My personal chatbot. The website where you can try the chatbot will be released quite soon, as the model is made, where you can talk to the chatbot . Don't expect a chatbot as complicated as Assistant or Siri, yet it is a powerful chatbot, you can surely have a good time chatting with. Also, this chatbot is trained on movie dialouges , so expect quite filmy dialogues .
Since I am in need of a personal GPU, I have been training the models on the free GPU sessions on Kaggle and colab. Now , since the free sessions have limits I could not train the model on massive datasets, but I managed to train the models on 1000 sentences of the CORNELL-MOVIE-DATASET dataset, since it contains both the text and summary. I also tried the model to run on the BBC news dataset, but , unfortunately the free session could not manage the massive size of the dataset. Thus I trained the model for 10 epochs , using the GRU cell and seq2seq with attention mechanism . So, being said that , these are the results I have got.
while True:
question = input("You: ")
if question.lower() == 'goodbye':
break
else:
result = answer(question,
encoder,
decoder,
inp_lang,
targ_lang,
max_length_inp,
max_length_targ)
print('Chatbot: ', result)
You: how are you?
Chatbot: it doesn t tell me about back i was it back when it . <end>
You: what's your name again ?
Chatbot: okay . what just worlds n . . . maybe you know you are eddie . <end>
You: okay, so you are eddie
Chatbot: oh , eddie . <end>
You: hi
Chatbot: tamina could know ! i supposed to doin me done , sweetheart ? <end>
You: is your name cameron ?
Chatbot: i ride for , just for your son , we questioned you wanna have , tabloids chosen for a good . like his get . i used out with me wrong . not s a hell , too . and by your america . <end>
You: how do you get your hair to look like that ?
Chatbot: again . <end>
You: yes
Chatbot: does . rights stupid ways run sinned . <end>
You: how do you get your hair like that ?
Chatbot: william son . i thought <end>
You: goodbye
I am completely estatic , that you have thought of helping me. If you wish to do so, kindly find me and my kernels on KAGGLE, at https://kaggle.com/justdvnsh