Desing, Development and Benchmarking of Algorithms for Conversational Search

Developing an intelligent dialog system that not only emulates human conversation, but also answers to difficult topics is one of the most important fields on several research area. In recent years, great strides have been made in this area and several companies and research groups create competitions that aim to find solutions to problems like Conversational Information Seeking and Natural Language Generation. On thisworkwe see one of them in particular: TREC CaST (Conversational Assistance Track). We analyze several techniques that allowto create a conversational system and how we can improve the results by using neural techniques. On this work we examine how to retrieve relevant documents by using Lucene and then to re-rank this documents by using a neural text-classifier like BERT.