Reviews Analysis Based on Sentence and Word Relevance

Assuming that one sentence in review expresses one opinion, LDA based on sentence is performed to analysis the massive online reviews. When computing the topic´s word, word relevance measure is designed which penalizes the word frequency by a factor that captures how much the word is shared across topics, words for topics can been selected more accurately. Experiments on massive review crawled from network show that the result of analyzing is better than the standard LDA, there is clearer topic cue, and recognition is improved among the topics.