Guided Summarization with Aspect Recognition

As a continuation of the summarization track of TAC 2010, the TAC 2011 summarization track pursues aspect-guided summarization. It is intended “to encourage a deeper linguistic (semantic) analysis of the source documents instead of relying only on document word frequencies to select important concepts”. The sustained interest in guided summarization marks a significant turn to semantically oriented end results that implicitly favor deep (semantically rich) NLP, domain-specific IE, NLG, among other techniques. The update summarization task is similar to that included in earlier TAC summarization tracks. The PolyCom (formerly PolyU) team has participated in the guided summarization task of both TAC 2010 and TAC 2011. Lacking experience with this new task, our team did not perform well with an IE-based system in 2010. This year, we draw on our learned lesson and resources built over months to build a system based on aspect recognition and a robust baseline. We achieve very competitive evaluation results released by NIST. In the following, we report the system design for our new system, including aspect recognition and aspect-guided summarization.