Subtopic Mining via Modifier Graph Clustering

Haitao Yu, Fuji Ren

Abstract: Understanding the information need encoded in a user query has long been regarded as a crucial step of effective information retrieval. In this paper, we focus on subtopic mining that aims at generating a ranked list of subtopic strings for a given topic. We propose the modifier graph based approach, under which the problem of subtopic mining reduces to that of graph clustering over the modifier graph. Compared with the existing methods, the experimental results show that our modifier-graph based approaches are robust to the sparseness problem. In particular, our approaches that perform subtopic mining at a fine-grained term-level outperform the baseline methods that perform subtopic mining at a whole query-level in terms of I-rec, D-nDCG and D#-nDCG.

External IDs:dblp:conf/pakdd/YuR14