Two Different Summarization Methods at NTCIR3-TSC2: Coverage Oriented and Focus Oriented

We are conducting research on multi-document summarization, participating in a competition of sum-marization, TSC (Text Summarization Challenge) task organized by NTCIR-3 project. In a dry run, we conceived a new extraction method for multi-document summarization which extracts a set of sentences that maximizes coverage of an original text and minimizes redundancy of a summary. Thinking over the result of the dry run, we decided to build another system for the formal run which generates a more focused summary. It employs a headline sentence and similarity of sentences to grasp the major points of original articles. In addition to them, we consider sentence ordering and reduction. We compare these summaries to discuss effectiveness of each method.