Team: Tongda Xu (Technical Prototype), Wenye (Echo) Lu (Storytelling)
Ideation: Siri God
Prototype Function Description: a chat robot talk to people like the God/his son to human-being in the holy bible
Training dataset: the conversation between human and God/his son in the holy babel King James Version
Framework: tensorflow-gpu + tf-seq2seq or keras seq2seq (more straightforward):
Reference:
https://github.com/nicolas-ivanov/seq2seq_chatbot_links (A general review of current implement)
https://google.github.io/seq2seq/nmt/#neural-machine-translation-background (Original seq2seq)
https://github.com/oswaldoludwig/Seq2seq-Chatbot-for-Keras
"tf-seq2seq is a general-purpose encoder-decoder framework for Tensorflow that can be used for Machine Translation, Text Summarization, Conversational Modeling, Image Captioning, and more."
Schedule:
1 week. Extract the training data, reformat and segment into data-set.
Digital Bible King James Version: https://www.bible.com/zh-CN/search/bible?q=Jesus%20says%20&category=bible&version_id=547
Reformatting: From word to vector: https://nlp.stanford.edu/projects/glove/
such as: “I have given you every herb yielding seed ... for food” (Gen. 1:29-30).
2 week. Building, training and fine tuning of model in tensorflow/keras.
1.5 week. Front-end development. (Including speech to text, text to speech, general user interface)
Komentáře