calling can be realized through an intelligent means, the customer service quality is improved, and a more scientific and efficient calling strategy is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of voice calling, and in particular to an intelligent voice calling method and system. Background Art

[0002] With the advent of the digital age, companies are facing increasing challenges in customer relationship management. Traditional telemarketing and customer service methods usually rely on manual judgment and experience, which is not only inefficient but also prone to customer churn. Customer needs change rapidly, and companies need to respond quickly to meet customers' personalized needs. However, traditional methods often fail to make full use of customer data, resulting in an inability to accurately identify potential customer needs. Summary of the invention

[0003] The purpose of the present invention is to provide an intelligent voice calling method and system to solve the deficiencies in the prior art, to achieve precise calling through intelligent means, to improve customer service quality, and to implement a more scientific and efficient calling strategy.

[0004] An embodiment of the present application provides an intelligent voice calling method, the method comprising: Extracting customer interaction data of the customer to be called from the customer database, wherein the customer interaction data includes: basic information of the customer, historical transaction records, call records and feedback information, wherein the historical transaction records are transaction records for corresponding products or services reached by the customer through historical calls, and the call records are historical call records between the calling end and the customer; Analyze the customer interaction data, identify potential customer needs, and generate customer demand characteristics; According to the customer demand characteristics, using a pre-trained customer demand prediction model, predict the customer's interest in a specific product or service and output the customer demand probability; Based on the customer interaction data and the customer demand probability, a call strategy for the customer to be called is set, and a call operation is performed according to the call strategy, wherein the call strategy includes: call time, call content and call style.

[0005] Optionally, analyzing the customer interaction data, identifying potential customer needs, and generating customer demand features includes: Extract user portrait features from basic customer information to obtain user portrait features; Use natural language processing technology to perform semantic analysis on call records and feedback information, extract keywords, and obtain keyword features; Use word frequency-inverse text frequency to quantify the extracted keywords and obtain keyword weights; The keyword features are weighted using the keyword weights to obtain weighted keyword features; The historical transaction features in the historical transaction records are extracted, and the customer demand features are determined according to the user portrait features, the weighted keyword features and the historical transaction features.

[0006] Optionally, the predicting the customer's interest in a specific product or service based on the customer demand characteristics using a pre-trained customer demand prediction model and outputting a customer demand probability includes: Inputting the customer demand features into a pre-trained customer demand prediction model, and outputting a series of interest scores related to specific products or services, wherein the customer demand prediction model includes: a random forest, XGBoost or a deep learning model, and the interest scores represent the potential demand intensity of the customer for each product or service; The interest level score output by the model is normalized through the sigmoid function or the softmax function and converted into the corresponding customer demand probability value.

[0007] Optionally, the setting of a calling strategy for the customer to be called based on the customer interaction data and the customer demand probability includes: According to the call records in the customer interaction data, select a time period with the highest answering rate as the call time; Prioritize each product or service based on the probability of customer demand for each product or service, and introduce the prioritized products or services in order as the content of the call; Determine a user portrait corresponding to the customer basic information in the customer interaction data, and select a chat style corresponding to the user portrait as the call style according to a mapping table between user portraits and chat styles.

[0008] Optionally, the use of natural language processing technology to perform semantic analysis on call records and feedback information, extract keywords, and obtain keyword features includes: Performing text preprocessing on call records and feedback information, wherein the text preprocessing includes: text data cleaning, stem extraction and word form restoration; Perform context-aware keyword extraction on the preprocessed text to obtain candidate keywords; Calculate the frequency of occurrence of each candidate keyword in the call records and feedback information, and assign a sentiment score to the candidate keyword according to its performance in the sentiment dictionary, wherein a sentiment dictionary is constructed based on the corpus, and the sentiment dictionary contains positive sentiment words, negative sentiment words and their sentiment values; Select candidate keywords whose occurrence frequency is higher than a preset frequency threshold and whose sentiment score is higher than a preset score threshold as final keywords; Use K-means or hierarchical clustering algorithm to group the final keywords according to similarity to identify potential topics or categories, and extract representative final keywords from each clustering result as feature words to form the final keyword feature set.

[0009] Optionally, the method of using word frequency-inverse text frequency to quantify the extracted keywords to obtain keyword weights includes: Use the TF-IDF algorithm to calculate the initial weight of the extracted keywords; An undirected graph is constructed according to the word co-occurrence relationship of the keywords. Based on the undirected graph, the importance of the words is iteratively calculated using the TextRank algorithm to obtain the final weight of the keywords, wherein each node of the undirected graph represents a keyword, the existence of an edge indicates that the two keywords have appeared in the same context, and each node is assigned an initial weight of the corresponding keyword.

[0010] Another embodiment of the present application provides an intelligent voice call system, the system comprising: An extraction module is used to extract customer interaction data of the customer to be called from the customer database, wherein the customer interaction data includes: basic information of the customer, historical transaction records, call records and feedback information, wherein the historical transaction records are transaction records for corresponding products or services reached by the customer through historical calls, and the call records are historical call records between the calling end and the customer; An analysis module, used to analyze the customer interaction data, identify potential customer needs, and generate customer demand characteristics; A prediction module is used to predict the customer's interest in a specific product or service based on the customer demand characteristics and use a pre-trained customer demand prediction model to output the customer demand probability; The calling module is used to set a calling strategy with the customer to be called based on the customer interaction data and the customer demand probability, and perform a calling operation according to the calling strategy, wherein the calling strategy includes: call time, call content and call style.

