Intelligent identification and early warning method and system for customer demands

Abstract

The present invention discloses a method and system for intelligent identification and early warning of customer demands, which relates to the technical field of customer service, including setting an initial sampling batch to collect customer demand data, a central platform extracting semantic features of customer demand data based on NLP technology; constructing a sentiment analysis model, dynamically adjusting the sampling batch to continuously sample and train the sentiment analysis model to obtain a final sentiment analysis model, collecting customer demand data in real time and inputting it into the final sentiment analysis model to obtain a sentiment analysis result; analyzing the priority of customer demands based on semantic features and sentiment analysis results, sorting and processing according to the priority, and collecting customer feedback data after processing to analyze customer satisfaction; the central platform calculates the proportion of customer demand data based on the customer demand data, and takes corresponding processing measures. The present invention improves the accuracy of customer demand emotion recognition and the accuracy of priority recognition, improves customer satisfaction, and realizes efficient and intelligent customer demand management.

Description

Technical Field

[0001] The present invention relates to the technical field of customer service, and in particular to a method and system for intelligently identifying and warning of customer demands. Background Art

[0002] With the rapid development of digital and intelligent technologies, the interaction between enterprises and customers has gradually evolved towards multi-channel and real-time directions. Customer demand management has become a key link for enterprises to improve customer experience and optimize operational efficiency. In order to effectively manage customer feedback, many companies have begun to introduce automation systems based on natural language processing (NLP) to improve the ability to identify and process customer demands. By collecting data from multiple channels, the core needs of customers are identified based on semantic analysis and sentiment analysis technology. However, the existing technology still has some significant shortcomings in responding to customer demands. When performing sentiment analysis, the existing technology only uses simple existing technology to obtain sentiment analysis results, and the sampling batches cannot be adjusted. The accuracy of sentiment analysis results is crossed when the data fluctuates greatly or the emotions are complex, and it is impossible to analyze the priority of customer demands in combination with semantic features, which affects the efficiency of responding to customer demands. Summary of the invention

[0003] In view of the problems existing in the above-mentioned existing customer demand intelligent identification and early warning methods and systems, the present invention is proposed.

[0004] Therefore, the problem to be solved by the present invention is that the prior art only uses simple prior art to obtain sentiment analysis results when performing sentiment analysis, and is unable to adjust the sampling batches. The accuracy of the sentiment analysis results is cross-correlated when the data fluctuates greatly or the emotions are complex, and it is unable to combine semantic features to analyze the priority of customer demands, which affects the efficiency of responding to customer demands.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: a method for intelligent identification and early warning of customer demands, which includes setting an initial sampling batch to collect customer demand data, and a central platform extracting semantic features of the customer demand data based on NLP technology; constructing a sentiment analysis model, dynamically adjusting the sampling batch to continuously sample and train the sentiment analysis model to obtain a final sentiment analysis model, and collecting customer demand data in real time to input the final sentiment analysis model to obtain a sentiment analysis result; analyzing the priority of customer demands based on semantic features and sentiment analysis results, sorting and processing them according to priority, and collecting customer feedback data after processing to analyze customer satisfaction; the central platform calculates the proportion of customer demand data based on the customer demand data, and takes corresponding processing measures.

[0006] As a preferred solution of the customer demand intelligent identification and early warning method described in the present invention, wherein: the initial sampling batch is set to collect customer demand data, and the central platform extracts the semantic features of the customer demand data based on NLP technology, which means collecting customer demand text data, the central platform cleans and filters the customer demand text data, uses a Gaussian filter to perform denoising, sets the initial sampling batch size to randomly allocate the collected customer demand text data to form an initial sampling batch, and uses NLP technology to extract semantic features in the customer demand text data.

[0007] As a preferred solution of the method for intelligent identification and early warning of customer demands of the present invention, wherein: the construction of the sentiment analysis model, the dynamic adjustment of the sampling batches for continuous sampling and training of the sentiment analysis model to obtain the final sentiment analysis model, and the real-time collection of customer demand data input into the final sentiment analysis model to obtain the sentiment analysis results include:

[0008] The HBRNN model is used as the sentiment analysis model, including a bidirectional recurrent neural network layer, a fully connected layer, and a softmax layer. The bidirectional recurrent neural network layer obtains the hidden state through forward propagation and backward propagation, and is connected to the softmax layer through the fully connected layer to generate the final sentiment analysis results, including positive sentiment state, negative sentiment state, and no sentiment state.

