Enterprise intelligent management method and system based on deep learning

Abstract

The invention relates to an enterprise intelligent management method and system based on deep learning. The method comprises the steps of performing semantic analysis on transaction records, service requests and product service feedback data of customers by adopting an artificial intelligence technology based on deep learning, mining a deep interaction mode between the transaction records and the service requests of the customers and the product service feedback data, deeply understanding customer behaviors and intelligently predicting customer loss risks. And making a customer retention strategy for the customer with high loss risk, continuously tracking the subsequent behavior of the user, and dynamically adjusting the customer retention strategy. Through the mode, an enterprise can be helped to accurately identify potential customer loss risks, so that corresponding strategies can be conveniently adopted for customer reservation, and customer satisfaction and enterprise operation benefits are improved.

Description

Technical Field

[0001] The present application relates to the field of enterprise digital data processing technology, and specifically, to an enterprise intelligent management method and system based on deep learning. Background Art

[0002] In the global economic wave of the 21st century, the rapid development of information technology and the wave of digital transformation of enterprises are reshaping the business environment at an unprecedented speed. In the current fierce market competition, the effective management and maintenance of customer relationships have become the key to the sustainable development of enterprises. Taking the retail industry as an example, according to a survey by Gartner, 60% of consumers will give up during the purchase process due to poor experience, and these experience problems often stem from the company's lack of understanding of customer behavior. In the face of ever-changing consumer needs, if companies cannot adjust their strategies in time and provide personalized and customized services, customers are likely to turn to competitors, resulting in a decline in the company's customer loyalty and market share.

[0003] However, in past business practices, companies usually rely on manual methods to collect, organize and analyze customer data in order to gain insight into customer needs and optimize service experience. However, this traditional model is powerless in the face of the explosive growth and complexity of current data. Manual analysis is not only time-consuming and labor-intensive, but also prone to information omissions and difficult to capture the dynamic changes in customer behavior and potential needs, which undoubtedly increases the risk of customer churn and weakens the market competitiveness of enterprises.

[0004] Therefore, we look forward to an intelligent enterprise management method based on deep learning. Summary of the invention

[0005] This summary is provided to introduce concepts in a brief form that will be described in detail in the detailed description below. This summary is not intended to identify key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0006] In a first aspect, the present application provides an enterprise intelligent management method based on deep learning, the method comprising: Collect transaction records, service requests, and product service feedback data of the analyzed customer objects through data collection tools; Transmitting the transaction records, service requests and product service feedback data of the analyzed customer object to the central platform to obtain a customer behavior data set; On the central platform, deep learning technology is used to analyze customer behavior patterns of the customer behavior data set to obtain prediction results of customer churn risk; Formulate customer retention strategies for customers with high churn risk based on the prediction results; Continue to track the subsequent behavior data of the high-churn risk customers, update the customer behavior data set, and regularly adjust and optimize the customer retention strategy based on the prediction results of the customer churn risk of the updated customer behavior data set.

[0007] Optionally, in the central platform, deep learning technology is used to perform customer behavior pattern analysis on the customer behavior data set to obtain a prediction result of customer churn risk, including: respectively extracting semantic features of service requests and transaction records in the customer behavior data set to obtain transaction record semantic encoding feature vectors and service request semantic encoding feature vectors; performing multimodal joint encoding on the image part and the text part in the product service feedback data in the customer behavior data set to obtain a service feedback multimodal joint representation matrix; inputting the transaction record semantic encoding feature vector and the service request semantic encoding feature vector into a fine-grained cross-modal saliency fusion module to obtain a product front-end multimodal fusion representation vector; inputting the product front-end multimodal fusion representation vector and the service feedback multimodal joint representation matrix into a local granularity interactive response network based on feature decomposition to obtain a product front-end-product back-end decoupled interaction global representation vector; based on the product front-end-product back-end decoupled interaction global representation vector, generating the prediction result, and the prediction result is used to represent a customer churn risk level label.

[0008] Optionally, the semantic features of the service request and the transaction record in the customer behavior dataset are extracted respectively to obtain a transaction record semantic coding feature vector and a service request semantic coding feature vector, including: inputting the service request into a semantic expression expander based on the AIGC model to obtain a semantically expanded service request; and semantically encoding the transaction record and the semantically expanded service request respectively to obtain the transaction record semantic coding feature vector and the service request semantic coding feature vector.

[0009] Optionally, multimodal joint encoding is performed on the image part and the text part in the product service feedback data in the customer behavior dataset to obtain a service feedback multimodal joint representation matrix, including: extracting the image part and the text part in the product service feedback data to obtain service feedback image data and service feedback text data; inputting the service feedback image data and the service feedback text data into a CLIP model including a text encoder and an image encoder to obtain the service feedback multimodal joint representation matrix.

[0010] Optionally, the transaction record semantic encoding feature vector and the service request semantic encoding feature vector are input into a fine-grained cross-modal saliency fusion module to obtain a product front-end multimodal fusion representation vector, including: calculating the global eigenvalue association matrix between the transaction record semantic encoding feature vector and the service request semantic encoding feature vector; inputting the global eigenvalue association matrix into a learnable gating function to obtain a correlation distinguishable weight matrix; using the correlation distinguishable weight matrix as a fusion mask matrix, respectively calculating the matrix product between the transaction record semantic encoding feature vector, the service request semantic encoding feature vector and the fusion mask matrix to obtain a distinguishing and enhancing transaction record semantic encoding feature vector and a distinguishing and enhancing service request semantic encoding feature vector; fusing the distinguishing and enhancing transaction record semantic encoding feature vector and the distinguishing and enhancing service request semantic encoding feature vector to obtain the product front-end multimodal fusion representation vector.

