Client information resource integration analysis method and system based on big data

Abstract

The invention discloses a customer information resource integration analysis method and system based on big data, and relates to the technical field of customer information services, and the method comprises the steps: collecting customer multi-modal data, carrying out the preprocessing, extracting data features, and carrying out the customer emotion analysis according to the data features; clustering and layering the clients according to the client sentiment analysis result to generate client tags, constructing a knowledge graph based on the client tags, and performing personalized recommendation based on the knowledge graph; monitoring personalized recommendation feedback of the client in real time, and optimizing the knowledge graph based on the feedback; and generating a customer recommendation strategy based on the knowledge graph, and comprehensively forming a customer decision support report. According to the method, the CNN model and quantum analysis are combined for emotion analysis, so that the efficiency and accuracy of emotion analysis are greatly improved, clustering analysis and knowledge graph construction are synchronously performed based on emotion analysis and customer data for personalized recommendation, the accuracy and correlation of customer recommendation are improved, and the user experience is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of customer information services, and in particular to a method and system for integrating and analyzing customer information resources based on big data. Background Art

[0002] With the rapid development of information technology and big data, the amount of customer information that enterprises can obtain and process in customer relationship management (CRM) is growing exponentially. This information includes various data types such as customer purchase history, browsing history, social media interaction, etc., forming a big data resource with customers as the core. In recent years, the collection and processing technology of multimodal data has gradually matured, allowing enterprises to comprehensively analyze customer behavior and emotional characteristics by comprehensively utilizing multiple data sources such as text, voice, and video. As an emerging technology, knowledge graphs can more deeply understand customer needs and explore potential business opportunities by establishing a relationship network between entities. Applying knowledge graphs to the integration and analysis of customer information resources can effectively improve the organization and retrieval efficiency of information. However, there are still deficiencies in existing technologies. Existing technologies often only use deep learning algorithms when analyzing customer emotional characteristics, and cannot fully utilize the complementarity of multiple data sources, resulting in insufficient accuracy of sentiment analysis results. In addition, customer stratification methods mainly rely on static clustering algorithms, which are difficult to dynamically adjust according to real-time changes in customer behavior and emotional state. Summary of the invention

[0003] In view of the problems existing in the above-mentioned existing customer information resource integration and analysis methods and systems based on big data, the present invention is proposed.

[0004] Therefore, the problem to be solved by the present invention is that the existing technology often only adopts deep learning algorithms when analyzing customer emotional characteristics, and cannot fully utilize the complementarity of multiple data sources, resulting in insufficient accuracy of sentiment analysis results, and the customer stratification method mainly relies on static clustering algorithms, which are difficult to dynamically adjust according to real-time changes in customer behavior and emotional state.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: a customer information resource integration and analysis method based on big data, which includes collecting customer multimodal data for preprocessing to extract data features and performing customer sentiment analysis based on the data features; clustering and stratifying customers according to the results of customer sentiment analysis to generate customer tags, and constructing a knowledge graph based on the customer tags, and performing personalized recommendations based on the knowledge graph; real-time monitoring of customer feedback on personalized recommendations, and optimizing the knowledge graph based on the feedback; generating customer recommendation strategies based on the knowledge graph, and comprehensively forming a customer decision support report.

[0006] As a preferred solution of the customer information resource integration and analysis method based on big data described in the present invention, wherein: the collecting of customer multimodal data for preprocessing and extracting data features refers to collecting customer text data, voice data, video data and customer basic information, and preprocessing them respectively;

[0007] Use the BERT model to convert text data into a text feature vector T;

[0008] Use the openSMILE model to extract the MFCC features of speech data, and convert the MFCC features into speech feature vectors S through the LSTM model;

[0009] ResNes-50 is used to extract the spatial features of video data, and the video feature vector V is obtained by combining LSTM to process the spatial features.

[0010] As a preferred solution of the customer information resource integration and analysis method based on big data described in the present invention, wherein: the customer sentiment analysis based on data features includes:

[0011] Based on the text feature vector T, the speech feature vector S and the video feature vector V, the autocovariance matrix C is constructed respectively. T , C S and C V , and construct the joint covariance matrix C of the three TSV ;

[0012] According to the text feature vector T, the speech feature vector S and the video feature vector V, the text feature projection vector a, the speech feature projection vector b and the video feature projection vector c are obtained respectively. The CCA model is used to maximize the correlation of the feature vectors to obtain the maximum correlation projection a of the text feature, speech feature and video feature in the high-dimensional space. T 、a S and a V ;

[0013] A T 、a S and a V By expressing the quantum state, we can get |T CCA >、|S CCA > and |V CCA >, define T CCA >、|S CCA > and |V CCA The weight coefficients of > are α, β and γ respectively, and the comprehensive quantum state representation is obtained through the weight coefficients

[0014]

