This paper describes the system submitted to the TREC 2016 LiveQA track. This year, the TREC 2016 LiveQA track consists of implementing a web service to answer user-submitted questions. The newest unanswered question from Yahoo! Answers will be posted to the web service, a question every minute for 24 hours. The implementation described in this paper uses natural language processing (NLP) to extract keywords from the question given as input. A web search together with a Yahoo! Answer search is

more »

... sed to select candidate answers. A latent dirichlet allocation (LDA) model is trained in order to compute a topic distribution of the different candidate answers. Finally, the Jensen-Shannon distance is used as similarity measure between the candidate answers and the question given as input. This implementation performed better than the average scores.