[0011] Optionally, the product front-end multimodal fusion representation vector and the service feedback multimodal joint representation matrix are input into a local granular interactive response network based on feature decomposition to obtain a product front-end-product back-end decoupled interaction global representation vector, including: performing feature decoupling on the product front-end multimodal fusion representation vector and the service feedback multimodal joint representation matrix to obtain a sequence of product front-end multimodal fusion local representation vectors and a sequence of service feedback multimodal joint local representation vectors; calculating the product front-end multimodal fusion local representation vector and the service feedback multimodal joint local representation vector sequence for each group of corresponding product front-end multimodal fusion local representation vectors and service feedback multimodal joint local representation vectors. Representing the local semantic association interaction vectors between the vectors to obtain a sequence of product front-end-product back-end local semantic association interaction vectors; calculating the fine-grained attention weights of each product front-end-product back-end local semantic association interaction vector in the sequence of product front-end-product back-end local semantic association interaction vectors to obtain a sequence of fine-grained attention weights; inputting the sequence of fine-grained attention weights into a random deactivation module to obtain a sequence of fine-grained attention weights after deactivation processing; using the sequence of fine-grained attention weights after deactivation processing as the weight distribution, calculating the weighted sum of the sequence of product front-end-product back-end local semantic association interaction vectors to obtain the product front-end-product back-end decoupled interaction global representation vector.

[0012] Optionally, the local semantic association interaction vector between each group of corresponding product front-end multimodal fusion local representation vectors and service feedback multimodal joint local representation vectors in the sequence of the product front-end multimodal fusion local representation vectors and the sequence of the service feedback multimodal joint local representation vectors is calculated to obtain a sequence of product front-end-product back-end local semantic association interaction vectors, including: calculating the position point addition, position point subtraction and position point multiplication between the product front-end multimodal fusion local representation vector and the corresponding service feedback multimodal joint local representation vector to obtain a first fusion result, a second fusion result and a third fusion result; after cascading the first fusion result, the second fusion result and the third fusion result into a product front-end-product back-end multi-scale fusion feature vector, one-dimensional convolution processing and maximum pooling processing are performed on them to obtain the product front-end-product back-end local semantic association interaction vector.

[0013] Optionally, the fine-grained attention weights of each product front-end-product back-end local semantic association interaction vector in the sequence of the product front-end-product back-end local semantic association interaction vectors are calculated to obtain a sequence of fine-grained attention weights, including: multiplying the product front-end-product back-end local semantic association interaction vector by a weight coefficient matrix, adding the result to the bias vector and then passing the result through a tanh function to obtain a product front-end-product back-end local semantic association interaction encoding feature vector; using an attention score reference vector multiplied by the product front-end-product back-end local semantic association interaction encoding feature vector to obtain the fine-grained attention weight.

[0014] Optionally, based on the product front-end-product back-end decoupled interaction global representation vector, the prediction result is generated, and the prediction result is used to represent the risk level label of customer churn, including: inputting the product front-end-product back-end decoupled interaction global representation vector into a classifier-based customer churn prediction module to obtain the prediction result.

[0015] In a second aspect, the present application provides an enterprise intelligent management system based on deep learning, the system comprising: The data collection module of the analyzed customer object is used to collect the transaction records, service requests and product service feedback data of the analyzed customer object through the data collection tool; A data transmission module, used to transmit the transaction records, service requests and product service feedback data of the analyzed customer object to the central platform to obtain a customer behavior data set; A customer behavior pattern analysis module is used to perform customer behavior pattern analysis on the customer behavior data set using deep learning technology on the central platform to obtain a prediction result of customer churn risk; A customer retention strategy formulation module, used to formulate a customer retention strategy for customers with high churn risk according to the prediction results; The customer churn risk result prediction module is used to continuously track the subsequent behavior data of the high churn risk customers, update the customer behavior data set, and regularly adjust and optimize the customer retention strategy based on the prediction results of the customer churn risk of the updated customer behavior data set.

[0016] The above technical solution uses artificial intelligence technology based on deep learning to perform semantic analysis on customer transaction records, service requests, and product service feedback data, dig out the deep interaction pattern between customer transaction records, service requests, and product service feedback data, deeply understand customer behavior, and intelligently predict customer churn risk, and then formulate customer retention strategies for customers with high churn risk, and continue to track user subsequent behavior and dynamically adjust customer retention strategies. In this way, it can help companies accurately identify potential customer churn risks so that they can adopt corresponding strategies for customer retention, thereby improving customer satisfaction and corporate operating efficiency.

[0017] Other features and advantages of the present application will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and other features, advantages and aspects of the embodiments of the present application will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale. In the drawings: Figure 1 It is a flowchart of an enterprise intelligent management method based on deep learning according to an exemplary embodiment.

[0019] Figure 2 is based on Figure 1 The illustrated embodiment shows a flowchart of step S103 of an enterprise intelligent management method based on deep learning.

[0020] Figure 3 It is a block diagram of an enterprise intelligent management system based on deep learning according to an exemplary embodiment.

[0021] Figure 4 It is a block diagram of an electronic device according to an exemplary embodiment.

[0022] Figure 5 This is an application scenario diagram of an enterprise intelligent management method based on deep learning according to an exemplary embodiment. The embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be construed as being limited to the embodiments described herein. Instead, these embodiments are provided to provide a more thorough and complete understanding of the present application. It should be understood that the drawings and embodiments of the present application are only for exemplary purposes and are not intended to limit the scope of protection of the present application.

[0023] It should be understood that the various steps described in the method implementation of the present application can be performed in different orders and / or performed in parallel. In addition, the method implementation may include additional steps and / or omit the steps shown. The scope of the present application is not limited in this respect.