[0009] The central platform divides each customer demand text data into two parts, the front part and the back part, and counts the number of words. It also marks the positive and negative words in the customer demand text data by searching the sentiment polarity dictionary, and adds a sentiment polarity label to each word.

[0010] According to the sentiment vocabulary frequency of the front and back segments of the customer demand text data, local Valence (q) and local Arousal (q) are calculated as local sentiment features:

[0011] Valence(q)=log(x 11 )-log(x 22 );

[0012] Arousal(q)=log(x 11 +x 22 );

[0013] where x 11 The frequency of positive sentiment words in the first part of the customer appeal text data, x 22 The frequency of negative sentiment words in the latter part of the customer appeal text data;

[0014] According to the overall frequency of positive and negative sentiment words in the customer demand text data, the global Arousal (p) and global sentiment tendency g are calculated as global sentiment features:

[0015]

[0016] where x p1 is the overall frequency of positive sentiment words in customer appeal text data, x p2 is the frequency of negative sentiment words in the customer appeal text data as a whole, w i is the sentiment polarity label of the i-th word, and n is the number of words in the customer demand text data;

[0017] Extract the local sentiment features and global sentiment features of all customer demand text data in the initial sampling batch, use a one-dimensional convolutional neural network to process the local sentiment features, and extract the high-dimensional features h of the local sentiment features. q , and apply the multi-head attention mechanism to obtain the attention weight matrix A of the local sentiment feature q , use the gating mechanism to get the final local sentiment feature Q:

[0018]

[0019] Use bidirectional long short-term memory network to process global sentiment features and extract hidden state h p , and introduces a space-based multi-layer attention mechanism to calculate the attention weight of the global sentiment feature, and uses the attention weight to calculate the hidden state h p Perform weighted processing to obtain the final global sentiment feature P;

[0020] The final local emotion feature Q and the final global emotion feature P are concatenated by feature vector concatenation operation to obtain the final emotion feature vector G;

[0021] Collect historical customer demand text data to extract sentiment feature vectors and annotate sentiment analysis results as training sets for model training, define cross entropy loss function and Adam optimizer for iterative optimization of model parameters, calculate model parameter gradients and model loss change rate after each iteration, and calculate the mean and standard deviation of gradient variance, and use the sum of the mean and standard deviation of gradient variance as the gradient threshold, simultaneously calculate the mean and standard deviation of loss change rate variance, and use the sum of the mean and standard deviation of loss change rate variance as the loss change rate threshold, and adjust the sampling batch size based on the gradient threshold and model loss change rate:

[0022]

[0023] Among them B t is the current sampling batch size, B t+1 is the next sampling batch size, Var(b) is the gradient variance, δ b is the gradient threshold, ΔL is the loss change rate, δ L is the loss change rate threshold;

[0024] After the model training is completed, the final sentiment feature vector G is input into the sentiment analysis model to output the sentiment analysis results, and the next customer demand data sampling is performed based on the updated sampling batch size.

[0025] As a preferred solution of the method for intelligent identification and early warning of customer demands described in the present invention, wherein: the analyzing the priority of customer demands based on semantic features and sentiment analysis results, and sorting them according to priority refers to constructing a convolutional neural network model, setting a loss function to train the convolutional neural network model, inputting the semantic features and sentiment analysis results in the customer demand data into the convolutional neural network model to obtain the priority of customer demands, sorting the customer demand text data from high to low according to the priority of the customer demands, and notifying the staff to process the customer demands according to the sorting.

[0026] As a preferred solution of the customer appeal intelligent identification and early warning method described in the present invention, wherein: the collecting of customer feedback data after processing to analyze customer satisfaction refers to collecting customer feedback text data after processing customer appeals and inputting it into a sentiment analysis model to analyze the customer feedback sentiment analysis results; if the customer feedback emotional state is positive or has no emotional state, it is marked as resolved; if the customer feedback emotional state is negative, it is marked as unresolved, and the staff is notified to re-process the customer appeal.

