Semantic search has emerged as a possible way for addressing the challenges of traditional keyword-based retrieval systems such as the vocabulary gap between the query and document spaces. In this paper, we propose a novel semantic retrieval framework that uses semantic entity linking systems for forming a graph representation of documents and queries, where nodes represent concepts extracted from documents and edges represent semantic relatedness between those concepts. The core of our proposed work is a semantic-enabled language model that estimates the probability of generating query concepts given values assigned to document concepts. The semantic retrieval framework also provides basis for interpolating keyword-based retrieval systems with the semantic-enabled language model. We conduct comprehensive experiments over several Trec document collections and ana- lyze the performance of different configurations of the framework across multiple retrieval measures. Our experimental results show that the proposed semantic retrieval model has a synergistic impact on the results obtained through the state-of-the-art keyword-based systems, and the consideration of semantic information can complement and enhance the performance of such retrieval models.

Keywords