Research on User Portrait Based on Tag Word Embedding

User portrait technology can bring huge commercial value to enterprises. Based on movie recommendation, this article extracts user tags from movie information based on the movie's viewing history, then designs and constructs user vectors, uses K-means to cluster users, and uses the clustering results to train the classifier. This article uses the traditional machine learning method Naive Bayes, BP neural network and deep learning method BiLSTM are used as the classifier. The construction of user vector adopts three methods, namely constructing user vector only by word vector, constructing user vector only by TF-IDF, and combining word vector and TD-IDF to construct user vector, and designed six sets of experiments for comparison. Experimental results show that the method of combining word vector and TF-IDF has achieved good results overall, and can further improve the quality of the recommendation system.