[0024] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0025] It should be noted that the concepts such as "first" and "second" mentioned in this application are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0026] It should be noted that the modifications of "one" and "plurality" mentioned in the present application are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0027] The names of the messages or information exchanged between multiple devices in the embodiments of the present application are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0028] The specific implementation of the present application is described in detail below with reference to the accompanying drawings.

[0029] In response to the above technical problems, the technical concept of this application is to use artificial intelligence technology based on deep learning to perform semantic analysis on customer transaction records, service requests and product service feedback data, dig out the deep interaction pattern between customer transaction records and service requests and product service feedback data, deeply understand customer behavior and intelligently predict customer churn risk, and then formulate customer retention strategies for customers with high churn risk, and continue to track user subsequent behavior and dynamically adjust customer retention strategies. In this way, it can help companies accurately identify potential customer churn risks so that they can adopt corresponding strategies for customer retention, thereby improving customer satisfaction and corporate operating efficiency.

[0030] Figure 1 is a flowchart of an enterprise intelligent management method based on deep learning according to an exemplary embodiment. Figure 1 As shown, the method includes: Step S101: collect transaction records, service requests and product service feedback data of the analyzed customer object through a data collection tool; Step S102: Transmit the transaction records, service requests and product service feedback data of the analyzed customer object to the central platform to obtain a customer behavior data set; Step S103: On the central platform, deep learning technology is used to analyze the customer behavior pattern of the customer behavior data set to obtain a prediction result of customer churn risk; Step S104: formulating a customer retention strategy for customers with high churn risk according to the prediction results; Step S105: Continue to track the subsequent behavior data of the high churn risk customers, update the customer behavior data set, and regularly adjust and optimize the customer retention strategy based on the prediction results of the customer churn risk of the updated customer behavior data set.

[0031] It should be understood that customer transaction records refer to various interaction data during the purchase process, including the products or services purchased, purchase time, purchase frequency, etc.; service requests reflect customers' needs for purchased products or services and the problems encountered, including information such as repair requests or consulting questions; and product service feedback data is a direct evaluation of customers' product or service experience, including customer satisfaction ratings, comments, complaints and other information. These customer behavior data together construct a multi-dimensional view of customer-enterprise interaction, providing enterprises with a window to fully understand customer behavior and needs. Through deep learning analysis of customer transaction records, service requests and product service feedback data, it is possible to predict customer churn risks, so as to formulate corresponding customer retention strategies, improve customer satisfaction and enterprise operational efficiency, and thus realize intelligent enterprise management.

[0032] Figure 2is based on Figure 1 The embodiment shown is a flowchart of step S103 of a method for intelligent enterprise management based on deep learning, as shown in FIG. Figure 2 As shown, step S103, on the central platform, using deep learning technology to analyze the customer behavior data set for customer behavior pattern analysis to obtain a prediction result of customer churn risk, including: Step S1031, respectively extracting semantic features of service requests and transaction records in the customer behavior dataset to obtain transaction record semantic coding feature vectors and service request semantic coding feature vectors; Step S1032: performing multimodal joint encoding on the image part and the text part in the product service feedback data in the customer behavior dataset to obtain a service feedback multimodal joint representation matrix; Step S1033: input the transaction record semantic encoding feature vector and the service request semantic encoding feature vector into a fine-grained cross-modal saliency fusion module to obtain a product front-end multimodal fusion representation vector; Step S1034: input the product front-end multimodal fusion representation vector and the service feedback multimodal joint representation matrix into a local granularity interactive response network based on feature decomposition to obtain a product front-end-product back-end decoupled interaction global representation vector; Step S1035: Generate the prediction result based on the product front-end-product back-end decoupling interaction global representation vector, where the prediction result is used to represent the risk level label of customer churn.

[0033] Specifically, in the process of predicting customer churn risk, first of all, it is considered that the original service request captured by the system may only contain relatively concise or vague requirements directly expressed by the user, and may show great differences due to various factors such as customer language habits and cultural background. Therefore, in order to improve the accuracy of data analysis, this application first uses a semantic expression expander based on the AIGC model to perform in-depth analysis and semantic expansion of the service request to generate a more detailed and accurate request description, thereby obtaining a semantically expanded service request, which provides higher quality input for subsequent customer behavior pattern analysis.

[0034] In one embodiment of the present application, semantic features of service requests and transaction records in the customer behavior dataset are extracted respectively to obtain transaction record semantic coding feature vectors and service request semantic coding feature vectors, including: inputting the service request into a semantic expression expander based on an AIGC model to obtain a semantically expanded service request; and semantically encoding the transaction record and the semantically expanded service request respectively to obtain the transaction record semantic coding feature vector and the service request semantic coding feature vector.

[0035] Next, the transaction record and the semantically expanded service request are further semantically encoded to extract the deep semantic information of the customer's purchasing behavior and service needs, respectively, to mine the behavioral characteristics of the customer during the purchase and use process, and to convert the original text data into a vector representation that can be understood by the deep learning model, thereby obtaining a transaction record semantic encoding feature vector and a service request semantic encoding feature vector. In an embodiment of the present application, a pre-trained BERT model can be used as a semantic encoder to utilize the bidirectional encoding capability of the BERT model to capture the deep contextual semantic information in the transaction record and the semantically expanded service request.