[0011] Yet another embodiment of the present application provides a storage medium, wherein the storage medium stores a computer program, wherein the computer program is configured to execute any of the above methods when running.

[0012] Yet another embodiment of the present application provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute any of the methods described above.

[0013] Compared with the prior art, the present invention provides an intelligent voice calling method, which extracts customer interaction data of customers to be called from a customer database, wherein the customer interaction data includes: customer basic information, historical transaction records, call records and feedback information; analyzes the customer interaction data, identifies customer potential needs, and generates customer demand characteristics; based on the customer demand characteristics, uses a pre-trained customer demand prediction model to predict the customer's interest in a specific product or service, and outputs the customer demand probability; based on the customer interaction data and the customer demand probability, sets a calling strategy with the customer to be called, and performs a calling operation according to the calling strategy, wherein the calling strategy includes: call time, call content and call style, so that precise calling can be achieved through intelligent means, customer service quality can be improved, and a more scientific and efficient calling strategy can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A hardware structure block diagram of a computer terminal for an intelligent voice calling method provided by an embodiment of the present invention; Figure 2 A flowchart of an intelligent voice calling method provided by an embodiment of the present invention; Figure 3 A schematic diagram of the structure of an intelligent voice calling system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, but should not be construed as limiting the present invention.

[0016] The embodiment of the present invention firstly provides an intelligent voice calling method, which can be applied to electronic devices such as computer terminals, specifically ordinary computers, etc.

[0017] The following describes it in detail by taking running on a computer terminal as an example. Figure 1 The hardware structure block diagram of a computer terminal of an intelligent voice calling method provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0018] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any intelligent voice call method.

[0019] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0020] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any intelligent voice call method.

[0021] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0022] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0023] See also Figure 2 , an embodiment of the present invention provides an intelligent voice calling method, which may include the following steps: S201, extracting customer interaction data of the customer to be called from a customer database, wherein the customer interaction data includes: basic information of the customer, historical transaction records, call records and feedback information; In the present invention, extracting the customer interaction data of the customer to be called from the customer database is the first step of the intelligent voice call method. The customer interaction data includes the customer's basic information, historical transaction records, call records and feedback information. The basic customer information provides the customer's identity information and preferences, helping the system to establish a user portrait; the historical transaction records reflect the customer's consumption behavior and preferences, providing a basis for further demand analysis; the call records record the communication between the customer and the enterprise, which helps to understand the customer's concerns and demand changes; the feedback information provides the customer's direct opinions and suggestions on the product or service. The comprehensive extraction of these data lays a solid foundation for subsequent demand analysis and call strategy formulation, ensuring that the system can fully understand the customer's personalized needs. Among them, the historical transaction records are specifically the transaction records for the corresponding products or services reached by the customer through historical calls, and the call records are specifically the historical call records between the call end and the customer. These data can be obtained from the customer's public database in actual applications to ensure data privacy.

[0024] In specific implementation, the system first accesses the customer database through the database interface to screen out all customers to be called. Then, the system uses preset query conditions to integrate each customer's basic information (such as name, contact information, age, gender, etc.), historical transaction records (such as purchased products, transaction time, transaction amount, etc.), call records (such as past call content, call duration, answering status, etc.) and feedback information (such as customer satisfaction survey results, product evaluation, etc.) into a complete customer interaction data set. The data set is then stored in memory for subsequent customer demand analysis and behavior prediction models. In this way, the intelligent voice call system can efficiently and accurately obtain customer interaction data, providing comprehensive information support for subsequent analysis and decision-making.

[0025] By extracting customer interaction data of customers to be called, the intelligent voice calling method can accurately identify and predict customer needs, and then formulate personalized calling strategies. This can not only improve customer service efficiency and reduce customer churn, but also have a positive impact on customer experience. Since each customer's needs and preferences are different, comprehensive analysis of customers' basic information, historical transaction records and call feedback information will help companies better identify customers' potential needs, thereby increasing the probability and satisfaction of customer responses. At the same time, this data-driven approach can also provide a more scientific basis for the company's marketing decisions, ultimately promoting sales conversion and enhancing corporate competitiveness.

[0026] S202, analyzing the customer interaction data, identifying potential customer needs, and generating customer demand characteristics; The process of analyzing customer interaction data aims to deeply explore the potential needs of customers and generate corresponding customer demand characteristics. This process usually includes comprehensive analysis of basic customer information to establish user portraits; processing of call records and feedback information to extract key semantic information; and combining historical transaction records to identify customer consumption behaviors and preferences. Through these analyses, the system can effectively capture the true intentions of customers in past interactions, thereby converting them into clear demand characteristics and providing data support for subsequent precise call strategies.

[0027] Through systematic analysis of customer interaction data and demand identification, companies can better understand and predict customer behavior and needs. This in-depth insight can not only improve customer service quality and satisfaction, but also provide a scientific basis for the company's marketing strategy, ensuring targeted product and service recommendations, thereby increasing customer purchase intention and conversion rate, and ultimately achieving a win-win situation for customers and companies.

[0028] Specifically, user portrait features can be extracted from the basic information of the customer to obtain user portrait features; In this step, the system extracts basic customer information from the customer database, analyzes information such as age, gender, geographic location, and occupation, and builds a user profile. By aggregating this information, companies can form a holistic understanding of their customers and lay the foundation for subsequent personalized services.

[0029] The establishment of user portraits enables companies to more accurately identify customer needs and preferences during interactions with customers, thereby optimizing marketing strategies and improving customer satisfaction. This not only improves customer experience, but also provides strong data support for companies to formulate targeted market strategies.

