Information Retrieval techniques for Question Answering

Abstract

Throughout History our ability to imprint Information and to access it, made science progress. First Information was made accessible through books and libraries, till Internet made an appearance and knowledge was made more accessible that ever. Now we are facing another big problem that if we let it grow it might even equalize the lack of knowledge. Ely et al. [1] have found that physicians spend an average of 2 min or less in seeking an answer, while Hersh et al. [2] have found that it takes more than 30 min on average for a health care professional to search for an answer. As a result many clinical questions go unanswered. “Better be ignorant of a matter than half know it.” Publilius Syrus said. So we know as a fact that efficient searching has gone far from the reach of plain human abilities. Thus, we need Question Answering (QA) Systems, which eventually will do the hard work of answering in a matter of few minutes. Question Answering is a multi-layered problem. In this work we will try to analyze the potential of re-ranking results retrieved from Information Retrieval techniques. More specifically the re-ranking is performed by machine learning models using word embedding that try to take into account the context of the sentences.