[0036] At the same time, it is considered that the product service feedback data usually includes multimodal information such as image comments and text comments. Therefore, in order to fully understand the customer's emotions and experience of the product and subsequent services, in the technical solution of the present application, first, the image part and the text part in the product service feedback data are extracted, and the two modal data are separated to obtain service feedback image data and service feedback text data. Then, the CLIP model is used to process the service feedback image data and the service feedback text data. Those of ordinary skill in the art should know that the CLIP model includes a text encoder and an image encoder, which can semantically encode text data and image data respectively, and perform cross-modal contrastive learning on the output of the text encoder and the image encoder to capture the deep association between images and texts, and ensure the consistency of information in multimodal data, thereby obtaining a service feedback multimodal joint representation matrix.

[0037] In one embodiment of the present application, multimodal joint encoding is performed on the image part and the text part in the product service feedback data in the customer behavior dataset to obtain a service feedback multimodal joint representation matrix, including: extracting the image part and the text part in the product service feedback data to obtain service feedback image data and service feedback text data; inputting the service feedback image data and the service feedback text data into a CLIP model including a text encoder and an image encoder to obtain the service feedback multimodal joint representation matrix.

[0038] Then, considering that the customer's transaction records and service requests are a direct reflection of the customer's behavior, and the product service feedback data is a supplement and verification of the former two, the three together constitute a complete picture of customer behavior. Therefore, in order to effectively utilize the above-mentioned multi-dimensional customer behavior data, in the technical solution of the present application, first, the transaction record semantic encoding feature vector and the service request semantic encoding feature vector are subjected to feature fusion processing. Specifically, the present application introduces a fine-grained cross-modal significant fusion module to realize this fusion process. The fine-grained cross-modal significant fusion module calculates the global eigenvalue association matrix between the transaction record semantic encoding feature vector and the service request semantic encoding feature vector to establish a global association between customer transaction records and service requests. Then, the gating function is used to perform feature learning on the global association information between the two, and the correlation distinguishable weight matrix of the eigenvalue granularity is adaptively generated, which is used as a mask matrix to perform feature distinction and enhancement processing on the transaction record semantic encoding feature vector and the service request semantic encoding feature vector, thereby improving the pertinence and effectiveness of feature fusion. Finally, the differentiated and enhanced transaction record semantic features and service request semantic features are integrated through cascading operations to obtain the product front-end multimodal fusion representation vector, thereby achieving a comprehensive understanding of customer behavior patterns.

[0039] In one embodiment of the present application, the transaction record semantic encoding feature vector and the service request semantic encoding feature vector are input into a fine-grained cross-modal saliency fusion module to obtain a product front-end multimodal fusion representation vector, including: calculating the global eigenvalue association matrix between the transaction record semantic encoding feature vector and the service request semantic encoding feature vector; inputting the global eigenvalue association matrix into a learnable gating function to obtain a correlation distinguishable weight matrix; using the correlation distinguishable weight matrix as a fusion mask matrix, respectively calculating the matrix product between the transaction record semantic encoding feature vector, the service request semantic encoding feature vector and the fusion mask matrix to obtain a distinguishing and enhancing transaction record semantic encoding feature vector and a distinguishing and enhancing service request semantic encoding feature vector; fusing the distinguishing and enhancing transaction record semantic encoding feature vector and the distinguishing and enhancing service request semantic encoding feature vector to obtain the product front-end multimodal fusion representation vector.

[0040] Specifically, the transaction record semantic encoding feature vector and the service request semantic encoding feature vector are processed by the following feature interaction response fusion formula to obtain the product front-end multimodal fusion representation vector, wherein the feature interaction response fusion formula is: in, represents the semantic encoding feature vector of the transaction record, represents the semantic encoding feature vector of the service request, represents the transpose of the eigenvector, represents the global eigenvalue incidence matrix, represents the exponential function operation with e as the base, represents the correlation distinguishable weight matrix, represents the matrix multiplication operation, represents the semantic encoding feature vector for distinguishing and strengthening transaction records, represents the semantic encoding feature vector of the differentiated enhanced service request, Indicates cascade operation, Represents the multimodal fusion representation vector of the product front end.

[0041] Since the product front-end multimodal fusion representation vector and the service feedback multimodal joint representation matrix represent the customer's behavior and feedback data before and after using the product, there is a close correlation and complementarity between the two. In the technical solution of the present application, in order to deeply understand the customer's behavior pattern throughout the product life cycle, the product front-end multimodal fusion representation vector and the service feedback multimodal joint representation matrix are further input into the local granularity interactive response network based on feature decomposition to obtain the product front-end-product back-end decoupled interaction global representation vector. Specifically, the local granularity interactive response network based on feature decomposition first decomposes the product front-end multimodal fusion representation vector and the service feedback multimodal joint representation matrix into a sequence of local feature vectors through a decoupling operation to increase the processing capability of detail information. Then, based on the fine-grained semantic association relationship between the sequences of the two sets of local feature vectors after decoupling, a local semantic association interaction vector sequence and an attention weight sequence are generated, so that the model can focus on the most important feature interactions, thereby improving the quality of feature representation and prediction accuracy. At the same time, in order to prevent overfitting, the generated attention weight sequence is further randomly deactivated to achieve weight sparsity and enhance the generalization ability of the model. Finally, the attention weight sequence after deactivation is used as the weight distribution to perform weighted summation on the local semantic association interaction vector sequence to synthesize the local interaction information in the product front-end multimodal fusion representation vector and the service feedback multimodal joint representation matrix, and generate a product front-end-product back-end decoupled interaction global representation vector, thereby realizing the understanding of the customer's full life cycle behavior pattern.

