An unsupervised method for attribute identification from a natural language query

Abstract

Identifying which attributes the user querying is an important step in providing a Natural language (NL) search interface to a relational database. In this paper, we discuss an unsupervised approach for identifying the target database attributes from natural language (NL) query. This is a knowledge base-driven method which can be adopted into any domain with very little effort. Initially, we created a knowledge base using background information about the database domain. Then used a probabilistic algorithm to calculate the semantic dependency between different nodes in the knowledge base. When processing the query, this dependency score will be used to resolve the target attribute. © Springer Nature Singapore Pte Ltd. 2018.