[0015] From the historical customer multimodal data, we obtain the |T representing positive emotion, neutral emotion and negative emotion respectively. CCA >1、|S CCA >1 and |V CCA >1, and substitute them into the comprehensive quantum state representation for weight coefficient training, use the cross entropy loss function and gradient descent method to adjust the weight coefficient, and obtain the weight coefficients of the comprehensive quantum state representation of positive emotion, neutral emotion and negative emotion respectively, and apply the weight coefficients of the comprehensive quantum state representation of positive emotion, neutral emotion and negative emotion to |T CCA >、|S CCA > and |V CCA > Obtain comprehensive quantum state representation of positive emotions, neutral emotions, and negative emotions

[0016] The maximum correlation projection a solved using the CCA model T 、a S and a V Fusion of text feature vector T, speech feature vector S and video feature vector V:

[0017]

[0018] Where F is the fusion feature vector;

[0019] Construct a CNN model and use the training set for training. Input the fused feature vector F into the CNN model to obtain the probability p of positive emotion, neutral emotion, and negative emotion of the fused feature vector F. i ;

[0020] According to the comprehensive quantum state representation of positive emotion, neutral emotion and negative emotion and the probability p of positive emotion, neutral emotion and negative emotion of the fused feature vector F i Construct the density matrix ρ:

[0021]

[0022] in for The accompanying state of

[0023] According to the comprehensive quantum state representation of positive emotion, neutral emotion and negative emotion Initialize the corresponding projection operator Π i , using the gradient descent algorithm to transform the projection operator Π i Perform training and calculate the objective function G:

[0024]

[0025] Among them, tr represents trace operation;

[0026] Calculate the gradient ▽G according to the objective function and calculate the gradient direction D k , according to the gradient direction D k Iteratively update the density matrix ρ:

[0027] ρ k+1 =ρ k +tk×D k ;

[0028] where ρ k is the density matrix updated at the kth iteration, ρ k+1 is the density matrix updated in the k+1th iteration, and tk is the gradient step size;

[0029] Calculate the probability p of positive emotion, neutral emotion and negative emotion for the density matrix after each iteration k+1 (i)

[0030] p k+1 (i) = tr(Π i ρ k+1 Π i );

[0031] Calculate p k+1 (i) and p i The cross entropy loss is calculated, and the iterative update of the density matrix is ​​stopped when the loss converges. The p calculated by the updated density matrix k+1 (i) is taken as the final probability, and the highest sentiment probability is selected as the sentiment analysis result.

[0032] As a preferred solution of the customer information resource integration and analysis method based on big data described in the present invention, wherein: the generation of customer labels by clustering and stratifying customers according to the customer sentiment analysis results refers to setting the initial number of cluster centers to K, calculating the Euclidean distance between each customer fusion feature vector F and each cluster center, assigning the customer fusion feature vector F to the nearest cluster center and updating the cluster center position;

[0033] Calculate the average distance x between the i-th customer and other customers in the same cluster center and the average distance y between the i-th customer and the customer in the nearest cluster center. Calculate the silhouette coefficient S of the i-th customer based on the average distance x and the average distance y, and then get the average silhouette coefficient of all customers in all cluster centers.

[0034] Set the iteration step to 1, iteratively increase the K value and repeatedly calculate the average silhouette coefficient corresponding to each K value until the maximum value is reached, and then stop the iteration. The K value with the largest average silhouette coefficient will be used as the latest number of cluster centers.

[0035] Repeat the customer fusion feature vector allocation and cluster center update until the cluster center position no longer changes significantly, and then stop updating;

[0036] Generate customer tags based on cluster centers and customer sentiment analysis results.

[0037] As a preferred solution of the customer information resource integration and analysis method based on big data described in the present invention, wherein: the construction of the knowledge graph based on customer tags and the personalized recommendation based on the knowledge graph refer to the construction of the knowledge graph based on the customer tags after obtaining the customer tags, taking the basic information of each customer, the customer tags and the fusion feature vector as nodes, and taking the Euclidean distance of the customer in the cluster as edges to construct the knowledge graph;

[0038] The nodes are updated through the graph convolutional neural network, the Pearson correlation coefficient between customers is calculated, the correlation coefficient threshold is set, customers with a correlation coefficient greater than the correlation coefficient threshold are marked as similar customers, and customer purchase and browsing information is obtained and recommended to similar customers.

[0039] As a preferred solution of the customer information resource integration and analysis method based on big data described in the present invention, wherein: the real-time monitoring of customer feedback on personalized recommendations and optimizing the knowledge graph based on feedback refers to collecting customer feedback data on personalized recommendations in real time after personalized recommendations are made, and obtaining sentiment analysis results of customer feedback data. If the customer feedback data is positive, positive feedback is generated to record the customer and the personalized recommendation content. If it is neutral or negative, negative feedback is generated, and the knowledge graph nodes are optimized again through the graph convolutional neural network.

[0040] As a preferred solution of the customer information resource integration and analysis method based on big data described in the present invention, wherein: the generating of customer recommendation strategy based on knowledge graph and comprehensively forming customer decision support report refers to comprehensively generating customer recommendation strategy based on purchase information and browsing information of all similar customers of the customer in the knowledge graph, adjusting the customer recommendation strategy based on customer feedback data to form a customer decision support report, and storing the customer decision support report.