[0027] As a preferred solution of the customer demand intelligent identification and early warning method described in the present invention, wherein: the central platform calculates the proportion of customer demand data based on the customer demand data, and takes corresponding processing measures, which means that the central platform implements the sentiment analysis results of the customer demand data obtained by sentiment analysis model analysis, sets a negative demand threshold, calculates the proportion of customer demand data with negative emotional state in the customer demand data to all customer demand data, and compares it with the negative demand threshold; when the proportion of customer demand data with negative emotional state is greater than the negative demand threshold, a negative demand early warning is triggered, and the staff is notified to analyze the customer demand data, determine the cause of the negative demand and deal with it.

[0028] As a preferred solution of the customer demand intelligent identification and early warning method described in the present invention, after collecting customer demand data, the central platform stores the customer demand data in a database according to time, and stores the sentiment analysis results and priority of the customer demand data in the database to form an association with the customer demand data.

[0029] Another object of the present invention is to provide a customer demand intelligent identification and early warning system, which comprises:

[0030] The data collection module is used to collect and preprocess customer demand data and simultaneously extract the semantic features of customer demand data;

[0031] Sentiment analysis module, used to build a sentiment analysis model and perform sentiment analysis on customer demand data to obtain sentiment analysis results;

[0032] The demand processing module is used to build a convolutional neural network to analyze the priority of customer demands based on the semantic features and sentiment analysis results of customer demand data, sort and process them according to the priority of customer demands, and simultaneously collect customer feedback data to analyze customer satisfaction;

[0033] The monitoring and storage module is used to monitor the proportion of customer demands with negative emotional states in customer demands, process them accordingly, and store customer demand data.

[0034] A computer device comprises: a memory and a processor; the memory stores a computer program, and the processor implements the steps of the above-mentioned customer demand intelligent identification and early warning method when executing the computer program.

[0035] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned customer demand intelligent identification and early warning method.

[0036] The beneficial effects of the present invention are as follows: the present invention collects customer demand data through batch sampling to extract semantic features, and constructs a sentiment analysis model, extracts local sentiment features and global sentiment to analyze the emotional state of customer demands, and synchronously adjusts the sampling batch size, and finally analyzes the priority of customer demands through semantic features and the emotional state of customer demands for sorting and processing, thereby improving the accuracy of customer demand emotion recognition and the accuracy of priority recognition, improving customer satisfaction, and realizing efficient and intelligent customer demand management. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0038] Figure 1 Flowchart of the intelligent identification and early warning method for customer demands.

[0039] Figure 2 Schematic diagram of the process of obtaining the final sentiment features for customer appeal data.

[0040] Figure 3 This is a structural diagram of the intelligent identification and early warning system for customer demands. DETAILED DESCRIPTION

[0041] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0042] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0043] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0044] Example 1

[0045] Reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a method for intelligently identifying and warning customer demands. The method for intelligently identifying and warning customer demands includes:

[0046] S1. Set the initial sampling batch to collect customer demand data, and the central platform extracts the semantic features of customer demand data based on NLP technology;

[0047] Specifically, an initial sampling batch is set to collect customer demand data. The central platform extracts the semantic features of the customer demand data based on NLP technology, which means collecting customer demand text data. The central platform cleans and filters the customer demand text data, uses a Gaussian filter to perform denoising, sets the initial sampling batch size, randomly allocates the collected customer demand text data to form an initial sampling batch, and uses NLP technology to extract semantic features in the customer demand text data.

[0048] By removing irrelevant noise data (such as advertisements, spelling errors, repeated characters, etc.), the quality of data entering the subsequent analysis steps is improved. The use of Gaussian filters for denoising can effectively smooth the data and reduce model errors caused by language inconsistencies or input errors. The denoised text data is purer, and the accuracy and reliability of semantic feature extraction are significantly improved. The setting of the initial sampling batch and the random allocation of data help to ensure the diversity and representativeness of the data in the early stages. Through random allocation, potential biases in the data are avoided, so that the sentiment analysis model can obtain widely covered customer demand information in the early stages of training. This processing method not only reduces the risk of overfitting in model training, but also provides a basic basis for subsequent dynamic sampling. Through NLP technology, the cleaned customer demand text data is deeply analyzed to extract key information representing the core demands of customers. This process not only improves the understanding ability of the sentiment analysis model, but also provides structured data for further training and optimization of the model.

