Chapter seven - Deep learning on information retrieval and its applications

Abstract

In the domain of information retrieval (IR), the matching of query and document relies on ranking models to calculate the degree of their relevance. Therefore, ranking models remain as the central component of the research. During the past decades, there has been a trend moving from traditional approaches to IR toward deep learning approaches to IR. Traditional IR models include basic handcrafted retrieval models, semantic-based models, term dependency-based models, and learning to rank models. The deep learning approaches, on the other hand, involve methods of representation learning, methods of matching function learning, and methods of relevance learning. Recently, we have seen a growing number of publications in both conferences and journals using deep learning techniques to solve the IR problems. The capability of neural ranking models to extract features directly from raw text inputs overcomes many limitations of traditional IR models that rely on handcrafted features. Moreover, the deep learning methods manage to capture complicated matching patterns for document ranking. In this chapter, we introduce a novel way of classifying these existing IR models, along with their recent improvements and developments. To the best of our knowledge, our approach is the first one to classify the existing work according to how they generate the features and the ranking functions. Moreover, we provide a review of these proposed models to discuss different dimensions and to make empirical comparisons, followed by a conclusion with possible directions of future work.