A Phrase Topic Model for Large-scale Corpus

Abstract

The topic model is an unsupervised learning model, one of the important tools for large-scale corpus analysis, widely used in information retrieval, natural language processing, and machine learning. Traditional topic models, such as Latent Dirichlet Allocation (LDA), ignore the order of words. However, in many text-mining tasks, word order and phrases are often crucial for capturing the meaning of texts efficiently. We propose a phrase topic model based on the LDA model, which integrates a regular expression constraint condition. Our model makes the topic more meaningful and interpretable based on a limited increase in the dimensions of the vocabulary. The experimental results show that our algorithm can find meaningful phrases and have generic applicability in our test data set.