A Machine Learning Approach for Ranking in Question Answering

Question Answering is a challenging topic and is gaining growing attention in last years being an interesting interdisciplinary research area and having practical application. In this paper we focus on the answer selection, a step of Question Answers that selects answers to the questions from among the answer candidates based on the result of question analysis. This process can be very challenging, as it often entails identifying correct answers amongst many incorrect ones. In particular, we focus on the ranking of the answers based on Italian language and referring to a dataset that is closed-domain, containing questions about cultural heritage with successive True or false answers. In this paper we demonstrate that, using an approach based on classification, we can reach a very high accuracy that is better than the accuracy reached with other approaches based on fuzzy logic.