Use Seq2seq to train a chatbot talk like Chandler and Angry Chandler
Data Process: Download friends scripts, processing the data. https://fangj.github.io/friends/ For finding the conversation talked by Chandler, I used the following strategy. I found that if a scene has many people, and one wants to talk to Chandler, not other people, it will have a "(to Chandler)" in the sentence. But there may be many "(to ***)" in one sentence, so I cut the other part, and only reserve the "talk to Chandler" part sentence and found the next Chandler answer as one pair of QA. If someone says something and then Chandler says something, I will reserve this kind of conversation. Although it may have some errors. (For example, They are answering the same question asked by some other people.) But I still consider this strategy is worked. Because most of these conversations are one talking to Chandler, and although there may have some errors, the two sentences must have the same topic. If Chandler's sentence is the first sente...