[0030] In this implementation step, the system first extracts the basic information of the customer to be analyzed from the customer database. This information may include the customer's name, age, gender, contact number, geographic location, occupation, hobbies, etc. Then, the system preprocesses this data to ensure the integrity and consistency of the data. For example, the system cleans up missing values ​​and outliers and standardizes all character data. Next, through data aggregation and classification technology, the system classifies similar features to form preliminary customer groups. At the same time, the system uses data mining algorithms to identify the common features of different customer groups, thereby constructing a representative user portrait feature set. These user portrait features will provide basic data for subsequent customer demand analysis, helping companies to better understand the characteristics and needs of different customers.

[0031] Use natural language processing technology to perform semantic analysis on call records and feedback information, extract keywords, and obtain keyword features; In this step, the system applies natural language processing (NLP) technology to conduct in-depth analysis of customer call records and feedback information, identify and extract key words and phrases. These keywords can effectively reflect customers' emotional attitudes and needs, helping companies better understand the voice of customers.

[0032] By extracting keywords, companies can quickly identify customers’ main concerns and potential needs, making subsequent communication and service more effective. This method can improve customer satisfaction, ensure that customer opinions and needs are responded to in a timely manner, and thus enhance customer loyalty.

[0033] Specifically, the call records and feedback information may be subjected to text preprocessing, wherein the text preprocessing includes: text data cleaning, stem extraction and word form restoration; Text preprocessing is a basic step in natural language processing, and its purpose is to improve the quality of text for subsequent analysis. Text data cleaning includes removing extra spaces, punctuation marks, and special characters to ensure the neatness of each text. Stemming and lemmatization normalize words with similar meanings, which can reduce the diversity of vocabulary. For example, "run" and "running" are classified as the same root "run". This can reduce the complexity of model training and improve the analysis effect. By cleaning and normalizing the text, the data noise that affects the analysis can be eliminated and the effectiveness of keyword extraction can be enhanced. On this basis, the model will be able to capture the main information in the text more accurately, and then identify customer needs and potential emotions.

[0034] When implementing text preprocessing, the collected call records and feedback information must first be cleaned. Regular expressions are used to remove non-alphanumeric characters and extra spaces. For example, for the call record "I'm going shopping today! Do you think it's okay?", it can be cleaned to "I'm going shopping today. Do you think it's okay?". Next, use natural language processing libraries such as NLTK or spaCy for stem extraction and word form restoration. By dividing the text into words, the Porter or Snowball stem extraction algorithm is applied respectively to classify "buy", "buy" and "bought" into the same root word "buy". Finally, a list of cleaned texts with noise removed and unified word forms is obtained, laying the foundation for subsequent keyword extraction.

[0035] Perform context-aware keyword extraction on the preprocessed text to obtain candidate keywords; Context-aware keyword extraction emphasizes the relative importance of words in a specific text environment. This step is based on the co-occurrence relationship of words in the text, taking into account the semantic association between words, and aims to extract candidate keywords that are most relevant to the topic. To achieve this goal, a sliding window mechanism can be used to check the 5 words before and after each word in the text to generate candidate keywords. Extracting keywords in a context-aware way can enhance the accuracy of keyword extraction and avoid missing words that are significant in a specific context. This ensures that in the subsequent steps, the candidate keywords used have higher information density and relevance, which helps identify the real needs and emotional state of customers.

[0036] When performing context-aware keyword extraction, the preprocessed text is first converted to a word vector. Using the Word2Vec model, a window (such as 5 words) is centered on each word to extract all the keywords in the window. Next, similarity calculations are performed on the word vectors in each window to identify words with higher frequency of occurrence. For example, for the text "I like this product, it is very useful", the word "like" is more related to "product" and "useful" in the context, so these words will be selected as candidate keywords. The final output set of candidate keywords will include words that are closely related and frequently appear in the context.

[0037] Calculate the frequency of occurrence of each candidate keyword in the call records and feedback information, and assign a sentiment score to the candidate keyword according to its performance in the sentiment dictionary, wherein a sentiment dictionary is constructed based on the corpus, and the sentiment dictionary contains positive sentiment words, negative sentiment words and their sentiment values; This step aims to quantify the frequency and sentiment analysis of candidate keywords to provide the necessary basis for subsequent screening. Recording the frequency of candidate keywords in all texts helps identify which words contribute most to the overall theme of the text. At the same time, by establishing a sentiment dictionary, each candidate keyword is assigned a sentiment score to gain a more comprehensive understanding of the emotional tendencies in the text. Through the combined evaluation of frequency and sentiment, irrelevant keywords can be effectively filtered out, and those keywords with positive or negative sentiment tendencies and frequent occurrences in the context can be retained. This lays a solid foundation for subsequent keyword screening and is conducive to improving the accuracy and effectiveness of the final keyword features.

[0038] In this stage, we first conduct a statistical analysis of the candidate keywords and calculate their frequency of appearance in all call records and feedback information. Next, we build a sentiment dictionary, including positive and negative sentiment words, such as "satisfied" and "like" as positive words, "dissatisfied" and "bad" as negative words, and each word is equipped with a corresponding sentiment value. Then, based on this sentiment dictionary, we calculate the sentiment score for each candidate keyword. Assume that "satisfied" appears 10 times in the candidate keywords and is assigned a sentiment value of 1, while "bad" appears 5 times and has a sentiment value of -1. When the frequency and sentiment score are combined, "satisfied" will be considered a keyword that is more likely to represent the customer's positive needs.