[0049] S2. Build a sentiment analysis model, dynamically adjust the sampling batches to continuously sample and train the sentiment analysis model to obtain the final sentiment analysis model, collect customer demand data in real time and input it into the final sentiment analysis model to obtain the sentiment analysis results;

[0050] Specifically, we build a sentiment analysis model, dynamically adjust the sampling batches to continuously sample and train the sentiment analysis model to obtain the final sentiment analysis model, collect customer demand data in real time and input it into the final sentiment analysis model to obtain sentiment analysis results including:

[0051] The HBRNN model is used as the sentiment analysis model, including a bidirectional recurrent neural network layer, a fully connected layer, and a softmax layer. The bidirectional recurrent neural network layer obtains the hidden state through forward propagation and backward propagation, and is connected to the softmax layer through the fully connected layer to generate the final sentiment analysis results, including positive sentiment state, negative sentiment state, and no sentiment state.

[0052] The central platform divides each customer demand text data into two parts, the front part and the back part, and counts the number of words. It also marks the positive and negative words in the customer demand text data by searching the sentiment polarity dictionary, and adds a sentiment polarity label to each word, where 1 represents positive sentiment, -1 represents negative sentiment, and 0 represents no sentiment polarity.

[0053] According to the sentiment vocabulary frequency of the front and back segments of the customer demand text data, local Valence (q) and local Arousal (q) are calculated as local sentiment features:

[0054] Valence(q)=log(x 11 )-log(x 22 );

[0055] Arousal(q)=log(x 11 +x 22 );

[0056] where x 11 The frequency of positive sentiment words in the first part of the customer appeal text data, x 22 The frequency of negative sentiment words in the latter part of the customer appeal text data;

[0057] According to the overall frequency of positive and negative sentiment words in the customer demand text data, the global Arousal (p) and global sentiment tendency g are calculated as global sentiment features:

[0058]

[0059] where x p1 is the overall frequency of positive sentiment words in customer appeal text data, x p2 is the frequency of negative sentiment words in the customer appeal text data as a whole, w i is the sentiment polarity label of the i-th word, and n is the number of words in the customer demand text data;

[0060] Extract the local sentiment features and global sentiment features of all customer demand text data in the initial sampling batch, use a one-dimensional convolutional neural network to process the local sentiment features, and extract the high-dimensional features h of the local sentiment features. q , and apply the multi-head attention mechanism to obtain the attention weight matrix A of the local sentiment feature q , use the gating mechanism to get the final local sentiment feature Q:

[0061]

[0062] Use bidirectional long short-term memory network to process global sentiment features and extract hidden state h p , and introduces a space-based multi-layer attention mechanism to calculate the attention weight of the global sentiment feature, and uses the attention weight to calculate the hidden state h p Perform weighted processing to obtain the final global sentiment feature P;

[0063] The final local emotion feature Q and the final global emotion feature P are concatenated by feature vector concatenation operation to obtain the final emotion feature vector G;

[0064] Collect historical customer demand text data to extract sentiment feature vectors and annotate sentiment analysis results as training sets for model training, define cross entropy loss function and Adam optimizer for iterative optimization of model parameters, calculate model parameter gradients and model loss change rate after each iteration, and calculate the mean and standard deviation of gradient variance, and use the sum of the mean and standard deviation of gradient variance as the gradient threshold, simultaneously calculate the mean and standard deviation of loss change rate variance, and use the sum of the mean and standard deviation of loss change rate variance as the loss change rate threshold, and adjust the sampling batch size based on the gradient threshold and model loss change rate:

[0065]

[0066] Among them B t is the current sampling batch size, B t+1 is the next sampling batch size, Var(b) is the gradient variance, δ b is the gradient threshold, ΔL is the loss change rate, δ L is the loss change rate threshold;

[0067] After the model training is completed, the final sentiment feature vector G is input into the sentiment analysis model to output the sentiment analysis results, and the next customer demand data sampling is performed based on the updated sampling batch size.

