Information Retrieval Based on Knowledge-Enhanced Word Embedding Through Dialog: A Case Study

Abstract

The aim of this paper is to provide a systematic route of information retrieval from a knowledge-based database (or domain knowledge) through a dialog system of natural language interaction. The application is about a comprehensive building at a university, with classrooms, laboratory rooms, meeting rooms, research rooms and offices, and is to present related information the user asks for. First, the domain knowledge is expressed with predicate expressions based on the ontology structure; then the vocabulary is presented distributedly with word embedding enhanced with the domain knowledge; queries from the user are then converted into the intent (general) and slot elements (specific) with the help of trained recurrent neural network (RNN). The system works smoothly. The key point is integrating the two methods of knowledge-based and data-driven natural language processing into one system, and the domain knowledge is in the central part which is incorporated into the word embedding to make it specifically fit the natural language in this application.