TREC-10 Experiments at KAIST: Batch Filtering and Question Answering

1.Introduction In TREC-10, we participated in two tasks: batch filtering task in the filtering task, and question answering task. In question answering task, we participated in three sub-tasks (main task, list task, and context task). In batching filtering task, we experimented a filtering technique, which unifies the results of support vector machines for subtopics subdivided by incremental clustering. For a topic, we generated subtopics by detecting similar documents in training relevant documents, and unified the results of SVM classifier for subtopics by OR set operation. In question answering task, we submitted two runs for main task (KAISTQAMAIN1, KAISTQAMAIN2), two runs for list task (KAISTQALIST1, KAISTQALIST2), and one run for context task (KAISTQACTX).