[0041] Another object of the present invention is to provide a customer information resource integration and analysis system based on big data, which includes: a data collection module for collecting customer multimodal data and performing preprocessing to extract data features;

[0042] Sentiment analysis module, used to analyze the data features of customer multimodal data to obtain sentiment analysis results;

[0043] Personalized recommendation module, which is used to perform cluster analysis based on sentiment analysis results and customer multimodal data, and build a knowledge graph for personalized recommendations;

[0044] The feedback optimization module is used to collect and analyze customer feedback data after personalized recommendations, optimize the knowledge graph, and simultaneously record personalized recommendation content.

[0045] 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 information resource integration and analysis method based on big data when executing the computer program.

[0046] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned customer information resource integration and analysis method based on big data.

[0047] The beneficial effects of the present invention are as follows: the present invention collects customer data and extracts features, and uses a CNN model combined with quantum analysis to perform sentiment analysis, thereby greatly improving the efficiency and accuracy of sentiment analysis. It simultaneously performs cluster analysis and builds a knowledge graph based on sentiment analysis and customer data for personalized recommendations, thereby improving the accuracy and relevance of customer recommendations and significantly improving user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] 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.

[0049] Figure 1 The figure is a flowchart of the customer information resource integration and analysis method based on big data.

[0050] Figure 2 Schematic diagram of the construction process of the quantum state density matrix for sentiment analysis.

[0051] Figure 3 This is a structural diagram of the customer information resource integration and analysis system based on big data. DETAILED DESCRIPTION

[0052] 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.

[0053] 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.

[0054] 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.

[0055] Example 1

[0056] Reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a customer information resource integration and analysis method based on big data. The customer information resource integration and analysis method based on big data includes:

[0057] S1, collect customer multimodal data, pre-process and extract data features, and perform customer sentiment analysis based on the data features;

[0058] Specifically, collecting multimodal data of customers and preprocessing and extracting data features means collecting text data, voice data, video data and basic information of customers, and preprocessing them respectively;

[0059] Use the BERT model to convert text data into a text feature vector T;

[0060] Use the openSMILE model to extract the MFCC features of speech data, and convert the MFCC features into speech feature vectors S through the LSTM model;

[0061] ResNes-50 is used to extract the spatial features of video data, and the video feature vector V is obtained by combining LSTM to process the spatial features.

[0062] By preprocessing and extracting features from multimodal data (text, voice, video), the effectiveness and availability of data are significantly improved. By processing text, voice, and video data separately, the system can extract the information features that best represent customer behavior and emotions in each data type. This method ensures the complementarity between data of different modalities and provides a more comprehensive and accurate data foundation for subsequent sentiment analysis and customer behavior prediction. The BERT model uses a bidirectional Transformer architecture to perform contextual understanding of text data. The generated text feature vector T can capture the deep semantics and emotional information in the text. Compared with traditional text processing methods, the BERT model significantly improves the accuracy of sentiment analysis, especially when processing complex natural languages. The feature vectors generated by BERT can more accurately predict customers' emotional tendencies and intentions, providing a more reliable foundation for personalized recommendations. enSMILE extracts MFCC features from speech and uses the LSTM model to convert these features into speech feature vectors S. The system can effectively capture emotional changes in speech signals. This process not only improves the processing accuracy of speech data, but also enables the system to dynamically analyze the customer's emotional state and potential information in the speech. Combined with the LSTM model of time series, the dynamic characteristics of speech are analyzed more carefully, enhancing the effect of emotion recognition. The ResNet-50 model extracts spatial features from video frames and combines the LSTM model with the processing of time series. The generated video feature vector V can fully capture the emotional information and behavioral clues in the video. Compared with traditional image processing methods, ResNet-50 has obvious advantages in extracting details and emotional patterns from complex video data. Especially when processing long-time series video content, the LSTM model can effectively capture emotional changes over time.

[0063] Further, customer sentiment analysis based on data features includes:

[0064] Based on the text feature vector T, the speech feature vector S and the video feature vector V, the autocovariance matrix C is constructed respectively. T , C S and C V , and construct the joint covariance matrix C of the three TSV ,in:

[0065] C T =Cov(T,T),C S = Cov(S,S), C V = Cov(V,V);

[0066] C TSV = Cov(T,S,V);

[0067] According to the text feature vector T, the speech feature vector S and the video feature vector V, the text feature projection vector a, the speech feature projection vector b and the video feature projection vector c are obtained respectively. The CCA model is used to maximize the correlation of the feature vectors to obtain the maximum correlation projection a of the text feature, speech feature and video feature in the high-dimensional space. T 、a S and a V ,in:

[0068]

[0069] where a z is the transpose of the text feature projection vector a, b z is the transpose of the speech feature projection vector b, c z is the transpose of the video feature projection vector c;