[0039] Select candidate keywords whose occurrence frequency is higher than a preset frequency threshold and whose sentiment score is higher than a preset score threshold as final keywords; The screening step aims to further narrow down the candidate keywords by setting thresholds. This process identifies the most representative keywords based on the frequency of keyword occurrence and sentiment score. This means that the keywords that best meet customer needs are screened through the dual criteria of quantity and sentiment. Setting preset frequency and sentiment score thresholds helps remove noise and irrelevant words, ensuring that the keywords in the final vocabulary set are both frequent and can accurately reflect the customer's emotional tendencies. This improves the accuracy and effectiveness of subsequent analysis, making the final extracted keywords more in line with customer needs.

[0040] When implementing the screening, first define the frequency threshold (for example: the number of occurrences must be at least 5 times) and the sentiment score threshold (for example: the sentiment score must be greater than 0). Then, traverse the candidate keywords and check the frequency of occurrence and sentiment score of each keyword. If a keyword such as "satisfied" has a frequency of 10 times and a sentiment score of 0.8, both of which are higher than the set thresholds, it will be included in the final keyword set. However, if a keyword such as "unsatisfied" only appears 3 times and has a sentiment score of -0.5, it will be excluded. Through this step, the system generates a streamlined set of final keywords with high relevance and sentiment representativeness.

[0041] Use K-means or hierarchical clustering algorithm to group the final keywords according to similarity to identify potential topics or categories, and extract representative final keywords from each clustering result as feature words to form the final keyword feature set.

[0042] Keyword clustering is the process of grouping keywords with high similarity, aiming to identify potential themes or categories in the text. Through K-means or hierarchical clustering algorithms, keywords can be effectively grouped into different groups to help clarify the relevance of various keywords and their role in customer needs. Cluster analysis helps to discover potential patterns and themes in customer needs and can further enrich user portraits. For example, through grouping, the analysis team can integrate customers' specific needs and emotional feedback for a certain type of product to provide a basis for subsequent market strategies. This method can improve the depth and accuracy of the overall analysis.

[0043] When implementing clustering, you first need to convert the final keywords into vector form, build a matrix representing the similarity between keywords, and then select K-means or hierarchical clustering algorithm to cluster these keywords. For example, assuming that the final keywords include "satisfaction", "trust", "quality", and "express delivery", these words can be converted into vectors and divided into several groups using the K-means algorithm. After clustering, the system will analyze the keywords in each group and extract the most representative keywords as feature words, such as extracting "satisfaction" from the "satisfaction" and "trust" groups as the feature word of the group. The final keyword feature set is not only directional, but also reflects the customer's preferences for different products or services.

[0044] Use word frequency-inverse text frequency to quantify the extracted keywords and obtain keyword weights; This step applies the term frequency-inverse document frequency (TF-IDF) algorithm to quantify the extracted keywords and assign a weight to each keyword by analyzing the frequency of occurrence of the keyword in a specific document and its rarity in the entire document set. This weight reflects the representativeness of the keyword for a specific customer interaction.

[0045] Quantifying keywords and calculating their weights enables companies to clearly identify the factors that have the greatest impact on customer demand. This data-driven insight not only helps companies focus on key needs, but also promotes more accurate product and service recommendations and improves customer purchasing experience.

[0046] Specifically, the TF-IDF algorithm can be used to calculate the initial weights of the extracted keywords; In this step, the extracted keywords are first quantified using the Term Frequency-Inverse Document Frequency (TF-IDF) algorithm to calculate the importance of each keyword in a specific document. TF represents the frequency of the word in the document, while IDF measures the uniqueness of the word in the entire corpus. By multiplying TF and IDF, we can get a preliminary weight value, which helps us identify keywords that are more important in a specific context. The initial weights obtained in this process lay the foundation for subsequent analysis and processing. By quantifying the weights of keywords using the TF-IDF algorithm, we can better capture the truly important words in the document. This process not only helps identify keywords that are critical to customer demand analysis, but also provides effective data support for subsequent model building and final analysis. The quantified keyword weights will make subsequent demand feature extraction and correlation analysis more accurate, thereby improving the accuracy and reliability of the overall model.

[0047] In the specific implementation, we first need to collect and organize the customer's interaction data, including call records and feedback information in text form. Suppose the keywords we extracted are "service quality", "response speed" and "reasonable price". Next, we need to calculate the TF value of these keywords in a specific document. For example, in the customer's call record, "service quality" appears 8 times, and the total number of words in the document is 200, so its TF value is 0.04 (i.e. 8 / 200). Then, we need to calculate the IDF value to determine the uniqueness of the word in the entire customer database. If "service quality" appears 20 times in 100 documents, the IDF calculation is log(100 / 20)=0.699. Finally, we can get the initial weight of "service quality" by multiplying TF and IDF, that is, 0.04 * 0.699 = 0.02796. Repeating this process for all extracted keywords will get the initial weight of each keyword. After calculating the weights of these keywords, we can further analyze the importance of keywords in the overall customer interaction data, and these weighted results will become an important basis for the subsequent use of the TextRank algorithm to construct an undirected graph.

[0048] An undirected graph is constructed according to the word co-occurrence relationship of the keywords. Based on the undirected graph, the importance of the words is iteratively calculated using the TextRank algorithm to obtain the final weight of the keywords, wherein each node of the undirected graph represents a keyword, the existence of an edge indicates that the two keywords have appeared in the same context, and each node is assigned an initial weight of the corresponding keyword.