[0042] In one embodiment of the present application, the product front-end multimodal fusion representation vector and the service feedback multimodal joint representation matrix are input into a local granularity interactive response network based on feature decomposition to obtain a product front-end-product back-end decoupled interaction global representation vector, including: performing feature decoupling on the product front-end multimodal fusion representation vector and the service feedback multimodal joint representation matrix to obtain a sequence of product front-end multimodal fusion local representation vectors and a sequence of service feedback multimodal joint local representation vectors; calculating the product front-end multimodal fusion local representation vector and the service feedback multimodal joint local representation vector in the sequence of the product front-end multimodal fusion local representation vector and the service feedback multimodal joint local representation vector. Combine the local semantic association interaction vectors between the local representation vectors to obtain a sequence of product front-end-product back-end local semantic association interaction vectors; calculate the fine-grained attention weights of each product front-end-product back-end local semantic association interaction vector in the sequence of product front-end-product back-end local semantic association interaction vectors to obtain a sequence of fine-grained attention weights; input the sequence of fine-grained attention weights into a random deactivation module to obtain a sequence of fine-grained attention weights after deactivation processing; use the sequence of fine-grained attention weights after deactivation processing as the weight distribution, calculate the weighted sum of the sequence of product front-end-product back-end local semantic association interaction vectors to obtain the product front-end-product back-end decoupled interaction global representation vector.

[0043] Furthermore, in one embodiment of the present application, the local semantic association interaction vector between each group of corresponding product front-end multimodal fusion local representation vectors and service feedback multimodal joint local representation vectors in the sequence of the product front-end multimodal fusion local representation vectors and the sequence of the service feedback multimodal joint local representation vectors is calculated to obtain a sequence of product front-end-product back-end local semantic association interaction vectors, including: calculating the position point addition, position point subtraction and position point multiplication between the product front-end multimodal fusion local representation vector and the corresponding service feedback multimodal joint local representation vector to obtain a first fusion result, a second fusion result and a third fusion result; after cascading the first fusion result, the second fusion result and the third fusion result into a product front-end-product back-end multi-scale fusion feature vector, one-dimensional convolution processing and maximum pooling processing are performed on them to obtain the product front-end-product back-end local semantic association interaction vector.

[0044] Furthermore, in one embodiment of the present application, the fine-grained attention weights of each product front-end-product back-end local semantic association interaction vector in the sequence of the product front-end-product back-end local semantic association interaction vectors are calculated to obtain a sequence of fine-grained attention weights, including: multiplying the product front-end-product back-end local semantic association interaction vector by a weight coefficient matrix, adding it to the bias vector and then passing it through a tanh function to obtain a product front-end-product back-end local semantic association interaction encoding feature vector; using an attention score reference vector multiplied by the product front-end-product back-end local semantic association interaction encoding feature vector to obtain the fine-grained attention weight. Specifically, the product front-end multimodal fusion representation vector and the service feedback multimodal joint representation matrix are processed with the following interaction response analysis formula to obtain the product front-end-product back-end decoupled interaction global representation vector: in, represents the multimodal fusion representation vector of the product front end, represents the multimodal joint representation matrix of the service feedback, Indicates decoupling processing, , and They represent the first, second, and third in the sequence of the multimodal fusion local representation vector of the product front end. and The multimodal fusion local representation vector of the product front end, The value of is the number of feature vectors in the sequence of the multimodal fusion local representation vector of the product front end, , , and Respectively represent the first, second, and third in the sequence of the multimodal joint local representation vector of the service feedback and service feedback multimodal joint local representation vector, Indicates cascade operation, represents the maximum pooling operation, represents a one-dimensional convolution operation, represents dot product, Indicates point addition, Indicates point reduction, The first in the sequence of the product front-end-product back-end local semantic association interaction vector product front-end-product back-end local semantic association interaction vector, is the weight coefficient matrix, is the bias vector, is the hyperbolic tangent function, is the attention score reference vector, Indicates fine-grained attention weights, represents random dropout processing, Indicates The fine-grained attention weights after dropout processing, Represents the product front-end-product back-end local semantic association interaction vector.

[0045] Then, the product front-end-product back-end decoupled interaction global representation vector is input into the classifier-based customer churn prediction module to obtain a prediction result, which is used to represent the risk level label of customer churn. That is, the product front-end-product back-end decoupled interaction global representation vector is feature parsed using a pre-trained classifier model, and potential patterns and features related to customer churn are mined through layer-by-layer abstraction and information extraction, and converted into probability distributions corresponding to different risk level labels, and the level label corresponding to the maximum probability value is used as the final prediction result to provide data support for business decisions.

[0046] In one embodiment of the present application, the prediction result is generated based on the product front-end-product back-end decoupled interaction global representation vector, and the prediction result is used to represent a risk level label of customer churn, including: inputting the product front-end-product back-end decoupled interaction global representation vector into a classifier-based customer churn prediction module to obtain the prediction result.

[0047] After obtaining the prediction results, the system will automatically identify customers with high churn risk and generate preliminary retention strategies based on predefined customer retention strategy templates, such as providing personalized offers, optimizing service processes, and strengthening customer care. At the same time, the subsequent behavior data of high-risk customers will be continuously tracked, and their customer churn prediction results will be dynamically updated based on new behavior data to reflect the latest changes in customer behavior, and customer retention strategies will be adjusted and optimized accordingly to ensure the timeliness and pertinence of the strategies.

[0048] In a preferred example, it is considered that the product front-end multimodal fusion representation vector and the service feedback multimodal joint representation matrix respectively represent the textual semantic multimodal fusion features of transaction records and service requests and the cross-domain joint coding features between image data and text data in service feedback. In this way, when the product front-end multimodal fusion representation vector and the service feedback multimodal joint representation matrix are input into the local granularity interactive response network based on feature decomposition, the obtained product front-end-product back-end decoupled interaction global representation vector will also fine-grainedly deconstruct the inconsistent feature granularity response and have a complex semantic interaction space structure. Therefore, when the product front-end-product back-end decoupled interaction global representation vector is input into the classifier-based customer churn prediction module to obtain the prediction result, it is expected to improve the classification regression convergence and generalization effect of the product front-end-product back-end decoupled interaction global representation vector under the complex semantic interaction space structure.