[0068] By dynamically adjusting the sampling batch size, the flexibility and adaptability of the model during the training process are guaranteed. Traditional sentiment analysis models often use fixed-size sampling batches, which is limited when processing large-scale or complex data and easily leads to overfitting or underfitting of the model. The present invention introduces a gradient threshold and a loss change rate threshold, combined with the fluctuations of the gradient variance and the loss change rate, to adjust the sampling batch size in real time to ensure that the model can maintain stable performance under different data volumes. This design not only improves the training efficiency of the model, but also enhances the generalization ability of the model when processing data of different scales. The local sentiment features extract high-dimensional features through a one-dimensional convolutional neural network, and combined with a multi-head attention mechanism and a gating mechanism, can accurately The key emotional parts in the text are identified, and the global emotional features are extracted through a bidirectional long short-term memory network (Bi-LSTM) and a space-based multi-layer attention mechanism to capture the emotional trend of the entire text. By splicing local features and global features, the present invention realizes a comprehensive analysis of customer demand emotions and can effectively identify subtle changes in emotions. The combination of local and global emotional features greatly improves the accuracy and reliability of emotional analysis. The bidirectional recursive neural network layer can simultaneously capture forward and backward contextual information from the input customer demand data, so that the model can fully consider the correlation between the front and back when analyzing emotions, thereby improving the accuracy of emotion recognition. The fully connected layer and the softmax layer are used to combine the Bi-RNN The hidden state is converted into the final sentiment classification output. The model can accurately judge whether the customer's emotional state is positive, negative or emotionless. This process ensures that the sentiment analysis is meticulous and comprehensive, especially when dealing with complex sentiment data. In the training process of the model, the cross entropy loss function is used to measure the gap between the model prediction and the true label. The cross entropy loss function can effectively handle multi-classification problems and performs well in the sentiment analysis task of the present invention. Combined with the Adam optimizer for gradient update and parameter adjustment, the model can converge quickly in each iteration and continuously optimize the accuracy of sentiment analysis. By splicing the local sentiment feature Q with the global sentiment feature P, the present invention can generate a more comprehensive sentiment The feature vector G not only contains the local details of the customer demand text, but also reflects the overall emotional tendency. The generation of this feature vector greatly improves the predictive ability of the sentiment analysis model and provides more comprehensive input data for the final sentiment classification. The splicing operation enables the model to take into account both local and global sentiment features to ensure more accurate analysis results. The sampling batch size is dynamically adjusted according to the changes in the data and the learning progress of the model. Through this mechanism, the model can adapt to the fluctuations of the data stream more flexibly, avoiding the problems of low training efficiency and poor model adaptability caused by fixed sampling. As the model is gradually optimized, it can respond quickly to changes in data-intensive environments, thereby ensuring the stability and efficiency of sentiment analysis.

[0069] S3. Analyze the priority of customer demands based on semantic features and sentiment analysis results, sort them according to the priority, and collect customer feedback data after processing to analyze customer satisfaction;

[0070] Specifically, analyzing the priority of customer demands based on semantic features and sentiment analysis results, and sorting them according to priority means constructing a convolutional neural network model, setting a loss function to train the convolutional neural network model, inputting the semantic features and sentiment analysis results in the customer demand data into the convolutional neural network model to obtain the priority of customer demands, sorting the customer demand text data from high to low according to the priority of the customer demands, and notifying the staff to process the customer demands according to the ranking.

[0071] Convolutional neural networks can extract distinguishing priority features from complex semantic features and sentiment analysis results. In customer demand management, it is crucial to accurately judge the urgency and importance of demands. By introducing the CNN model, the present invention can extract higher-dimensional semantic and sentiment features from customer demand data, providing strong support for accurate prediction of priorities. This method can automatically process a large amount of customer demand data, reducing the need for human intervention and greatly improving the efficiency of customer demand processing. The loss function guides the model to continuously optimize weight parameters during training by calculating the gap between the priority output by the model and the true label. The advantage of using the cross entropy loss function is that it can effectively handle multi-classification problems, especially in the scenario of customer demand priority classification, it can ensure that each customer demand is given an appropriate priority assessment. In addition, by continuously updating the model parameters, the prediction accuracy of the model has been significantly improved in iterative training. Combining the semantic features and sentiment analysis results in customer demands, multi-modal fusion is input into the CNN model. This process greatly enhances the model's accuracy in customer demand. The ability to identify the priority of customer demands. Semantic features reflect the main content and theme of customer demands, while sentiment analysis results reflect the customer's emotions and attitudes. By fusing information from these two different modalities, the model can more comprehensively understand the true intention and urgency of customer demands, automatically sort customer demands, and feed back these sorting results to the company's customer service personnel. Priority sorting can not only help customer service personnel quickly locate the most urgent and important demands, but also reduce the situation where low-priority demands occupy too many resources, thereby improving the efficiency of the entire demand processing process. This process greatly reduces the time for manual judgment and improves the overall intelligence level and response speed of customer management. Staff can prioritize urgent customer demands according to the sorting results to ensure that customers' urgent needs can be responded to in a timely manner. This notification mechanism combines intelligent sorting and manual intervention to form an efficient customer service model. In addition, after the processing is completed, the system will automatically collect the processing results and continuously optimize the priority prediction effect of the CNN model by further analyzing customer feedback data, forming a virtuous circle.

