Finding titles representing segments of Wikipedia Articles from keyphrases

Wikipedia is a free online encyclopedia that aims to allow anyone to edit any article or create them. However, articles tend to become long and complex, so giving appropriate titles or key phrases to untitled segments is necessary for reader assistance. In this paper, we show methods to select titles for representing article segments. Key phrase extraction has been studied for years, but we concentrate on selecting a title phrase for a given target segment from candidate phrases, which needs to reflect local and global contexts. In this paper, we evaluate five features we proposed before, and one new feature which is based on word embedding. These features are combined to produce a ranked list of candidate titles. We construct over a candidate title set consisting of titles of articles, sections and subsections, and anchor texts of inner links (inter-article links) where the hidden title of the target segment is the ground truth. We compare performance of various feature combinations by precision@K, reciprocal rank and average precision. Keyword: Wikipedia, Finding Representative Title, Candidate Ranking