Personalized multi-document summarization in information retrieval

This research is directed towards automating open-domain multi-document summarization in the framework of Web search. We present a novel approach to achieve this object. Given an unrestricted user query, our system retrieves documents related to and summarizes them. In the process of summarization, the sentences in a given document are scored based on the relevant value and the informativeness value, which are realized by using word overlap and semantic graph. Then, the sentences with highest scores are incorporated into the output summary together with their structural context. Experimental results show that our query-topic focused summary could return a topically relevant extractive summary. And the summarization quality is relatively competitive.