PTTChatBot_DL2017

Deep Learning and Practice course (NCTU, 2017 spring) final project

• Group members: Mion (ninetf135246), thisray (thisray)

Introduction

In one of the course assignments, we have experimented with RNN to generate and process text sequences. It was a very interesting experience and motivated us to study whether it is possible to train talkshow chatbots.

In order to build chatbots capable of handling a wide range to topics, we trained our chatbots using the Chinese corpus from two of the PTT gossip forums: Gossiping and Women Talk. Both forums have their own styles of opinionated expressions. We aim to create chatbots to deliver quarrelsome or sarcastic responds to input queries.

Directory instruction

• /1_ptt_crawler: crawl Chinese articles from PTT.
• /2_make_QA: make data into Q&A corpus type.
• /3_chatbot_model: train & use chatbot. (We provide the dataset & model we have trained here.)

ChatBot environment

• python version: > 3.0
• requirement python package: tensorflow, numpy, jieba
• tensorflow version: 1.0

Reference (Codes credits)

• https://github.com/fxsjy/jieba
• https://github.com/ldkrsi/jieba-zh_TW
• https://github.com/zake7749/PTT-Chat-Generator
• https://www.tensorflow.org/tutorials/seq2seq
• https://github.com/sherjilozair/char-rnn-tensorflow
• https://github.com/Marsan-Ma/tf_chatbot_seq2seq_antilm