TFIDF based Feature Words Extraction and Topic Modeling for Short Text

In this paper, feature words extraction and topic modeling based on Term Frequency times In-verse Document Frequency (TFIDF) and Latent Dirichlet Allocation (LDA) is achieved aiming at short titles text of The National Institutes of Health (NIH) supported research. After preprocess of raw text, distinct terms extracted from titles of NIH supported research compose the Bag of Words. And then TFIDF is used in order to re-weight the count features for reducing the influence of more frequent yet less valuable terms and enhancing the influence of rarer yet more valuable terms. Via topic modeling with Latent Dirichlet Allocation, ten topics and corresponding three feature words are extracted, and each topic is characterized by the three special words. As a result, it can achieve topic extraction of NIH supported research according to titles and reveal the most probable research area.