[0072] Furthermore, collecting customer feedback data after processing and analyzing customer satisfaction means collecting customer feedback text data after processing customer demands and inputting the data into the sentiment analysis model to analyze the customer feedback sentiment analysis results. If the customer feedback emotional state is positive or has no emotional state, it is marked as solved; if the customer feedback emotional state is negative, it is marked as unsolved, and the staff is notified to re-process the customer demand.

[0073] After the customer's demands are processed, the enterprise will collect text data of customer feedback, which can reflect the customer's satisfaction or dissatisfaction with the processing results. By introducing the sentiment analysis model, the present invention can automatically identify the emotional state in the customer feedback and mark the processing status of the customer's demands according to the sentiment classification results. The sentiment analysis model can effectively distinguish the customer's positive, negative or neutral emotions, and use this as a basis to judge whether the customer is satisfied with the processing results. This process greatly improves the efficiency of customer feedback data processing and reduces the time and cost of manual analysis. By collecting customer feedback data and using the sentiment analysis model to monitor the customer's emotional state, it can continuously track and analyze the customer's feedback on the processing results, thereby providing the enterprise with a more accurate customer satisfaction evaluation. This automated monitoring mechanism can not only discover problems in the service, but also provide the enterprise with feedback basis for improving service quality. By continuously optimizing the processing process and improving customer service, the present invention will greatly improve customer satisfaction, enabling the enterprise to manage customer relationships more efficiently. By introducing the sentiment analysis model and the automatic marking system, the full automation of customer feedback processing is realized. This intelligent processing method can not only reduce the manpower investment of the enterprise in customer management, but also ensure that all customer demands are properly handled, thereby improving the overall operational efficiency and customer experience of the enterprise.

[0074] S4. The central platform calculates the proportion of customer demand data based on customer demand data and takes corresponding measures;

[0075] Specifically, the central platform calculates the proportion of customer demand data based on the customer demand data, and takes corresponding processing measures. The central platform implements the sentiment analysis results of the customer demand data through the sentiment analysis model, sets the negative demand threshold, calculates the proportion of customer demand data with negative emotional state in the customer demand data to all customer demand data, and compares it with the negative demand threshold. When the proportion of customer demand data with negative emotional state is greater than the negative demand threshold, a negative demand warning is triggered, and the staff is notified to analyze the customer demand data, determine the causes of the negative demands and deal with them.

[0076] The sentiment analysis model can efficiently process large amounts of text data and automatically determine the emotional state of each customer demand. This process can not only clearly classify the customer's emotional state, but also provide basic support for subsequent negative sentiment data analysis. By classifying the emotions of customer demands, companies can more intuitively understand the emotional trends of customers and provide data support for optimizing customer service strategies. Based on the sentiment analysis results, the central platform will calculate the proportion of customer demand data with negative emotional states to all customer demand data. This proportion reflects the severity of potential problems in the current customer service process. When the proportion of negative emotions is lower than the preset threshold, it means that the customer's overall emotional state is relatively stable; however, when the proportion exceeds the threshold, it indicates that there may be systemic problems or specific products and services that have caused more negative emotions. After the negative demand warning is triggered, the system will notify the staff to further analyze the customer demand data with negative emotional states. The staff can view these data to gain an in-depth understanding of the causes of customer dissatisfaction, such as product problems, service defects or communication barriers. By analyzing these negative demand data, companies can formulate targeted solutions to improve customer experience and increase customer satisfaction.

[0077] Furthermore, after collecting the customer demand data, the central platform stores the customer demand data in a database according to time, and stores the sentiment analysis results and priorities of the customer demand data in the database to form an association with the customer demand data.

