Vector-based Ranking Techniques for Identifying the Topical Anchors of a Context

Lexical cooccurrence in textual data is not uniformly random. The statistics inferred from the termcooccurrence data enable us to model dependencies between terms as graphs, somewhat resembling the way semantic memory is organised in human beings. In this paper we look at cooccurrence patterns to identify topical anchors of a given context. A context is defined as a collection of terms and topical anchors are those terms whose semantics represent the topic of the whole context. This work is based on computing a stationary distribution in the cooccurrence graph. Different vector-based techniques are proposed to compute and threshold the stationary distribution. Experiments were conducted on 100 distinct human defined contexts. The topical anchors generated were evaluated against those chosen manually by 86 volunteers and the results show that the algorithm correctly identifies the topical anchors 70% of the time.