Improving Keyphrase Extraction Using Wikipedia Semantics

Keyphrase extraction plays a key role in various fields such as information retrieval, text classification etc. However, most traditional keyphrase extraction methods relies on word frequency and position instead of document inherent semantic information, often results in inaccurate output. In this paper, we propose a novel automatic keyphrase extraction algorithm using semantic features mined from online Wikipedia. This algorithm first identifies candidate keyphrases based on lexical methods, and then a semantic graph which connects candidate keyphrases with document topics is constructed. Afterwards, a link analysis algorithm is applied to assign semantic feature weight to the candidate keyphrases. Finally, several statistical and semantic features are assembled by a regression model to predict the quality of candidates. Encouraging results are achieved in our experiments which show the effectiveness of our method.