User tag extraction method, device and electronic equipment based on embedding vector

Abstract

The invention discloses a user label extraction method, device and electronic equipment based on embedding vectors. The method includes: obtaining embedding vectors of all user equipment data after being authorized by all users, and generating training data; using the training data and multiple A plurality of candidate clustering models corresponding to the number of candidate cluster centers are trained; the candidate clustering models are tuned according to the tuning data containing positive and negative samples to obtain the optimal clustering model; based on the optimal clustering The class model clusters users and extracts user labels. The optimal clustering model of the present invention can do a layer of semantic clustering on the embedding vector, thereby using the vector distance between users to fully mine the semantic similarity between users, clustering users, and ensuring the extraction of user The accuracy of the label can avoid misjudgment of user value, willingness and risk, especially the misjudgment of user risk, and improve business security.

Description

technical field [0001] The present invention relates to the technical field of computer information processing, in particular to a method, device, electronic equipment and computer-readable medium for extracting user tags based on embedding vectors. Background technique [0002] With the advent of the era of Internet big data, Internet companies hope to dig out highly accurate, high-coverage, and multi-dimensional label information from massive data to describe users, so as to distinguish the value, willingness, and risks of potential users. [0003] In the prior art, user data is converted into embedding vectors and input into machine learning models (such as marketing models, risk models, etc.), and user descriptions are obtained through these models. However, these models use features such as embedding vectors directly, and cannot make full use of the rich information hidden in the embedding vectors to mine the semantic similarity between users, which affects the accuracy...

AI Technical Summary

Problems solved by technology

However, these models use features such as embedding vectors directly, and cannot make full use of the rich information hidden in the embedding vectors to mine the semantic similarity between users, which affects the accuracy of user characterization, resulting in the loss of user value and willingness. Misjudgment of risk and risk, especially misjudgment of user risk, will cause hidden dangers to business security

Images