Neural Network Accelerated Tuple Search For Relational Data

In this paper, we study the relational data search problem in which users can issue semi-structured keyword queries to search through a database of multiple relations. Traditional text indexes suffer from performance issues when we allow search queries with fuzzy text matching. Using NLP inspired neural networks, we are able to overcome the performance issues by a predictive index access optimization. The neural network that performs the query optimization is learned in an unsupervised fashion. As we show, the text corpus itself is used as the training data for self-learning query optimization networks. Our experimental evaluation shows that the predictive query optimizer is effective, and can yield over 40% performance boost.