Chatbot seq2seq (RNN)

Chatbot seq2seq consists of 2 components - an RNN encoder and decoder. The system has been implemented using the Tensorflow library. The task of the encoder is to create an internal vector representation of typed text (recognized by the speech recognition module). The decoder, on the other hand, after receiving this sequence, generates the output text that is most appropriate for the text string entered at the input. In order to train the seq2seq model, a set of data was used, including subtitles (from which conversations between heroes were extracted) and Beksinski's letters. The database consists of pairs: {question, answer} and contains more than 100MB of text data.

Technologies: Tensorflow, RNN, Google Cloud