[0049] Therefore, when the product front-end-product back-end decoupled interaction global representation vector is input into a classifier-based customer churn prediction module to obtain a prediction result, the application optimizes the product front-end-product back-end decoupled interaction global representation vector, and the optimization process includes: calculating the absolute value sum of each eigenvalue of the product front-end-product back-end decoupled interaction global representation vector to obtain a first product front-end-product back-end decoupled interaction global representation vector space structure value, and calculating the square root of the sum of the squares of each eigenvalue of the product front-end-product back-end decoupled interaction global representation vector to obtain a second product front-end-product back-end decoupled interaction global representation vector space structure value: in, The first one represents the global representation vector of the product front-end-product back-end decoupling interaction eigenvalues, represents the global representation vector of the product front-end-product back-end decoupled interaction, represents the set of real numbers, represents the length of the global representation vector of the product front-end-product back-end decoupling interaction, It represents the global representation vector space structure value of the first product front-end-product back-end decoupling interaction, Indicates the second product front-end-product back-end decoupling interaction global representation vector space structure value; Multiply each eigenvalue of the product front-end-product back-end decoupling interaction global representation vector by the first product front-end-product back-end decoupling interaction global representation vector space structure value and the second product front-end-product back-end decoupling interaction global representation vector space structure value to obtain a first product front-end-product back-end decoupling interaction global representation vector structure reference value and a second product front-end-product back-end decoupling interaction global representation vector structure reference value corresponding to each eigenvalue; Multiply each eigenvalue of the product front-end-product back-end decoupling interaction global representation vector by the length of the product front-end-product back-end decoupling interaction global representation vector and the square root of the length to obtain a first product front-end-product back-end decoupling interaction global representation vector scale transformation value and a second product front-end-product back-end decoupling interaction global representation vector scale transformation value corresponding to each eigenvalue; Dividing the first product front-end-product back-end decoupling interaction global representation vector structure reference value by the difference between the first product front-end-product back-end decoupling interaction global representation vector space structure value and the first product front-end-product back-end decoupling interaction global representation vector scale transformation value to obtain the first product front-end-product back-end decoupling interaction global representation vector transformation adjustment value; Dividing the second product front-end-product back-end decoupling interaction global representation vector structure reference value by the difference between the second product front-end-product back-end decoupling interaction global representation vector space structure value and the second product front-end-product back-end decoupling interaction global representation vector scale transformation value to obtain the second product front-end-product back-end decoupling interaction global representation vector transformation adjustment value; The weighted sum of the first product front-end-product back-end decoupling interaction global representation vector transformation adjustment value and the second product front-end-product back-end decoupling interaction global representation vector transformation adjustment value is calculated to obtain each feature value of the optimized product front-end-product back-end decoupling interaction global representation vector. Finally, the optimized product front-end-product back-end decoupling interaction global representation vector is input into a classifier-based customer churn prediction module to obtain the prediction result.

[0050] Here, the optimized product front-end-product back-end decoupled interaction global representation vector It is expressed as: in, represents the second product front-end-product back-end decoupled interaction global representation transformation adjustment vector composed of the second product front-end-product back-end decoupled interaction global representation vector transformation adjustment value, represents the first product front-end-product back-end decoupled interaction global representation transformation adjustment vector composed of the first product front-end-product back-end decoupled interaction global representation vector transformation adjustment value, represents the weighted hyperparameter, It means adding by position. It means point multiplication by position. Represents the optimized product front-end-product back-end decoupled interaction global representation vector; ,and In the case of: in, It represents the global representation vector space structure value of the first product front-end-product back-end decoupling interaction, It represents the global representation vector space structure value of the second product front-end-product back-end decoupling interaction, The first one represents the global representation vector of the product front-end-product back-end decoupling interaction eigenvalues, represents the length of the global representation vector of the product front-end-product back-end decoupling interaction, Indicates the second product front-end-product back-end decoupling interaction global representation vector transformation adjustment value, It represents the first product front-end-product back-end decoupling interaction global representation vector transformation adjustment value.

[0051] That is, with respect to the spatial structure information of the feature set of the global representation vector of the product front-end-product back-end decoupled interaction in the high-dimensional space, a scale-based frame transformation of each eigenvalue of the global representation vector of the product front-end-product back-end decoupled interaction is performed by taking the class-norm space structured representation of the global representation vector of the product front-end-product back-end decoupled interaction as a reference window, and a spatial structure-based frame attention weight adjustment of each eigenvalue of the global representation vector of the product front-end-product back-end decoupled interaction is implemented to ensure the spatial transformation invariance of the global representation vector of the product front-end-product back-end decoupled interaction under the feature space interaction, thereby improving the convergence and generalization effects of the classification regression of the feature set of the global representation vector of the product front-end-product back-end decoupled interaction under the complex semantic space structure representation, and improving the accuracy of the prediction results obtained by the customer churn prediction module based on the classifier.

[0052] In summary, the above solution uses artificial intelligence technology based on deep learning to perform semantic analysis on customer transaction records, service requests, and product service feedback data, dig out the deep interaction pattern between customer transaction records, service requests, and product service feedback data, deeply understand customer behavior, and intelligently predict customer churn risk, and then formulate customer retention strategies for customers with high churn risk, and continue to track user subsequent behavior and dynamically adjust customer retention strategies. In this way, it can help companies accurately identify potential customer churn risks so that they can adopt corresponding strategies for customer retention, thereby improving customer satisfaction and corporate operating efficiency.

