- Group members: Mion (ninetf135246), thisray (thisray)
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.
/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.)
- python version: > 3.0
- requirement python package: tensorflow, numpy, jieba
- tensorflow version: 1.0