Learning to Match Using Local and Distributed Representations of Text for Web Search

Models such as latent semantic analysis and those based on neural embeddings distributed representations of text, and match the query against the in the latent semantic space. In traditional information retrieval, on the other hand, terms have discrete or local representations, and relevance of a document is determined by the exact matches of query terms the body text. We hypothesize that matching with distributed representations matching with traditional local representations, and that a of the two is favorable. We propose a novel document ranking model of two separate deep neural networks, one that matches the query and document using a local representation, and another that matches the query the document using learned distributed representations. The two networks jointly trained as part of a single neural network. We show that this or `duet' performs significantly better than either neural network on a Web page ranking task, and also significantly outperforms baselines and other recently proposed models based on neural.