[0053] Figure 3 FIG. 1 is a block diagram of an enterprise intelligent management system based on deep learning according to an exemplary embodiment. Figure 3 As shown, the system 200 includes: The data collection module 201 of the analyzed customer object is used to collect the transaction records, service requests and product service feedback data of the analyzed customer object through a data collection tool; Data transmission module 202, used to transmit the transaction records, service requests and product service feedback data of the analyzed customer object to the central platform to obtain a customer behavior data set; A customer behavior pattern analysis module 203 is used to perform customer behavior pattern analysis on the customer behavior data set using deep learning technology on the central platform to obtain a prediction result of customer churn risk; A customer retention strategy formulation module 204, for formulating a customer retention strategy for customers with high churn risk according to the prediction results; The customer churn risk result prediction module 205 is used to continuously track the subsequent behavior data of the high churn risk customers, update the customer behavior data set, and regularly adjust and optimize the customer retention strategy based on the prediction results of the customer churn risk of the updated customer behavior data set.

[0054] Reference below Figure 4 , which shows a schematic diagram of the structure of an electronic device 600 suitable for implementing the embodiment of the present application. The terminal device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 4 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0055] like Figure 4 As shown, the electronic device 600 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the electronic device 600 are also stored. The processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0056] Typically, the following devices may be connected to the I / O interface 605: an input device 606 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 608 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 609. The communication device 609 may allow the electronic device 600 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 4The electronic device 600 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0057] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 609, or installed from the storage device 608, or installed from the ROM 602. When the computer program is executed by the processing device 601, the above-mentioned functions defined in the method of the embodiment of the present application are executed.

[0058] It should be noted that the computer-readable medium mentioned above in the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0059] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0060] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0061] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including, but not limited to, object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0062] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a sequence different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0063] The modules involved in the embodiments described in this application can be implemented by software or hardware. The name of the module does not limit the module itself in some cases. For example, the test parameter acquisition module can also be described as a "module for acquiring device test parameters corresponding to the target device".

[0064] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0065] In the context of the present application, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0066] Figure 5 FIG. 1 is an application scenario diagram of an enterprise intelligent management method based on deep learning according to an exemplary embodiment. Figure 5 As shown, in this application scenario, first, the transaction records of the analyzed customer object are collected through a data collection tool (for example, Figure 5 C1 as indicated in the figure), service requests (e.g. Figure 5 C2) and product service feedback data (e.g. Figure 5 Then, the acquired transaction records, service requests, and product service feedback data are input into a server that is deployed with an enterprise intelligent management algorithm based on deep learning (for example, Figure 5 In S), the server is capable of processing the transaction records, the service requests and the product service feedback data based on a deep learning-based enterprise intelligent management algorithm, so as to regularly adjust and optimize the customer retention strategy according to the prediction results of the customer churn risk of the updated customer behavior data set.

[0067] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in this application (but not limited to) by each other to form a technical solution.

[0068] In addition, although each operation is described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or to be performed in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be construed as limiting the scope of the application. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0069] Although the subject matter has been described in language specific to structural features and / or method logic actions, it should be understood that the subject matter defined is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely exemplary forms of implementation. With respect to the apparatus in the above-mentioned embodiments, the specific manner in which each module performs the operation has been described in detail in the embodiments related to the method, and will not be described in detail here.

Claims

1. A method for intelligent enterprise management based on deep learning, characterized in that: include: Collect transaction records, service requests, and product service feedback data of the analyzed customer objects through data collection tools; Transmitting the transaction records, service requests and product service feedback data of the analyzed customer object to the central platform to obtain a customer behavior data set; On the central platform, deep learning technology is used to analyze customer behavior patterns of the customer behavior data set to obtain prediction results of customer churn risk; Formulate customer retention strategies for customers with high churn risk based on the prediction results; Continue to track the subsequent behavior data of the high-churn risk customers, update the customer behavior data set, and regularly adjust and optimize the customer retention strategy based on the prediction results of the customer churn risk of the updated customer behavior data set.

2. The enterprise intelligent management method based on deep learning according to claim 1 is characterized in that: On the central platform, deep learning technology is used to analyze the customer behavior patterns of the customer behavior data set to obtain prediction results of customer churn risk, including: Separately extracting semantic features of service requests and transaction records in the customer behavior dataset to obtain a transaction record semantic encoding feature vector and a service request semantic encoding feature vector; Performing multimodal joint encoding on the image part and the text part in the product service feedback data in the customer behavior dataset to obtain a service feedback multimodal joint representation matrix; Inputting the transaction record semantic encoding feature vector and the service request semantic encoding feature vector into a fine-grained cross-modal saliency fusion module to obtain a product front-end multimodal fusion representation vector; Input the product front-end multimodal fusion representation vector and the service feedback multimodal joint representation matrix into a local granularity interactive response network based on eigendecomposition to obtain a product front-end-product back-end decoupled interaction global representation vector; The prediction result is generated based on the product front-end-product back-end decoupled interaction global representation vector, and the prediction result is used to represent the risk level label of customer churn.

3. The enterprise intelligent management method based on deep learning according to claim 2 is characterized in that: Separately extracting semantic features of service requests and transaction records in the customer behavior dataset to obtain transaction record semantic encoding feature vectors and service request semantic encoding feature vectors, including: Inputting the service request into a semantic expression expander based on the AIGC model to obtain a semantically expanded service request; The transaction record and the semantically expanded service request are semantically encoded respectively to obtain the transaction record semantic encoding feature vector and the service request semantic encoding feature vector.