[0070] A T 、a S and a V By expressing the quantum state, we can get |T CCA >、|S CCA > and |V CCA >, define T CCA >、|S CCA > and |V CCA The weight coefficients of > are α, β and γ respectively, and the comprehensive quantum state representation is obtained through the weight coefficients

[0071]

[0072] From the historical customer multimodal data, we obtain the |T representing positive emotion, neutral emotion and negative emotion respectively. CCA >1、|S CCA >1 and |V CCA >1, and substitute them into the comprehensive quantum state representation for weight coefficient training, use the cross entropy loss function and gradient descent method to adjust the weight coefficient, and obtain the weight coefficients of the comprehensive quantum state representation of positive emotion, neutral emotion and negative emotion respectively, and apply the weight coefficients of the comprehensive quantum state representation of positive emotion, neutral emotion and negative emotion to |T CCA >、|S CCA > and |V CCA > Obtain comprehensive quantum state representation of positive emotions, neutral emotions, and negative emotions

[0073] The maximum correlation projection a solved using the CCA model T 、a S and a VFusion of text feature vector T, speech feature vector S and video feature vector V:

[0074]

[0075] Where F is the fusion feature vector;

[0076] Construct a CNN model and use the training set for training. Input the fused feature vector F into the CNN model to obtain the probability p of positive emotion, neutral emotion, and negative emotion of the fused feature vector F. i ;

[0077] According to the comprehensive quantum state representation of positive emotion, neutral emotion and negative emotion and the probability p of positive emotion, neutral emotion and negative emotion of the fused feature vector F i Construct the density matrix ρ:

[0078]

[0079] in for The accompanying state of

[0080] According to the comprehensive quantum state representation of positive emotion, neutral emotion and negative emotion Initialize the corresponding projection operator Π i , using the gradient descent algorithm to transform the projection operator Π i Perform training and calculate the objective function G:

[0081]

[0082] Among them, tr represents trace operation;

[0083] Calculate the gradient ▽G according to the objective function and calculate the gradient direction D k :

[0084]

[0085] Where q(tk) is the step size function, tk is the gradient step size, and ρ k is the density matrix updated at the kth iteration, and is the direction of gradient ascent, ▽G k The objective function gradient of the density matrix updated for the kth iteration;

[0086] in and They are as follows:

[0087]

[0088] The step size function q(tk) is as follows:

[0089] q(tk)=1+2tk+tk 2 tr(▽G k ρ k ▽G k );

[0090] According to the gradient direction D k Iteratively update the density matrix ρ:

[0091] ρ k+1 =ρ k +tk×D k ;

[0092] where ρ k is the density matrix updated at the kth iteration, ρ k+1 is the density matrix updated in the k+1th iteration, and tk is the gradient step size;

[0093] Calculate the probability p of positive emotion, neutral emotion and negative emotion for the density matrix after each iteration k+1 (i)

[0094] p k+1 (i) = tr(Π i ρ k+1 ∏ i );

[0095] Calculate p k+1 (i) and p i The cross entropy loss is calculated, and the iterative update of the density matrix is ​​stopped when the loss converges. The p calculated by the updated density matrix k+1 (i) is taken as the final probability, and the highest sentiment probability is selected as the sentiment analysis result.

[0096] By constructing the autocovariance matrix and the joint covariance matrix, the present invention can fully explore the internal feature correlation of text, voice and video data and the correlation between them, and combine the CCA model to maximize the correlation projection, which effectively improves the feature fusion effect of multimodal data. Compared with the traditional feature extraction method, this step can more comprehensively capture the complex information in the multimodal data and improve the accuracy and robustness of the subsequent sentiment analysis. In the present invention, the quantum state representation is used to quantize the fusion results of the text, voice and video feature vectors into quantum states. This innovative step introduces the concept of state vectors in quantum mechanics, so that the fusion of multimodal data is no longer limited to traditional linear combinations, but through quantum state superposition and weight coefficient optimization, more complex feature fusion is achieved. This greatly improves the accuracy of sentiment analysis, especially when processing nonlinear and high-dimensional data. By extracting samples of positive, neutral and negative emotions from historical customer multimodal data and substituting them into the quantum state representation for weight coefficient training, the present invention achieves higher accuracy in sentiment classification. The cross entropy loss function and the gradient descent method are used to optimize the weight coefficient of the quantum state, so that the system can accurately identify and distinguish different emotional states. The advantage of this method is that it can Dynamically adjust and optimize the sentiment classification model to ensure its adaptability and accuracy. The text, voice and video feature vectors are fused through the maximum correlation projection vector solved by the CCA model. The present invention realizes unified sentiment analysis of multimodal data. Compared with traditional methods, feature fusion based on CCA not only improves the accuracy of sentiment classification, but also makes the sentiment analysis process more efficient and consistent. This feature fusion strategy effectively solves the information redundancy problem of multimodal data in sentiment classification and improves the overall performance of the system. By constructing a density matrix and combining the sentiment classification results represented by quantum states, the present invention can accurately calculate the probability distribution of positive, neutral and negative emotions, optimize the density matrix using a gradient descent algorithm, and stop updating when the loss converges to ensure that the final calculated sentiment probability has a high degree of credibility. The projection operator optimization and objective function calculation in this process further improve the accuracy and efficiency of sentiment analysis. Through the convergence calculation of the cross entropy loss function, the updated density matrix sentiment probability is used as the final sentiment analysis result, and the sentiment category with the highest probability is selected as the output. This method can not only ensure the accuracy of sentiment classification, but also provide high stability, so that the system maintains consistent performance when processing different types of multimodal data.