[0049] In this step, we construct an undirected graph by analyzing the co-occurrence relationships between keywords, and apply the TextRank algorithm to further calculate the importance of keywords. The formal structure of the undirected graph makes each keyword a node, and the co-occurrence relationship connects these nodes through edges. This graph construction allows us to observe and understand the relationship between different keywords, and then identify keywords that are more important in a specific context. The iterative calculation method of TextRank can effectively integrate these relationships to evaluate the final weight of each keyword.

[0050] By constructing an undirected graph of keywords and applying the TextRank algorithm, we can not only mine the potential semantic relationships between keywords, but also quantify their importance. This process helps to form a more comprehensive customer demand analysis framework, which greatly improves the accuracy of identifying potential customer needs. The final keyword weights will guide the formulation of call strategies, improve the effectiveness of customer communication, and thus create more business opportunities for enterprises.

[0051] In terms of specific implementation, first, we can use the keywords extracted in the first step to construct an undirected graph. Suppose our keywords include "service quality", "response speed" and "reasonable price". If "service quality" and "response speed" co-occur in the same call record, we will draw an edge connecting these two nodes in the graph. This process is similarly performed for all keywords to construct an undirected graph containing all keywords. Next, the TextRank algorithm is applied to iteratively process the graph. Initially, we assign the same weight (such as 1) to each node (keyword) and then calculate the weight update of each node. Specifically, the weight update of a node depends on the weights of its neighboring nodes and the number of connections to these neighboring nodes. For example, if the neighboring nodes of "service quality" are "response speed" and "reasonable price", then the new weight of "service quality" will be a certain proportion of the sum of the weights of these neighboring nodes. Through multiple rounds of iterations, we can converge to stable weight values ​​that reflect the relative importance of keywords in the entire corpus. The final keyword weights will be used to support subsequent demand analysis and call strategy formulation, thereby improving the effectiveness of customer interaction.

[0052] The keyword features are weighted using the keyword weights to obtain weighted keyword features; In this step, the keyword features are weighted according to the keyword weights calculated in the previous step to generate a weighted keyword feature set. This process ensures that the keyword features can better reflect the real needs and concerns expressed by customers during the interaction process.

[0053] The weighted keyword features can provide companies with more accurate analysis results, helping them to more effectively respond to customer needs in customer service and marketing. At the same time, this method also improves the efficiency of data analysis, allowing companies to quickly respond to market changes.

[0054] In this step, the system uses the previously calculated keyword weights to perform weighted processing on the extracted keyword features. Specifically, the system multiplies each keyword feature by its corresponding weight to obtain a weighted keyword feature. This process emphasizes the importance of words in customer interaction and ensures that customer demand analysis is more targeted and effective. To achieve this process, the system will build a weighted model and use the higher information density contained in the weighted features for data integration. Finally, the calculated weighted keyword features will be stored in the database in the form of structured data for subsequent comprehensive analysis of demand features.

[0055] The historical transaction features in the historical transaction records are extracted, and the customer demand features are determined according to the user portrait features, the weighted keyword features and the historical transaction features.

[0056] In the last step, we extract the customer's historical transaction records, analyze their consumption habits, frequency and other characteristics, and conduct a comprehensive analysis based on the user portrait characteristics and weighted keyword characteristics to accurately identify the customer's demand characteristics. This process ensures the accurate identification and dynamic update of customer needs.

[0057] By establishing accurate customer demand characteristics, enterprises can provide personalized services in their interactions with customers, improve customer satisfaction and loyalty. Such dynamic identification and analysis capabilities enable enterprises to better meet market demand and enhance their competitive advantage.

[0058] In this step, the system will extract the customer's historical transaction records and analyze the historical transaction characteristics, including transaction time, transaction amount, purchase product category, purchase frequency, payment method, etc. Through these characteristics, the system can identify the customer's consumption habits, preferences and their changing trends. For example, the system will analyze the customer's consumption behavior in different time periods to determine the customer's active period and potential demand, thereby providing important data support for demand feature analysis.

[0059] Next, the system comprehensively considers the extracted user portrait features, weighted keyword features, and extracted historical transaction features to determine the customer's demand characteristics. Specifically, the system integrates user portrait features (such as the customer's basic information and interests), weighted keyword features (reflecting the needs and concerns expressed by the customer in the interaction), and historical transaction features (such as purchasing patterns and consumption habits). The system can use machine learning algorithms, such as cluster analysis or decision trees, to comprehensively evaluate these features to identify the customer's potential needs. For example, the system can compare the characteristics of similar users, identify the customer's strong demand for a certain type of product in a specific time period, and generate a detailed customer demand feature report. This report will provide a clear basis for subsequent personalized marketing and services, and support companies to adopt more targeted strategies in customer interactions.

[0060] S203, predicting the customer's interest in a specific product or service based on the customer demand characteristics using a pre-trained customer demand prediction model, and outputting a customer demand probability; In this step, the system inputs the previously generated customer demand features into a pre-trained customer demand prediction model. This model may use a variety of algorithms, such as random forest, XGBoost, or deep learning, and aims to predict the customer's interest in a specific product or service by analyzing the customer's demand characteristics. This process combines multiple factors such as the customer's historical transaction records, call records, and feedback information to generate a set of interest scores to quantify the customer's potential demand for each product or service.

[0061] By leveraging customer demand prediction models, companies can gain in-depth data-driven insights to help identify products or services that customers may be interested in. This prediction not only enables companies to develop more precise marketing and call strategies, but also significantly improves customer satisfaction and conversion rates, thereby improving overall business performance and market competitiveness.