4. The enterprise intelligent management method based on deep learning according to claim 3 is characterized in that: Performing multimodal joint encoding on the image part and the text part in the product service feedback data in the customer behavior dataset to obtain a service feedback multimodal joint representation matrix, including: Extracting an image portion and a text portion from the product service feedback data to obtain service feedback image data and service feedback text data; The service feedback image data and the service feedback text data are input into a CLIP model including a text encoder and an image encoder to obtain the service feedback multimodal joint representation matrix.

5. The enterprise intelligent management method based on deep learning according to claim 4 is characterized in that: Inputting the transaction record semantic encoding feature vector and the service request semantic encoding feature vector into a fine-grained cross-modal saliency fusion module to obtain a product front-end multimodal fusion representation vector, including: Calculating a global eigenvalue association matrix between the transaction record semantic encoding feature vector and the service request semantic encoding feature vector; Inputting the global eigenvalue association matrix into a learnable gating function to obtain a correlation distinguishable weight matrix; Using the correlation distinguishable weight matrix as a fusion mask matrix, respectively calculating the matrix products of the transaction record semantic coding feature vector, the service request semantic coding feature vector and the fusion mask matrix to obtain a distinguishing and enhancing transaction record semantic coding feature vector and a distinguishing and enhancing service request semantic coding feature vector; The distinguishing and strengthening transaction record semantic coding feature vector and the distinguishing and strengthening service request semantic coding feature vector are fused to obtain the product front-end multimodal fusion representation vector.

6. The enterprise intelligent management method based on deep learning according to claim 5 is characterized in that: Inputting the product front-end multimodal fusion representation vector and the service feedback multimodal joint representation matrix into a local granularity interactive response network based on feature decomposition to obtain a product front-end-product back-end decoupled interaction global representation vector, including: Performing feature decoupling on the product front-end multimodal fusion representation vector and the service feedback multimodal joint representation matrix to obtain a sequence of product front-end multimodal fusion local representation vectors and a sequence of service feedback multimodal joint local representation vectors; Calculate the local semantic association interaction vector between each corresponding product front-end multimodal fusion local representation vector and the service feedback multimodal joint local representation vector in the sequence of the product front-end multimodal fusion local representation vector and the sequence of the service feedback multimodal joint local representation vector to obtain a sequence of product front-end-product back-end local semantic association interaction vectors; Calculating the fine-grained attention weight of each product front-end-product back-end local semantic association interaction vector in the sequence of product front-end-product back-end local semantic association interaction vectors to obtain a sequence of fine-grained attention weights; Inputting the sequence of fine-grained attention weights into a random dropout module to obtain a sequence of fine-grained attention weights after dropout processing; Using the sequence of fine-grained attention weights after the deactivation process as the weight distribution, the weighted sum of the sequence of product front-end-product back-end local semantic association interaction vectors is calculated to obtain the product front-end-product back-end decoupled interaction global representation vector.

7. The enterprise intelligent management method based on deep learning according to claim 6 is characterized in that: Calculating the local semantic association interaction vector between each corresponding product front-end multimodal fusion local representation vector and the service feedback multimodal joint local representation vector in the sequence of the product front-end multimodal fusion local representation vector and the sequence of the service feedback multimodal joint local representation vector to obtain a sequence of product front-end-product back-end local semantic association interaction vectors, including: Calculate the position point addition, position point subtraction and position point multiplication between the product front-end multimodal fusion local representation vector and the corresponding service feedback multimodal joint local representation vector to obtain a first fusion result, a second fusion result and a third fusion result; After the first fusion result, the second fusion result and the third fusion result are cascaded into a product front-end-product back-end multi-scale fusion feature vector, one-dimensional convolution processing and maximum pooling processing are performed on them to obtain the product front-end-product back-end local semantic association interaction vector.

8. The enterprise intelligent management method based on deep learning according to claim 7 is characterized in that: Calculating the fine-grained attention weight of each product front-end-product back-end local semantic association interaction vector in the sequence of product front-end-product back-end local semantic association interaction vectors to obtain a sequence of fine-grained attention weights, including: The product front-end-product back-end local semantic association interaction vector is multiplied by a weight coefficient matrix, and then added to the bias vector and passed through a tanh function to obtain a product front-end-product back-end local semantic association interaction encoding feature vector; The attention score reference vector is multiplied by the product front-end-product back-end local semantic association interaction encoding feature vector to obtain the fine-grained attention weight.

9. The enterprise intelligent management method based on deep learning according to claim 8 is characterized in that: Based on the product front-end-product back-end decoupled interaction global representation vector, the prediction result is generated, and the prediction result is used to represent the risk level label of customer churn, including: The product front-end-product back-end decoupled interaction global representation vector is input into a classifier-based customer churn prediction module to obtain the prediction result.

10. An enterprise intelligent management system based on deep learning, characterized in that: include: The data collection module of the analyzed customer object is used to collect the transaction records, service requests and product service feedback data of the analyzed customer object through the data collection tool; A data transmission module, used to transmit the transaction records, service requests and product service feedback data of the analyzed customer object to the central platform to obtain a customer behavior data set; A customer behavior pattern analysis module is used to perform customer behavior pattern analysis on the customer behavior data set using deep learning technology on the central platform to obtain a prediction result of customer churn risk; A customer retention strategy formulation module, used to formulate a customer retention strategy for customers with high churn risk according to the prediction results; The customer churn risk result prediction module is used to continuously track the subsequent behavior data of the high churn risk customers, update the customer behavior data set, and regularly adjust and optimize the customer retention strategy based on the prediction results of the customer churn risk of the updated customer behavior data set.

Images