[0097] S2. Cluster and stratify customers based on the results of customer sentiment analysis to generate customer tags, build a knowledge graph based on customer tags, and make personalized recommendations based on the knowledge graph;

[0098] Specifically, clustering and stratifying customers to generate customer labels according to the customer sentiment analysis results means setting the number of initial cluster centers to K, calculating the Euclidean distance between each customer fusion feature vector F and each cluster center, assigning the customer fusion feature vector F to the nearest cluster center and updating the cluster center position;

[0099] Calculate the average distance x between the i-th customer and other customers in the same cluster center and the average distance y between the i-th customer and the customer in the nearest cluster center. Calculate the silhouette coefficient S of the i-th customer based on the average distance x and the average distance y, and then get the average silhouette coefficient of all customers in all cluster centers.

[0100]

[0101] The silhouette coefficient is calculated using the following formula:

[0102] Set the iteration step to 1, iteratively increase the K value and repeatedly calculate the average silhouette coefficient corresponding to each K value until the maximum value is reached, and then stop the iteration. The K value with the largest average silhouette coefficient will be used as the latest number of cluster centers.

[0103] Repeat the customer fusion feature vector allocation and cluster center update until the cluster center position no longer changes significantly, and then stop updating;

[0104] Generate customer tags based on cluster centers and customer sentiment analysis results.

[0105] By calculating the Euclidean distance between the customer fusion feature vector F and each cluster center, the system can quickly and accurately assign customers to the most appropriate cluster. Compared with other distance measurement methods, the Euclidean distance calculation is simple and effective, and is particularly suitable for data analysis in high-dimensional space. In this way, customers with similar emotional characteristics can be effectively clustered together, laying the foundation for subsequent customer label generation. Each step in this process ensures the accuracy of customer allocation, thereby improving the quality of the entire cluster analysis. By calculating the average distance between each customer and other customers in its cluster and the average distance to the customer in the nearest cluster, the system can obtain the silhouette coefficient of each customer. This method enables the system to dynamically adjust the number of cluster centers K, and by comparing the average silhouette coefficients under different K values, finally select the K value that maximizes the average silhouette coefficient as the optimal number of clusters. Through this optimization process, the system can achieve more reasonable customer stratification, so that the clustering results have higher cohesion and separation, and improve the accuracy of customer group analysis. By setting the iteration step size to 1, gradually increasing the number of cluster centers and calculating the corresponding average silhouette coefficient after each iteration, the system can effectively search for the optimal K value. Compared with a one-time setting, the system can effectively search for the optimal K value. A fixed K value is set. This step-by-step iteration method can more flexibly adapt to the characteristics of different data sets and avoid the problem of over-clustering or under-clustering. By gradually optimizing the K value, the clustering results finally determined by the system are more stable and reliable, ensuring the scientificity and rationality of customer stratification. Through the allocation of customer fusion feature vectors and the update of cluster center positions, the system can continuously optimize the clustering results until the cluster center positions no longer change significantly. Each update of the cluster center makes the center position closer to the actual data distribution, thereby improving the accuracy of clustering. In this process, the system continuously converges through multiple iterations and finally obtains the cluster center that best represents the characteristics of the customer group. Compared with the static clustering method, this dynamic update method can better adapt to the changes in customer data and ensure the accuracy of the clustering results. Based on the final cluster center and customer sentiment analysis results, the present invention can generate accurate customer labels. These labels not only reflect the customer's emotional state, but also combine customer behavior data and clustering results, providing important support for personalized marketing and customer relationship management. Through this labeled management method, enterprises can more accurately locate target customer groups and formulate more personalized service strategies, thereby improving customer satisfaction and loyalty.

[0106] Furthermore, a knowledge graph is constructed based on customer tags. Personalized recommendation based on the knowledge graph means that after obtaining customer tags, a knowledge graph is constructed based on the customer tags, and each customer's basic information, customer tags, and fused feature vector are used as nodes, and the customer's Euclidean distance in the cluster is used as an edge to construct the knowledge graph;

[0107] The nodes are updated through the graph convolutional neural network, the Pearson correlation coefficient between customers is calculated, the correlation coefficient threshold is set, customers with a correlation coefficient greater than the correlation coefficient threshold are marked as similar customers, and customer purchase and browsing information is obtained and recommended to similar customers.