[0062] Specifically, the customer demand features may be input into a pre-trained customer demand prediction model to output a series of interest scores related to specific products or services, wherein the customer demand prediction model includes: a random forest, XGBoost or a deep learning model, and the interest score represents the customer's potential demand intensity for each product or service; In this step, the enterprise organizes customer demand characteristics and inputs them into the pre-trained customer demand prediction model. The model infers the input data characteristics and outputs interest scores related to specific products or services. These scores are based on the customer's historical data and behavior patterns, reflecting the customer's potential demand for different products or services.

[0063] This step is designed to provide an objective, data-driven assessment to help companies understand the latest customer demand dynamics. By generating interest scores, companies can more efficiently allocate resources and strategies in customer-targeted marketing activities, ensuring that customers are provided with the products or services they are most interested in, thereby improving market response speed.

[0064] During the implementation process, the system inputs customer demand characteristics into the selected customer demand prediction model in the form of structured data. The model will process the input characteristics based on the rules learned during its training process and ultimately output a score of each customer's interest in a specific product or service. These scores will provide an important basis for the company's subsequent decision-making and marketing strategies.

[0065] The interest level score output by the model is normalized through the sigmoid function or the softmax function and converted into the corresponding customer demand probability value.

[0066] In this step, the system normalizes the interest scores generated by the model, usually using a sigmoid function or a softmax function. The purpose of this process is to convert the original scores into probability values ​​between 0 and 1, making them easier to interpret and to compare different products or services. This normalization ensures that the expression of customer needs is more intuitive and facilitates subsequent analysis and decision-making.

[0067] Through this normalization process, companies can understand the intensity of customer demand for products or services in the form of probability. This form of expression not only facilitates the marketing team to prioritize and allocate resources, but also helps companies optimize customer relationship management, ensuring that they can provide services that customers really need and are interested in during interactions, thereby improving customer satisfaction.

[0068] During the implementation process, the system will select an appropriate normalization method (such as sigmoid or softmax) and substitute the interest score output by the model into the selected function for calculation. This process will organize and record the generated probability values ​​to form a database of customer demand probabilities to provide data support for subsequent call strategies and marketing activities.

[0069] S204, based on the customer interaction data and the customer demand probability, set a call strategy with the customer to be called, and perform a call operation according to the call strategy, wherein the call strategy includes: call time, call content and call style.

[0070] In this step, the system uses customer interaction data and the probability of customer demand generated by the predictive model to develop call strategies. These strategies cover call time (selecting a time period when customers are more likely to answer the call), call content (introducing products or services with priority based on customer demand probability), and call style (selecting the appropriate communication method based on the customer's user profile). This comprehensive consideration ensures the effectiveness and pertinence of the call process.

[0071] By setting a clear calling strategy, companies can maximize customer answering rates and satisfaction. Accurate calling strategies can not only increase the chances of successful marketing, but also improve customer experience and enhance the trust between customers and companies. The effective implementation of this process directly affects sales conversion rates and customer loyalty, helping companies gain an advantage in a highly competitive market.

[0072] Specifically, based on the customer interaction data and the customer demand probability, a calling strategy for the customer to be called is set, and a time period with the highest answering rate is selected as the calling time according to the call records in the customer interaction data; In this step, the system analyzes the customer's historical call records to identify the time periods when the customer answers the phone most frequently. This analysis can be statistically analyzed based on past call data to help determine the best contact time so that calls can be made during the time period when the customer is most likely to answer the call, thereby increasing the success rate.

[0073] Choosing the right call time is crucial to improving customer answering rates. By calling during the time when customers are most likely to answer, companies can effectively increase the chances of communication and thus increase the likelihood of customer response. This approach helps reduce call failure rates, improve resource utilization efficiency, and ultimately improve customer relationship management effectiveness.

[0074] During the implementation process, the system will collect and organize the customer's call record data, and apply data analysis methods to find out the customer's peak call time. This data will assist the system in selecting the best call time period and be used in the formulation of call strategies to ensure that customers are contacted at critical moments.

[0075] Prioritize each product or service based on the probability of customer demand for each product or service, and introduce the prioritized products or services in order as the content of the call; In this step, the system ranks each product or service according to its probability of customer demand, and determines which products or services have a higher demand in the minds of customers. Based on this priority, the call content will focus on the products or services that customers are most likely to be interested in, making the marketing information more targeted and attractive.

[0076] A prioritization strategy can significantly improve the relevance of calls, ensuring that customers hear the product information they are most interested in. This precision marketing can enhance customer engagement and increase their attention to the content of the call, which may in turn increase conversion rates and customer satisfaction.

[0077] The system will use the customer demand probability value to sort all relevant products or services and record the intensity of demand. When calling, the system will introduce relevant product information to customers in order according to this ranking, ensuring that each call can maximize customer needs and highlight the products that customers are most interested in.

[0078] Determine a user portrait corresponding to the customer basic information in the customer interaction data, and select a chat style corresponding to the user portrait as the call style according to a mapping table between user portraits and chat styles.

[0079] In this step, the system analyzes the customer's basic information to generate a user profile, which may include information such as the customer's age, gender, purchasing habits, preferences, etc. The system then searches the pre-defined chat style mapping table to match the call style that matches the customer profile, thereby customizing a personalized communication method for the call.

[0080] By applying a personalized calling style, companies can establish better customer connections and increase the intimacy and effectiveness of communication. Personalized calling strategies not only improve customer acceptance, but also make customers feel valued and understood, thereby enhancing customer loyalty and satisfaction and further promoting sales.

