Semantic Networks and Spreading Activation Process for QA improvement on text answers

Abstract

Question Answering (QA) systems try to find precise answers to natural language questions. QA extraction result is often an amount of text candidate answers which requires some validation and ranking criteria. This paper presents an automatic answer appreciation technique where extracted candidate answers are represented in a question dedicated associative knowledge base, a semantic network. A spreading activation algorithm looks for semantically related candidate answers, that reinforce each other. The purpose is to enhance the best answers by rising their weight. This article concludes with evaluation details for an experiment with text answers to Portuguese questions.