A Diversity-Promoting Objective Function for Neural Conversation Models

Code

• github.com/aswinapk/chatbot_rnn
  tf★ 0
• github.com/sseol11/DialoGPT
  pytorch★ 0
• github.com/kennethjones17/chatbot
  tf★ 0
• github.com/ankita1404/ConversationalChatBot
  tf★ 0
• github.com/souvikdgp16/dialo_gpt_daily_dialog
  pytorch★ 0
• github.com/manu-sk/rnn_chatbot
  tf★ 0
• github.com/smruthidabbiru/Advanced-ChatBot
  tf★ 0
• github.com/neeleshnegi/chat-bot
  tf★ 0
• github.com/GraphGrailAi/DialoGPT
  pytorch★ 0
• github.com/UladzimirBaklan/qwerty
  tf★ 0

Abstract

Sequence-to-sequence neural network models for generation of conversational responses tend to generate safe, commonplace responses (e.g., "I don't know") regardless of the input. We suggest that the traditional objective function, i.e., the likelihood of output (response) given input (message) is unsuited to response generation tasks. Instead we propose using Maximum Mutual Information (MMI) as the objective function in neural models. Experimental results demonstrate that the proposed MMI models produce more diverse, interesting, and appropriate responses, yielding substantive gains in BLEU scores on two conversational datasets and in human evaluations.

Tasks

Reproductions