A Text based Natural Dialogue System using Entity Retrieval for News Documents

Abstract

A typical Dialogue system uses Natural Language Processing (NLP) to analyze human language and responds to user queries. The system relies on the current user query for answering instead relying on previous mentions. While this paradigm is sufficient for simple usage, it cannot be used for systems that rely on the entities previously mentioned in the conversation or even for a long term conversation. This Dialogue system uses a knowledge base consisting of user mentioned entities and their properties in order to reply user queries. The system has a contextual approach in dialogue delivery trying to achieve a memory retrieval answering paradigm. It detects entities on their first occurrence in the conversation and stores it in the knowledge base along with its attributes. While the future conversations have the entities that have been already mentioned in the previous conversations, the system generates appropriate replies for a given query by making use of the corresponding attributes which are treasured in the knowledge base.