[0108] By taking the basic information of customers, customer labels and fused feature vectors as nodes, and the Euclidean distance between customers as edges, a knowledge graph is constructed. The graph can intuitively represent the relationship and similarity between customers, laying the foundation for subsequent personalized recommendations. Compared with traditional rule-based recommendation methods, knowledge graphs can adapt to changes in customer behavior and emotions more dynamically and flexibly, providing enterprises with a more comprehensive customer relationship management tool. The nodes in the knowledge graph are updated through a graph convolutional neural network (GCN). The present invention can effectively capture the complex relationships between customers. GCN gradually updates the feature vectors of nodes through convolution operations in each iteration, so that each customer node not only retains its own information, but also can learn more contextual information from its neighboring nodes. Compared with traditional graph structure analysis methods, GCN has stronger expression and learning capabilities, and can capture deeper relationships and patterns in the graph, thereby improving the accuracy of similar customer identification. By calculating the Pearson correlation coefficient between customers, the system can accurately identify similar customers. After setting the correlation coefficient threshold, the system will add the correlation coefficient greater than the threshold to the customer node. Customers who meet the threshold are marked as similar customers. This method not only takes into account the labels and behavioral characteristics between customers, but also introduces correlation analysis in statistics, making the identification of similar customers more scientific and accurate. In this way, enterprises can better understand the correlation between customers, so as to better match customer needs in personalized recommendations. By obtaining the purchase information and browsing information of similar customers, the system can recommend products or services that are more in line with their interests and needs to target customers. This recommendation method not only improves the relevance of recommendations, but also significantly improves user experience and customer satisfaction. Compared with traditional recommendation algorithms, recommendation methods based on knowledge graphs and GCN can more accurately reflect customer preferences and interests, and win more customer loyalty for enterprises in the fiercely competitive market. Knowledge graphs and GCN models can be continuously updated and optimized according to new customer data and behaviors. This dynamic optimization mechanism enables the system to adapt to changes in customer needs in real time and ensure the accuracy and timeliness of recommendations. At the same time, through the continuously updated knowledge graph, the system can continuously accumulate and mine new customer relationships and patterns, thereby providing more powerful data support for personalized recommendations.

[0109] S3. Monitor customer feedback on personalized recommendations in real time and optimize the knowledge graph based on the feedback;

[0110] Specifically, real-time monitoring of customer feedback on personalized recommendations and optimizing the knowledge graph based on feedback means collecting customer feedback data on personalized recommendations in real time after making personalized recommendations, and obtaining sentiment analysis results of customer feedback data. If the customer feedback data is positive, positive feedback is generated to record the customer and the personalized recommendation content. If it is neutral or negative, negative feedback is generated, and the knowledge graph nodes are optimized again through the graph convolutional neural network.

[0111] By monitoring customer feedback data on personalized recommendations in real time, it is possible to collect customer emotional reactions immediately after the recommendation content is released. This real-time monitoring mechanism ensures that the system can quickly identify customer satisfaction and demand changes, and provide timely data support for subsequent optimization. In this way, enterprises can respond to customer emotional changes more quickly, improve the timeliness and effectiveness of recommendations. After collecting customer feedback data, the present invention uses sentiment analysis technology to process these data to determine the customer's emotional state. Positive emotional feedback usually indicates that the customer accepts and is satisfied with the recommended content, while neutral or negative emotional feedback may indicate customer dissatisfaction or indifference to the recommended content. Through this sentiment analysis, the system can automatically classify and record customer feedback, providing a clear direction for further personalized recommendation optimization. Compared with traditional user satisfaction surveys, sentiment analysis can more directly and accurately reflect the customer's emotional state, improving the system's The system's ability to perceive customer needs. Customer feedback data is classified as positive or negative feedback based on the results of sentiment analysis. Positive feedback will record the correspondence between the customer and the recommended content, forming a positive reinforcement mechanism, thereby providing support for the continued use of similar content in subsequent recommendations. For neutral or negative feedback, the system will generate negative feedback records and further analyze the reasons for these feedbacks. Through this distinction and recording mechanism, the system can effectively capture and utilize customer satisfaction and dissatisfaction information, thereby avoiding similar mistakes in future recommendations and improving customer overall satisfaction. By continuously collecting customer feedback, conducting sentiment analysis, recording positive and negative feedback, and dynamically optimizing the knowledge graph based on negative feedback, the system can gradually improve the quality of recommendations after each recommendation. Such dynamic adjustment and continuous optimization mechanisms ensure that the recommendation system can keep pace with changes in customer needs, thereby maintaining a high level of user experience and satisfaction over the long term.

[0112] S4. Generate customer recommendation strategies based on knowledge graphs and comprehensively form customer decision support reports;

[0113] Specifically, generating a customer recommendation strategy based on the knowledge graph and comprehensively forming a customer decision support report means comprehensively generating a customer recommendation strategy based on the purchase information and browsing information of all similar customers of the customer in the knowledge graph, adjusting the customer recommendation strategy based on customer feedback data to form a customer decision support report, and storing the customer decision support report.

