Opinion Question Answering: Towards a Unified Approach

Nowadays, the Web contains large amounts of heterogeneous (factual and opinionated) data, which is becoming equally important for users to access. The need to efficiently manage this information leads to the necessity of building automatic systems that efficiently process it. In this paper, we propose and evaluate a series of techniques whose aim is to improve the performance of an Opinion Question Answering (OQA) system. We include additional resources and processes with the objective of limiting the sources of errors in the different stages involved - question analysis, answer retrieval and filtering, answer re-ranking. We propose new elements that are significant in these stages and show that their use improves the performance of the system. We conclude that the suggested techniques help to influences the task in a positive manner.