Increase Diversity in Chatbots’ Responses

May 12, 2020
Duration: One Semester
Lab: HCI/IC/EPFL
Goals: Build a chatbot that is able to generate diverse responses.
Assistant: Yubo Xie (yubo DOT xie AT epfl DOT ch)
Student Name: Open
Keywords: Generative dialog model; natural language processing
Abstract:

By training a generative neural model on massive dialog data, a chatbot can already respond in a way that makes sense. However, sometimes the replies generated by chatbots are generic and dull (for example “I don’t know”), and lack some specificity according to the conversation context. One explanation is that these utterances appear more frequently in the training set, which leads to a higher probability of being generated. This project aims at increasing the diversity of the responses generated by chatbots, while still keeping them on the conversation topic.

The student is expected to:

  • do some literature review on the diversity issue of open-domain chatbots;
  • build a generative dialog model and implement a way to increase the diversity of the chatbot’s responses (for example ranking the beam search results according to mutual information);
  • evaluate your approach using both automatic metrics and human judgement.
Related Skills: Basic knowledge in natural language processing and neural networks
Suitable for: Master student. Interested student should contact Yubo Xie (yubo DOT xie AT epfl DOT ch) and Pearl Pu (pearl DOT pu AT epfl DOT ch) along with a copy of your CV as well as your transcript.