University of Lethbridge’s Participation in DUC-2007 Main Task

This paper describes a query-focused multi-document summarizer based on two distinct but complementary concepts: a) how much the sentence is related to the user query and b) how much the sentence is salient to the overall concept. Keeping these in focus we consider 6 important features: (1)Cosine Similarity (2)Lexical chain (3)BE overlaps (4)Question Focus overlap (5) Previous Sentence overlaps and (6)Document overlap. We consider Cosine Similarity measure, for computing sentence importance based on the concept of eigenvector centrality in a graph representation of sentences (Erkan and Radev, 2004). Lexical chains efficiently identify the theme of the document. An additional argument for the chain representation to consider as opposed to a simple word frequency model is the case when a single concept is represented by a number of words, each with relatively low frequency. Because the chain combines the number of occurrences of all its members, it can overcome the weight of the single word (Chali and Kolla, 2004). With BE 2 represented as a head-modifier-relation triple, one can quite easily decide whether any two units match (express the same meaning) or not–considerably more easily than with longer units (Hovy et al., 2005). We consider Question Focus Overlap feature to extract the sentences, which are relevant to the topic and narration. We consider the other two features: Previous Sentence overlaps and Document overlap in order to increase the coherence among the sentences in the summary.