[0081] During the implementation process, the system first sorts out the information of the customer's basic data to generate a detailed user portrait. Then, it uses this portrait to match the chat style mapping table established in advance to select a call style that best suits the customer to guide the tone, content and method of the call, so that the customer can feel a more personalized experience during the call.

[0082] It can be seen that the customer interaction data of the customer to be called is extracted from the customer database, wherein the customer interaction data includes: customer basic information, historical transaction records, call records and feedback information; the customer interaction data is analyzed to identify customer potential needs and generate customer demand characteristics; according to the customer demand characteristics, a pre-trained customer demand prediction model is used to predict the customer's interest in a specific product or service, and the customer demand probability is output; based on the customer interaction data and the customer demand probability, a call strategy with the customer to be called is set, and a call operation is performed according to the call strategy, wherein the call strategy includes: call time, call content and call style, so that precise calls can be achieved through intelligent means, customer service quality can be improved, and a more scientific and efficient call strategy can be achieved.

[0083] Another embodiment of the present invention provides an intelligent voice calling system. Figure 3 , the system may include: Extraction module 301, used to extract customer interaction data of the customer to be called from the customer database, wherein the customer interaction data includes: customer basic information, historical transaction records, call records and feedback information, wherein the historical transaction records are transaction records for corresponding products or services reached by the customer through historical calls, and the call records are historical call records between the calling end and the customer; An analysis module 302 is used to analyze the customer interaction data, identify potential customer needs, and generate customer demand characteristics; Prediction module 303, used to predict the customer's interest in a specific product or service based on the customer demand characteristics and using a pre-trained customer demand prediction model, and output the customer demand probability; The calling module 304 is used to set a calling strategy with the customer to be called based on the customer interaction data and the customer demand probability, and perform a calling operation according to the calling strategy, wherein the calling strategy includes: call time, call content and call style.

[0084] It can be seen that the customer interaction data of the customer to be called is extracted from the customer database, wherein the customer interaction data includes: customer basic information, historical transaction records, call records and feedback information; the customer interaction data is analyzed to identify customer potential needs and generate customer demand characteristics; according to the customer demand characteristics, a pre-trained customer demand prediction model is used to predict the customer's interest in a specific product or service, and the customer demand probability is output; based on the customer interaction data and the customer demand probability, a call strategy with the customer to be called is set, and a call operation is performed according to the call strategy, wherein the call strategy includes: call time, call content and call style, so that precise calls can be achieved through intelligent means, customer service quality can be improved, and a more scientific and efficient call strategy can be achieved.

[0085] An embodiment of the present invention further provides a storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0086] Specifically, in this embodiment, the above storage medium may be configured to store a computer program for performing the following steps: S201, extracting customer interaction data of the customer to be called from a customer database, wherein the customer interaction data includes: basic information of the customer, historical transaction records, call records and feedback information, wherein the historical transaction records are transaction records for corresponding products or services reached by the customer through historical calls, and the call records are historical call records between the calling end and the customer; S202, analyzing the customer interaction data, identifying potential customer needs, and generating customer demand characteristics; S203, predicting the customer's interest in a specific product or service based on the customer demand characteristics using a pre-trained customer demand prediction model, and outputting a customer demand probability; S204, based on the customer interaction data and the customer demand probability, set a call strategy with the customer to be called, and perform a call operation according to the call strategy, wherein the call strategy includes: call time, call content and call style.

[0087] It can be seen that the customer interaction data of the customer to be called is extracted from the customer database, wherein the customer interaction data includes: customer basic information, historical transaction records, call records and feedback information; the customer interaction data is analyzed to identify customer potential needs and generate customer demand characteristics; according to the customer demand characteristics, a pre-trained customer demand prediction model is used to predict the customer's interest in a specific product or service, and the customer demand probability is output; based on the customer interaction data and the customer demand probability, a call strategy with the customer to be called is set, and a call operation is performed according to the call strategy, wherein the call strategy includes: call time, call content and call style, so that precise calls can be achieved through intelligent means, customer service quality can be improved, and a more scientific and efficient call strategy can be achieved.

[0088] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0089] Specifically, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0090] Specifically, in this embodiment, the processor may be configured to perform the following steps through a computer program: S201, extracting customer interaction data of the customer to be called from a customer database, wherein the customer interaction data includes: basic information of the customer, historical transaction records, call records and feedback information, wherein the historical transaction records are transaction records for corresponding products or services reached by the customer through historical calls, and the call records are historical call records between the calling end and the customer; S202, analyzing the customer interaction data, identifying potential customer needs, and generating customer demand characteristics; S203, predicting the customer's interest in a specific product or service based on the customer demand characteristics using a pre-trained customer demand prediction model, and outputting a customer demand probability; S204, based on the customer interaction data and the customer demand probability, set a call strategy with the customer to be called, and perform a call operation according to the call strategy, wherein the call strategy includes: call time, call content and call style.

[0091] It can be seen that the customer interaction data of the customer to be called is extracted from the customer database, wherein the customer interaction data includes: customer basic information, historical transaction records, call records and feedback information; the customer interaction data is analyzed to identify customer potential needs and generate customer demand characteristics; according to the customer demand characteristics, a pre-trained customer demand prediction model is used to predict the customer's interest in a specific product or service, and the customer demand probability is output; based on the customer interaction data and the customer demand probability, a call strategy with the customer to be called is set, and a call operation is performed according to the call strategy, wherein the call strategy includes: call time, call content and call style, so that precise calls can be achieved through intelligent means, customer service quality can be improved, and a more scientific and efficient call strategy can be achieved.

