CN-121981780-A - Multi-language user data integration marketing method based on natural language processing

CN121981780ACN 121981780 ACN121981780 ACN 121981780ACN-121981780-A

Abstract

The invention discloses a multi-language user data integration marketing method based on natural language processing, which relates to the technical field of natural language processing and comprises the steps of carrying out entity recognition and relation extraction on multi-language user data through natural language processing, constructing a multi-language enterprise knowledge graph, carrying out representation learning on the multi-language enterprise knowledge graph through a multi-scale integral graph attention network, constructing a potential association strength matrix, mining the potential association strength matrix, identifying a community structure, calculating steady-state influence values of nodes based on the potential association strength matrix and the community structure, screening target client groups through an influence analysis algorithm, and generating a context feature vector of a target client. The invention realizes the transition of the quality of multi-language enterprise data from shallow information association to deep potential relation mining, and improves the prospective and accuracy of identifying the target customer group.

Inventors

YU LI
ZHAO HONGBIN
ZHANG BAO

Assignees

福禄娃(丽水)大数据管理有限公司

Dates

Publication Date: 20260505
Application Date: 20260105

Claims (10)

1.A multi-language user data integration marketing method based on natural language processing is characterized by comprising the following steps of, Performing entity identification and relation extraction on the multilingual user data through natural language processing, and constructing a multilingual enterprise knowledge graph; Performing representation learning on the multi-language enterprise knowledge graph through a multi-scale integral graph attention network, constructing a potential association strength matrix, mining the potential association strength matrix, and identifying a community structure; Calculating a steady-state influence value of a node based on the potential association strength matrix and the community structure, screening a target client group through an influence analysis algorithm, and generating a context feature vector of a target client; inputting the context feature vector of the target client into a generated type countermeasure network, outputting an evaluation score, and performing multi-language grammar checking and style optimization to generate a personalized marketing content package; and sending the personalized marketing content to the corresponding target client, and simultaneously collecting feedback data of the target client and updating the multilingual enterprise knowledge graph.
2. The method for integrated marketing of multilingual user data based on natural language processing according to claim 1, wherein the steps of performing entity recognition and relationship extraction on the multilingual user data by natural language processing and constructing a multilingual enterprise knowledge graph are as follows, Cleaning and standardizing the multilingual user data to generate a standardized multilingual text data set; performing entity identification operation on the normalized multilingual text data set, and extracting enterprise, product, personnel and place entities; Identifying provisioning, collaboration, and competition relationships between entities from the normalized multilingual text data; Taking enterprises, products, personnel and place entities as nodes, taking supply, cooperation and competition relations as edges, and constructing an initial multilingual enterprise knowledge graph; and performing entity alignment and disambiguation on the initial multi-language enterprise knowledge graph to generate a disambiguated multi-language enterprise knowledge graph.
3. The method for integrated marketing of multilingual user data based on natural language processing according to claim 2, wherein the method comprises the steps of performing representation learning on a multilingual enterprise knowledge graph through a multiscale integrated graph attention network, constructing a potential correlation strength matrix, Preparing graph data of the disambiguated multi-language enterprise knowledge graph, extracting multi-language feature vectors from the nodes, and generating a node feature vector set; Inputting the node characteristic vector set into a multiscale integral graph attention network, and fusing multiscale neighbor information through integral transformation to generate a final representation vector of the node; Carrying out standardization processing on the final representation vector to obtain a standardized node representation vector; Potential correlation strength scalar values between nodes are calculated from the normalized node representation vectors and a potential correlation strength matrix is constructed.
4. The method for integrated marketing of multilingual user data based on natural language processing of claim 3, wherein the mining of the potential correlation strength matrix, the identification of the community structure, comprises the following steps, Carrying out symmetry and normalization treatment on the potential association intensity matrix to obtain a standardized association matrix; and identifying a community structure of the multilingual enterprise knowledge graph in the standardized incidence matrix.
5. The method for integrated marketing of multilingual user data based on natural language processing of claim 4, wherein the calculating the steady-state influence value of the node based on the potential correlation strength matrix and the community structure comprises the following steps, Identifying an optimal community division scheme of a community structure through a graph neural network, and acquiring a community division result matrix of a node-community membership; extracting multi-modal influence features of each node from the potential association strength matrix and the community division result matrix to generate a multi-modal feature vector set; And dynamically evolving the multi-mode feature vector set, and calculating the steady-state influence value of the node.
6. The method for integrated marketing of multilingual user data based on natural language processing of claim 5, wherein the method comprises the steps of screening a target customer group by an influence analysis algorithm to generate a context feature vector of the target customer, Calculating the frequency domain influence score of the node according to the steady state influence value scalar; and screening the target customer group by adopting an influence analysis algorithm based on the frequency domain influence score, and acquiring the context feature vector of the target customer.
7. The method for integrated marketing of multilingual user data based on natural language processing of claim 6, wherein the step of inputting the contextual feature vector of the target client into the generated countermeasure network to output the evaluation score comprises the steps of, Performing feature cross enhancement processing on the context feature vector of the target client to generate an enhanced context feature matrix vector; inputting the enhanced context feature matrix into a content generator of a generated type countermeasure network, performing feature vector decoding, conditional text sequence generation and countermeasure quality optimization, and outputting an initial marketing content text; and inputting the initial marketing content text into a content discriminator of the generated type countermeasure network, carrying out three-dimensional evaluation of semantic consistency, cultural adaptability and marketing effect, and outputting evaluation scores.
8. The method for integrated marketing of multilingual user data based on natural language processing of claim 7, wherein the steps of performing multilingual grammar examination and style optimization to generate personalized marketing content packages are as follows, According to the evaluation score, performing back propagation and updating network parameters of the content generator to generate optimized marketing content; performing multilingual language law examination and style optimization on the optimized marketing content to generate a compliance marketing content text conforming to the language habit of the target client; Binding the compliance marketing content text with the corresponding target client information to generate a personalized marketing content package.
9. The method for integrated marketing of multilingual user data based on natural language processing of claim 8, wherein the step of transmitting personalized marketing content to a corresponding target client while collecting feedback data of the target client comprises the steps of, Distributing personalized marketing content packages to a plurality of communication channels, and generating a sending state matrix and a customer contact point log; Based on the sending state matrix and the customer contact point log, the multi-mode feedback data are collected in real time and are subjected to fusion processing, and a fusion feedback data stream is generated.
10. The method for integrated marketing of multilingual user data based on natural language processing of claim 9, wherein the updating of the knowledge graph of multilingual enterprises comprises the following steps, Inputting the fused feedback data stream into a point reinforcement learning decision frame, calculating a feedback value score through a point reward function, and generating a map updating decision; and carrying out incremental updating and consistency verification on the multilingual enterprise knowledge graph according to the graph updating decision.

