Experiences from Developing the Domain-Specific Entity Search Engine GeneView

.GeneView is a semantic search engine for the Life Sciences. Unlike traditional search engines, GeneView searches indexed documents not only at the textual (syntactic) level, but analyzes texts upon import to recognize and properly handle biomedical entities, relationships between those entities, and the structure of documents. This allows for a number of advanced features required to work effectively with scientific texts, such as precise search despite large numbers of synonyms and homonyms, entity disambiguation, ranking of documents by entity content, linking to structured knowledge about entities, user-friendly highlighting of recognized entities etc. As of now, GeneView indexes approximately ~21,4 million abstracts and ~358.000 full texts with more than 200 Million entities of 11 different types and more than 100,000 relationships of three different types. In this paper, we describe the architecture underlying the system with a focus on the complex pipeline of advanced NLP and information extraction tools necessary for achieving the above functionality. We also discuss open challenges in developing and maintaining a semantic search engine over a large (though not web-scale) corpus.