A Natural Language and Interactive End-to-End Querying and Reporting System

Natural language query understanding for unstructured textual sources has seen significant progress over the last couple of decades. For structured data, while the ecosystem has evolved with regard to data storage and retrieval mechanisms, the query language has remained predominantly SQL (or SQL-like). Towards making the latter more natural there has been recent research emphasis on Natural Language Interface to DataBases (NLIDB) systems. Piggybacking on the rise of 'deep learning' systems, the state-of-the-art NLIDB solutions over large parallel and standard benchmarks (viz, WikiSQL and Spider) primarily rely on attention based sequence-to-sequence models. Building industry grade NLIDB solutions for making big data ecosystem accessible by truly natural and unstructured querying mechanism presents several challenges. These include lack of availability of parallel corpora, diversity in underlying data schema, wide variability in the nature of queries to context and dialog management in interactive systems. In this paper, we present an end-to-end system Query Enterprise Data (QED) towards making enterprise descriptive analytics and reporting easier and natural. We elaborate in detail how we addressed the challenges mentioned above and novel features such as handling incomplete queries in incremental fashion as well as highlight the role of an assistive user interface that provides a better user experience. Finally, we conclude the paper with observations and lessons learnt from the experience of transferring and deploying a research solution to industry grade practical deployment.