[0078] By associating and storing customer demand data, sentiment analysis results, and priority information, the database not only stores the content of customer demands, but also includes sentiment analysis results and priority information for these demands. This multi-dimensional data storage method can significantly improve data utilization efficiency, allowing companies to more quickly find customer demand data related to specific emotional states or priorities. For customer service personnel, this mechanism helps to more comprehensively understand the customer's emotional background and the urgency of their demands when handling customer demands, so as to make more appropriate decisions. Storing sentiment analysis results in association with customer demand data can provide companies with more accurate emotional insights. For example, when the sentiment analysis results in customer demand data are negative, the system can prioritize marking these data and push them to relevant departments for priority processing. By storing sentiment analysis results and associating them with demand data, companies can more quickly identify and respond to customer dissatisfaction and avoid the accumulation and expansion of problems.

[0079] Example 2

[0080] Reference Figure 3 , which is the second embodiment of the present invention, is different from the previous embodiment and provides a customer demand intelligent identification and early warning system, which includes:

[0081] The data collection module is used to collect and preprocess customer demand data and simultaneously extract the semantic features of customer demand data;

[0082] Sentiment analysis module, used to build a sentiment analysis model and perform sentiment analysis on customer demand data to obtain sentiment analysis results;

[0083] The demand processing module is used to build a convolutional neural network to analyze the priority of customer demands based on the semantic features and sentiment analysis results of customer demand data, sort and process them according to the priority of customer demands, and simultaneously collect customer feedback data to analyze customer satisfaction;

[0084] The monitoring and storage module is used to monitor the proportion of customer demands with negative emotional states in customer demands, process them accordingly, and store customer demand data.

[0085] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0086] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0087] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0088] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0089] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for intelligently identifying and warning customer demands, characterized in that: include, Set the initial sampling batch to collect customer demand data, and the central platform extracts the semantic features of customer demand data based on NLP technology; Build a sentiment analysis model, dynamically adjust the sampling batches to continuously sample and train the sentiment analysis model to obtain the final sentiment analysis model, collect customer demand data in real time and input it into the final sentiment analysis model to obtain the sentiment analysis results; Analyze the priority of customer demands based on semantic features and sentiment analysis results, sort them according to priority, and collect customer feedback data after processing to analyze customer satisfaction; The central platform calculates the proportion of customer demand data based on customer demand data and takes corresponding measures; The setting of the initial sampling batch to collect customer demand data, the central platform extracting the semantic features of the customer demand data based on the NLP technology means collecting customer demand text data, the central platform cleaning and filtering the customer demand text data, using a Gaussian filter to perform denoising, setting the initial sampling batch size to randomly allocate the collected customer demand text data to form an initial sampling batch, and using the NLP technology to extract semantic features in the customer demand text data; The sentiment analysis model is constructed, the sampling batch is dynamically adjusted to continuously sample and train the sentiment analysis model to obtain the final sentiment analysis model, and the customer demand data is collected in real time and input into the final sentiment analysis model to obtain the sentiment analysis result, including: The HBRNN model is used as the sentiment analysis model, including a bidirectional recurrent neural network layer, a fully connected layer, and a softmax layer. The bidirectional recurrent neural network layer obtains the hidden state through forward propagation and backward propagation, and is connected to the softmax layer through the fully connected layer to generate the final sentiment analysis results, including positive sentiment state, negative sentiment state, and no sentiment state. The central platform divides each customer demand text data into two parts, the front part and the back part, and counts the number of words. It also marks the positive and negative words in the customer demand text data by searching the sentiment polarity dictionary, and adds a sentiment polarity label to each word. According to the sentiment word frequencies of the front and back segments of the customer demand text data, local Valence (q) and local Arousal (q) are calculated as local sentiment features: Valence(q)=log(x 11 )-log(x 22 ); Arousal(q)=log(x 11 +x 22 ); where x 11 The frequency of positive sentiment words in the first part of the customer appeal text data, x 22 The frequency of negative sentiment words in the latter part of the customer appeal text data; According to the overall frequency of positive and negative sentiment words in the customer demand text data, the global Arousal (p) and global sentiment tendency g are calculated as global sentiment features: where x p1 is the overall frequency of positive sentiment words in customer appeal text data, x p2 is the frequency of negative sentiment words in the customer appeal text data as a whole, w i is the sentiment polarity label of the i-th word, and n is the number of words in the customer demand text data; Extract the local sentiment features and global sentiment features of all customer demand text data in the initial sampling batch, use a one-dimensional convolutional neural network to process the local sentiment features, and extract the high-dimensional features h of the local sentiment features. q , and apply the multi-head attention mechanism to obtain the attention weight matrix A of the local sentiment feature q , use the gating mechanism to get the final local sentiment feature Q: Use bidirectional long short-term memory network to process global sentiment features and extract hidden state h p , and introduces a space-based multi-layer attention mechanism to calculate the attention weight of the global sentiment feature, and uses the attention weight to calculate the hidden state h p Perform weighted processing to obtain the final global sentiment feature P; The final local emotion feature Q and the final global emotion feature P are concatenated by feature vector concatenation operation to obtain the final emotion feature vector G; Collect historical customer demand text data to extract sentiment feature vectors and annotate sentiment analysis results as training sets for model training, define cross entropy loss function and Adam optimizer for iterative optimization of model parameters, calculate model parameter gradients and model loss change rate after each iteration, and calculate the mean and standard deviation of gradient variance, and use the sum of the mean and standard deviation of gradient variance as the gradient threshold, simultaneously calculate the mean and standard deviation of loss change rate variance, and use the sum of the mean and standard deviation of loss change rate variance as the loss change rate threshold, and adjust the sampling batch size based on the gradient threshold and model loss change rate: Among them B t is the current sampling batch size, B t+1 is the next sampling batch size, Var(b) is the gradient variance, δ b is the gradient threshold, ΔL is the loss change rate, δ L is the loss change rate threshold; After the model training is completed, the final sentiment feature vector G is input into the sentiment analysis model to output the sentiment analysis results, and the next customer demand data sampling is performed based on the updated sampling batch size.

