Topic segmentation using weighted lexical links (WLL)

Laurianne Sitbon, Patrice Bellot

Abstract: This paper presents two new approaches of lexical chains for topic segmentation using weighted lexical chains (WLC) or weighted lexical links (WLL) between repeated occurrences of lemmas along the text. The main advantage of using these new approaches is the suppression of the empirical parameter called hiatus in lexical chain processing. An evaluation according to the WindowDiff measure on a large automatically built corpus shows slight improvements in WLL compared to state-of-the-art methods based on lexical chains.