NLP-based artificially intelligent chat-bot

Abstract

Tools such as search engines and personal agents lack personality, evidenced by their mechanical acceptance of input and production of output, and practicality, demonstrated by their in-take of a single type of input, like keywords. To assuage these issues, techniques developed in Natural Language Processing (NLP) are used to produce a model of a system which combines a normal chatter-bot with a more intelligent document categorization and retrieval system, thereby creating a new digital assistant system. The NLP techniques used in this project include topic modeling, recursive distributed representation, which is a form of connectionist modeling, and Weizenbaum's ELIZA. The first of the three is used to categorize a corpus of documents, while the other techniques are used to interact with the user. Reliance on keywords was found to be unavoidable, but modeling the corpus to topics rather than a spatial distribution, such as that of frequency, proved to be successful at retrieving the relevant documents, to a certain extent. The topic modeling mechanism appeared to be highly influenced by the number of words in each document. The connectionist modeling was worse than ELIZA at recognizing trained phrases, but was better at dealing with unknown words.