[0092] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings. The above is only a preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the drawings. Any changes made according to the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which still do not exceed the spirit covered by the description and drawings, should be within the protection scope of the present invention.

Claims

1. An intelligent voice calling method, characterized in that: The method comprises: Extracting customer interaction data of the customer to be called from the customer database, wherein the customer interaction data includes: basic information of the customer, historical transaction records, call records and feedback information, wherein the historical transaction records are transaction records for corresponding products or services reached by the customer through historical calls, and the call records are historical call records between the calling end and the customer; Analyze the customer interaction data, identify potential customer needs, and generate customer demand characteristics; According to the customer demand characteristics, using a pre-trained customer demand prediction model, predict the customer's interest in a specific product or service and output the customer demand probability; Based on the customer interaction data and the customer demand probability, a call strategy for the customer to be called is set, and a call operation is performed according to the call strategy, wherein the call strategy includes: call time, call content and call style.

2. The method according to claim 1, characterized in that The analyzing the customer interaction data, identifying potential customer needs, and generating customer demand characteristics includes: Extract user portrait features from basic customer information to obtain user portrait features; Use natural language processing technology to perform semantic analysis on call records and feedback information, extract keywords, and obtain keyword features; Use word frequency-inverse text frequency to quantify the extracted keywords and obtain keyword weights; The keyword features are weighted using the keyword weights to obtain weighted keyword features; The historical transaction features in the historical transaction records are extracted, and the customer demand features are determined according to the user portrait features, the weighted keyword features and the historical transaction features.

3. The method according to claim 2, characterized in that The method of predicting the customer's interest in a specific product or service based on the customer demand characteristics and outputting the customer demand probability includes: Inputting the customer demand features into a pre-trained customer demand prediction model, and outputting a series of interest scores related to specific products or services, wherein the customer demand prediction model includes: a random forest, XGBoost or a deep learning model, and the interest scores represent the potential demand intensity of the customer for each product or service; The interest level score output by the model is normalized through the sigmoid function or the softmax function and converted into the corresponding customer demand probability value.

4. The method according to claim 3, characterized in that The step of setting a calling strategy for the customer to be called based on the customer interaction data and the customer demand probability includes: According to the call records in the customer interaction data, select a time period with the highest answering rate as the call time; Prioritize each product or service based on the probability of customer demand for each product or service, and introduce the prioritized products or services in order as the content of the call; Determine a user portrait corresponding to the customer basic information in the customer interaction data, and select a chat style corresponding to the user portrait as the call style according to a mapping table between user portraits and chat styles.

5. The method according to claim 2, characterized in that: The method of using natural language processing technology to perform semantic analysis on call records and feedback information, extract keywords, and obtain keyword features includes: Performing text preprocessing on call records and feedback information, wherein the text preprocessing includes: text data cleaning, stem extraction and word form restoration; Perform context-aware keyword extraction on the preprocessed text to obtain candidate keywords; Calculate the frequency of occurrence of each candidate keyword in the call records and feedback information, and assign a sentiment score to the candidate keyword according to its performance in the sentiment dictionary, wherein a sentiment dictionary is constructed based on the corpus, and the sentiment dictionary contains positive sentiment words, negative sentiment words and their sentiment values; Select candidate keywords whose occurrence frequency is higher than a preset frequency threshold and whose sentiment score is higher than a preset score threshold as final keywords; Use K-means or hierarchical clustering algorithm to group the final keywords according to similarity to identify potential topics or categories, and extract representative final keywords from each clustering result as feature words to form the final keyword feature set.

6. The method according to claim 2, characterized in that The method of using word frequency-inverse text frequency to quantify the extracted keywords to obtain keyword weights includes: Use the TF-IDF algorithm to calculate the initial weight of the extracted keywords; An undirected graph is constructed according to the word co-occurrence relationship of the keywords. Based on the undirected graph, the importance of the words is iteratively calculated using the TextRank algorithm to obtain the final weight of the keywords, wherein each node of the undirected graph represents a keyword, the existence of an edge indicates that the two keywords have appeared in the same context, and each node is assigned an initial weight of the corresponding keyword.

7. An intelligent voice calling system, characterized in that: The system comprises: An extraction module is used to extract customer interaction data of the customer to be called from the customer database, wherein the customer interaction data includes: basic information of the customer, historical transaction records, call records and feedback information, wherein the historical transaction records are transaction records for corresponding products or services reached by the customer through historical calls, and the call records are historical call records between the calling end and the customer; An analysis module, used to analyze the customer interaction data, identify potential customer needs, and generate customer demand characteristics; A prediction module is used to predict the customer's interest in a specific product or service based on the customer demand characteristics and use a pre-trained customer demand prediction model to output the customer demand probability; The calling module is used to set a calling strategy with the customer to be called based on the customer interaction data and the customer demand probability, and perform a calling operation according to the calling strategy, wherein the calling strategy includes: call time, call content and call style.

8. The system according to claim 5, characterized in that The analysis module is specifically used for: Extract user portrait features from basic customer information to obtain user portrait features; Use natural language processing technology to perform semantic analysis on call records and feedback information, extract keywords, and obtain keyword features; Use word frequency-inverse text frequency to quantify the extracted keywords and obtain keyword weights; The keyword features are weighted using the keyword weights to obtain weighted keyword features; The historical transaction features in the historical transaction records are extracted, and the customer demand features are determined according to the user portrait features, the weighted keyword features and the historical transaction features.

9. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 6 when executed.

10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 6.

Images