Query Focused Language Independent Multi-document Summarization

A query focused multi-document automatic summarizer has been described. The system clusters similar texts in multiple related documents having related (sub) topical features. A document graph is constructed, where nodes are sentences and edge scores reflect correlation measure between nodes. Then clusters are constructed from the graph. Each cluster gets a weight and has a cluster center. Next, query dependent weights for each sentence are added to the edge score as well as to the cluster score. Top ranked sentence of each cluster in order is identified for inclusion in the output summary. It was tested on the standard TAC (formerly DUC) 2008 data sets of the Update Summarization Track and evaluated by ROUGE 1.5.5 where ROUGE-2 and ROUGE-SU-4 scores of 0.103 and 0.14 have been obtained. Then the experiments carried out at Jadavpur University as part of the participation in FIRE 2010 in the ad-hoc mono-lingual information retrieval task for English and Bengali languages, has been described. The experiments are based on stemming, zonal indexing, theme identification, TF-IDF based ranking model and positional information. Each query was specified using title, narration and description