[0114] By analyzing the purchase and browsing information of customers and their similar customers in the knowledge graph, a personalized customer recommendation strategy is generated. Compared with the traditional recommendation system, this recommendation method based on the knowledge graph can more comprehensively consider the behavioral correlation between customers and their similar customers, thereby improving the accuracy of the recommended content. By combining the multi-dimensional information of customers and similar customers, the system can better match the interests and needs of customers in the recommended content, significantly improving user experience and customer satisfaction. After generating the preliminary customer recommendation strategy, the present invention collects and analyzes customer feedback data on the recommended content, and adjusts the recommendation strategy in real time. Whether it is positive feedback or negative feedback, the system will use this data to optimize the recommendation algorithm to ensure that subsequent recommendations are more accurate and personalized. Through this continuous optimization mechanism, the system can not only dynamically adapt to changes in customer needs, but also gradually improve the effectiveness of recommendations in long-term use. This adjustment process improves the responsiveness of the recommendation system, so that the recommended content can always meet the latest customer needs. On the basis of generating and adjusting customer recommendation strategies, the present invention further integrates the recommendation results and customer feedback data to form a customer decision support report. The report not only includes the current recommendation strategy, but also provides an in-depth analysis of customer behavior and future trend forecasts. Through this report, enterprises can obtain key insights into customer needs and market trends, so as to make more informed decisions. As an intelligent tool, the decision support report can significantly improve the market response capabilities and strategic planning levels of enterprises.

[0115] Example 2

[0116] Reference Figure 3 , which is the second embodiment of the present invention, which is different from the previous embodiment and provides a customer information resource integration and analysis system based on big data, which includes:

[0117] Data collection module, used to collect multimodal customer data and perform preprocessing to extract data features;

[0118] Sentiment analysis module, used to analyze the data features of customer multimodal data to obtain sentiment analysis results;

[0119] Personalized recommendation module, which is used to perform cluster analysis based on sentiment analysis results and customer multimodal data, and build a knowledge graph for personalized recommendations;

[0120] The feedback optimization module is used to collect and analyze customer feedback data after personalized recommendations, optimize the knowledge graph, and simultaneously record personalized recommendation content.

[0121] 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 a computer device (which can be a personal computer, a server, or a 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, etc., which can store program codes.

[0122] 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.

[0123] 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.

[0124] 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.

[0125] 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 customer information resource integration and analysis method based on big data, characterized by: include, Collect multimodal customer data, perform preprocessing to extract data features, and perform customer sentiment analysis based on the data features; According to the results of customer sentiment analysis, customers are clustered and stratified to generate customer tags, and a knowledge graph is built based on the customer tags, and personalized recommendations are made based on the knowledge graph; Monitor customer feedback on personalized recommendations in real time and optimize the knowledge graph based on the feedback; Generate customer recommendation strategies based on the knowledge graph and comprehensively form a customer decision support report.

2. The method for integrating and analyzing customer information resources based on big data according to claim 1, characterized in that: The collecting of multimodal data of customers for preprocessing and extracting data features refers to collecting text data, voice data, video data and basic information of customers, and preprocessing them respectively; Use the BERT model to convert text data into a text feature vector T; Use the openSMILE model to extract the MFCC features of speech data, and convert the MFCC features into speech feature vectors S through the LSTM model; ResNes-50 is used to extract the spatial features of video data, and the video feature vector V is obtained by combining LSTM to process the spatial features.

3. The method for integrating and analyzing customer information resources based on big data according to claim 2, characterized in that: The customer sentiment analysis based on data features includes: Based on the text feature vector T, the speech feature vector S and the video feature vector V, the autocovariance matrix C is constructed respectively. T , C S and C V , and construct the joint covariance matrix C of the three TSV ; According to the text feature vector T, the speech feature vector S and the video feature vector V, the text feature projection vector a, the speech feature projection vector b and the video feature projection vector c are obtained respectively. The CCA model is used to maximize the correlation of the feature vectors to obtain the maximum correlation projection a of the text feature, speech feature and video feature in the high-dimensional space. T 、a S and a V ; A T 、a S and a V By expressing the quantum state, we can get |T CCA >、|S CCA > and |V CCA >, define T CCA >、|S CCA > and |V CCA The weight coefficients of > are α, β and γ respectively, and the comprehensive quantum state representation is obtained through the weight coefficients From the historical customer multimodal data, we obtain the |T representing positive emotion, neutral emotion and negative emotion respectively. CCA >1、|S CCA >1 and |V CCA >1, and substitute them into the comprehensive quantum state representation for weight coefficient training, use the cross entropy loss function and gradient descent method to adjust the weight coefficient, and obtain the weight coefficients of the comprehensive quantum state representation of positive emotion, neutral emotion and negative emotion respectively, and apply the weight coefficients of the comprehensive quantum state representation of positive emotion, neutral emotion and negative emotion to |T CCA >、|S CCA > and |V CCA > Obtain comprehensive quantum state representation of positive emotions, neutral emotions, and negative emotions The maximum correlation projection a solved using the CCA model T 、a S and a V Fusion of text feature vector T, speech feature vector S and video feature vector V: Where F is the fusion feature vector; Construct a CNN model and use the training set for training. Input the fused feature vector F into the CNN model to obtain the probability p of positive emotion, neutral emotion, and negative emotion of the fused feature vector F. i ; According to the comprehensive quantum state representation of positive emotion, neutral emotion and negative emotion and the probability p of positive emotion, neutral emotion and negative emotion of the fused feature vector F i Construct the density matrix ρ: in for The accompanying state of According to the comprehensive quantum state representation of positive emotion, neutral emotion and negative emotion Initialize the corresponding projection operator Π i , using the gradient descent algorithm to project the operator Π i Perform training and calculate the objective function G: Among them, tr represents trace operation; Calculate the gradient ▽G according to the objective function and calculate the gradient direction D k , according to the gradient direction D k Iteratively update the density matrix ρ: r k+1 =ρ k +tk×D k ; where ρ k is the density matrix updated at the kth iteration, ρ k+1 is the density matrix updated in the k+1th iteration, and tk is the gradient step size; Calculate the probability p of positive emotion, neutral emotion and negative emotion for the density matrix after each iteration k+1 (i) p k+1 (i)=tr(Π i r k+1 P i ); Calculate p k+1 (i) and p i The cross entropy loss is calculated, and the iterative update of the density matrix is ​​stopped when the loss converges. The p calculated by the updated density matrix k+1 (i) is taken as the final probability, and the highest sentiment probability is selected as the sentiment analysis result.

