The Italica System at TAC 2008 Opinion Summarization Task
Fermín L. Cruz, José A. Troyano, F. Javier Ortega, Fernando Enríquez · Theory and applications of categories · 2008
texts. Second, the goal is not to obtain asummary of the complete documents, but the answersgiven in the documents to several questions. As inputto the system, apart from the documents themselves,the participants are provided with the questions andalso the snippets that answer those questions, beingthe latter generated by the QA systems participatingin another TAC 2008 task.Oursystemisbasedonthecombinationofthesnip-pets provided for the summary construction. In orderto make the text more readable and complete the in-formation, the process starts looking for the most re-levant sentences in relation to the snippets, which arethen used to generate the text that will flnally com-pose the summary. Our system is therefore focusedon the sentence extraction phase. We have obtainedgood results with the pyramid F-score and overall res-ponsiveness measures, achieving the second and flrstplace respectively among the participating systems.In the following sections we describe the architec-ture of the system in flrst place, and continue dis-cussing the results obtained in the evaluation process.Finally, we conclude discussing the strong and weakpoints of our system.