2. The method for intelligent identification and early warning of customer demands according to claim 1, characterized in that: The analyzing the priority of customer demands based on semantic features and sentiment analysis results and sorting them according to priority refers to constructing a convolutional neural network model, setting a loss function to train the convolutional neural network model, inputting the semantic features and sentiment analysis results in the customer demand data into the convolutional neural network model to obtain the priority of customer demands, sorting the customer demand text data from high to low according to the priority of the customer demands, and notifying the staff to process the customer demands according to the ranking.

3. The method for intelligent identification and early warning of customer demands according to claim 2, characterized in that: The collecting of customer feedback data after processing and analyzing customer satisfaction refers to collecting customer feedback text data after processing customer demands and inputting it into a sentiment analysis model to analyze the customer feedback sentiment analysis results; if the customer feedback sentiment state is positive or has no sentiment state, it is marked as resolved; if the customer feedback sentiment state is negative, it is marked as unresolved, and the staff is notified to re-process the customer demand.

4. The method for intelligent identification and early warning of customer demands according to claim 3, characterized in that: The central platform calculates the proportion of customer demand data based on the customer demand data and takes corresponding processing measures, which means that the central platform implements the sentiment analysis results of the customer demand data obtained by sentiment analysis model analysis, sets a negative demand threshold, calculates the proportion of customer demand data with negative sentiment status in the customer demand data to all customer demand data, and compares it with the negative demand threshold. When the proportion of customer demand data with negative sentiment status is greater than the negative demand threshold, a negative demand warning is triggered, and the staff is notified to analyze the customer demand data, determine the cause of the negative demand and handle it.

5. The method for intelligent identification and early warning of customer demands according to claim 4, characterized in that: After collecting the customer demand data, the central platform stores the customer demand data in a database according to time, and stores the sentiment analysis results and priorities of the customer demand data in the database to form an association with the customer demand data.

6. A customer demand intelligent identification and early warning system according to any one of claims 1 to 5, characterized in that: include, The data collection module is used to collect and preprocess customer demand data and simultaneously extract the semantic features of customer demand data; Sentiment analysis module, used to build a sentiment analysis model and perform sentiment analysis on customer demand data to obtain sentiment analysis results; The demand processing module is used to build a convolutional neural network to analyze the priority of customer demands based on the semantic features and sentiment analysis results of customer demand data, sort and process them according to the priority of customer demands, and simultaneously collect customer feedback data to analyze customer satisfaction; The monitoring and storage module is used to monitor the proportion of customer demands with negative emotional states in customer demands, process them accordingly, and store customer demand data.

7. A computer device comprising: Memory and processor; The memory stores a computer program, characterized in that when the processor executes the computer program, the steps of the customer demand intelligent identification and early warning method described in any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the customer demand intelligent identification and early warning method described in any one of claims 1 to 5 are implemented.

Images