4. The method for integrating and analyzing customer information resources based on big data according to claim 3, characterized in that: The step of clustering and stratifying customers according to the customer sentiment analysis results to generate customer labels refers to setting the number of initial cluster centers to K, calculating the Euclidean distance between each customer fusion feature vector F and each cluster center, assigning the customer fusion feature vector F to the nearest cluster center and updating the cluster center position; Calculate the average distance x between the i-th customer and other customers in the same cluster center and the average distance y between the i-th customer and the customer in the nearest cluster center. Calculate the silhouette coefficient S of the i-th customer based on the average distance x and the average distance y, and then get the average silhouette coefficient of all customers in all cluster centers. Set the iteration step to 1, iteratively increase the K value and repeatedly calculate the average silhouette coefficient corresponding to each K value until the maximum value is reached, and then stop the iteration. The K value with the largest average silhouette coefficient will be used as the latest number of cluster centers. Repeat the customer fusion feature vector allocation and cluster center update until the cluster center position no longer changes significantly, and then stop updating; Generate customer tags based on cluster centers and customer sentiment analysis results.

5. The method for integrating and analyzing customer information resources based on big data according to claim 4, characterized in that: The construction of the knowledge graph based on customer tags and the personalized recommendation based on the knowledge graph refer to the construction of the knowledge graph based on the customer tags after obtaining the customer tags, taking the basic information of each customer, the customer tags and the fusion feature vector as nodes, and taking the Euclidean distance of the customer in the cluster as edges to construct the knowledge graph; The nodes are updated through the graph convolutional neural network, the Pearson correlation coefficient between customers is calculated, the correlation coefficient threshold is set, customers with a correlation coefficient greater than the correlation coefficient threshold are marked as similar customers, and customer purchase information and browsing information are obtained and recommended to similar customers.

6. The method for integrating and analyzing customer information resources based on big data according to claim 5, characterized in that: The real-time monitoring of customer feedback on personalized recommendations and optimizing the knowledge graph based on feedback refers to collecting customer feedback data on personalized recommendations in real time after personalized recommendations are made, and obtaining sentiment analysis results of customer feedback data. If the customer feedback data is positive, positive feedback is generated to record the customer and the personalized recommendation content. If it is neutral or negative, negative feedback is generated, and the knowledge graph nodes are optimized again through the graph convolutional neural network.

7. The method for integrating and analyzing customer information resources based on big data according to claim 6, characterized in that: Generating a customer recommendation strategy based on the knowledge graph and comprehensively forming a customer decision support report refers to comprehensively generating a customer recommendation strategy based on the purchase information and browsing information of all similar customers of the customer in the knowledge graph, adjusting the customer recommendation strategy based on customer feedback data to form a customer decision support report, and storing the customer decision support report.

8. A customer information resource integration and analysis system based on big data according to any one of claims 1 to 7, characterized in that: include, Data collection module, used to collect multimodal customer data and perform preprocessing to extract data features; Sentiment analysis module, used to analyze the data features of customer multimodal data to obtain sentiment analysis results; Personalized recommendation module, which is used to perform cluster analysis based on sentiment analysis results and customer multimodal data, and build a knowledge graph for personalized recommendations; The feedback optimization module is used to collect and analyze customer feedback data after personalized recommendations, optimize the knowledge graph, and simultaneously record personalized recommendation content.

9. 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 information resource integration and analysis method based on big data described in any one of claims 1 to 7 are implemented.

10. 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 information resource integration and analysis method based on big data described in any one of claims 1 to 7 are implemented.

Images