PolyU's Experimentation with Guided Summarization

In addition to the familiar tasks of update summarization and automatic summary evaluation, the TAC 2010 summarization track includes a new task: guided summarization. It is designed to “encourage a deeper linguistic (semantic) analysis of the source documents” in order to generate a summary guided by a given set of aspects. The new task is an upgrade of query-focused summarization in that an automatic summary is to meet more specific and semantically structured user need. The upgrade demonstrates the growing inclination for semantically oriented summarization, whilst posing new challenges to traditional frameworks and techniques. The update summarization and the automatically evaluating summaries of peers (AESOP) tasks are similar to those included in the TAC 2009 summarization track. The PolyU team has participated in both tasks of the TAC 2010 summarization track, including the guided summarization task and the AESOP. The guided summarization task is an unfamiliar one, with few success or failure stories. Our team takes up this new challenge, not widely studied in the summarization community, with techniques based on Information Extraction (IE). For the AESOP task, we also experiment with some new techniques. In the following, we report the system design for the tasks as well as findings and diagnoses from our experimentation.