A Deep Learning Approach for Polite Dialogue Response Generation
5 Pages Posted: 16 Aug 2019
Date Written: August 16, 2019
Dialogue generation is the act of creating response sentences that are contextually relevant, meaning full and grammatically correct. A natural language sentence is characterized by different features including linguistic features and stylistic features. In a dialogue generation system, the generated response can be polite, rude or neutral. The main objective of this proposed system is to generate polite dialogue responses by using deep learning techniques. The system mainly contains two modules: the first module implements a politeness classifier and the second module implements a response generation system. Politeness classifier calculates the politeness score for the input sentence and by using this score the response is generated. For implementing the classifier the Stanford politeness corpus is used and for implementing dialogue generation model the Cornell movie dialogue corpus is used. The advantages of Bi-directional Long Short Term Memory (BLSTM) is used for building the system. Input to the system is a sentence which will be processed by the classifier and the output of the classifier and the sentence after vectorization is given to the dialogue generation module which will generate the final output. The main advantage of this system is that it is created by a deep learning model which will perform well in large dataset.
Keywords: Linguistic Features, Dialogue Generation, Deep Learning, BLSTM (Bidirectional Long Short Term Memory, Politeness, Politeness Score
Suggested Citation: Suggested Citation