Description

Multi-language user data integration marketing method based on natural language processing Technical Field The invention relates to the technical field of natural language processing, in particular to a multilingual user data integration marketing method based on natural language processing. Background Along with the acceleration of globalization process and the deep digital transformation of enterprises, multilingual user data integration and accurate marketing technology have become important research directions in the field of business intelligence. The prior art mainly relies on two main core directions of Natural Language Processing (NLP) and Graph Neural Network (GNN), namely, on one hand, based on entity recognition and relation extraction technologies of pretrained models such as multilingual BERT and XLM-R, structured knowledge can be extracted from heterogeneous multilingual data to initially construct an enterprise knowledge graph, on the other hand, the knowledge graph is expressed and learned through a graph attention network (GAT) or a graph rolling network (GCN), dominant association relations among enterprises are mined, and community discovery algorithms (such as Louvain or label propagation algorithms) are utilized to identify group structures. Despite the stepwise progress made in the prior art, there are two major core limitations. Firstly, there are significant shortcomings in depth association mining and dynamic updating of multi-language knowledge maps. The existing method is dependent on a static graph structure, potential association strength among enterprises is difficult to quantify through a multi-scale integral attention mechanism, and incremental learning capacity based on feedback data is lacking, so that graph updating is delayed from market dynamic change. Secondly, there is a technical dislocation in the generation and optimization links of personalized marketing content. The existing generation model often operates independently, is not deeply coupled with structural information (such as community influence and cultural background) of the knowledge graph, and lacks a fine evaluation mechanism for multilingual language law compliance and cultural adaptability, so that the generated content lacks context correlation and cross-cultural persuasion. Disclosure of Invention The present invention has been made in view of the above-described problems occurring in the prior art. Therefore, the invention provides a multi-language user data integration marketing method based on natural language processing, which solves the problems of knowledge graph staticization and content generation dislocation in the prior art. In order to solve the technical problems, the invention provides the following technical scheme: The invention provides a multi-language user data integration marketing method based on natural language processing, which comprises the steps of carrying out entity recognition and relation extraction on multi-language user data through natural language processing, constructing a multi-language enterprise knowledge graph, carrying out representation learning on the multi-language enterprise knowledge graph through a multi-scale integration graph attention network, constructing a potential correlation strength matrix, mining the potential correlation strength matrix, identifying a community structure, calculating a steady-state influence value of a node based on the potential correlation strength matrix and the community structure, screening a target client group through an influence analysis algorithm, generating a context feature vector of a target client, inputting the context feature vector of the target client into a generation type countermeasure network, outputting an evaluation score, carrying out multi-language grammar checking and style optimization, generating a personalized marketing content package, sending personalized marketing content to a corresponding target client, and simultaneously collecting feedback data of the target client and updating the multi-language enterprise knowledge graph. As an optimal scheme of the multi-language user data integration marketing method based on natural language processing, the method comprises the steps of carrying out entity identification and relation extraction on the multi-language user data through natural language processing, constructing a multi-language enterprise knowledge graph, specifically comprising the following steps of, Cleaning and standardizing the multilingual user data to generate a standardized multilingual text data set; performing entity identification operation on the normalized multilingual text data set, and extracting enterprise, product, personnel and place entities; Identifying provisioning, collaboration, and competition relationships between entities from the normalized multilingual text data; Taking enterprises, products, personnel and place entities as nodes, taking supply, cooperation and com