Node relationship modeling method based on word vector model

Abstract

The invention discloses a node relationship modeling method based on a word vector model. The node relationship modeling method comprises the following steps of: a data preprocessing stage: pulling user flow data of a certain number of days, generating a user data sequence by taking a user as a carrier, and sorting according to time to obtain a node sequence based on the user; a vector training stage: carrying out word vector training through word2vec according to the node sequence with the user as the carrier, and obtaining vector representation of a single node through training; a similar node clustering stage: on the basis of the word vectors obtained through training, conducting similar entry clustering, and obtaining clustering result clusters of similar nodes; and a label reasoning and transmitting stage: transmitting and reasoning the labels in the clustering result cluster. Based on the word vector method, the node relationship modeling method establishes an intra-cluster structure chart and gives the edge weight information according to the posterior behavior of the user and the interaction relationship between the articles, considers the association information between the individuals, and performs label reasoning transmission through comprehensive judgment, and the labels of the articles are more accurate in combination with the group information.

Description

technical field [0001] The invention relates to word vector technology in the field of natural language processing, in particular to a node relationship modeling method based on a word vector model. Background technique [0002] In recent years, with the rapid development of the Internet, the interaction between people and data has become more and more frequent. With the popularity of e-commerce platforms and news websites. At the same time, people have also formed a great dependence on mobile phones, generating hundreds of millions of flow data on various application software on mobile phones, such as purchasing behaviors, such as news browsing behaviors. In terms of purchase behavior, the number and types of items are becoming more and more diverse, and the items also have more and more detailed labels. In browsing behavior, all kinds of news also have categories and numerous tags. High-quality labeling and classification can help improve business efficiency and optimize...

AI Technical Summary

Problems solved by technology

However, for the division of personality tags, there are often strong group characteristics, and the traditional feature-based classification model fails to fully consider the impact of potential connections between user groups on the interaction data category.

Images