Overview / Usage

The goals include attempting to reduce the feelings of loneliness and ideally to be a companion. The challenges include curating a dataset based on interactions with a counsellor and patient (anonymously collected). And building a cutting edge conversational chatbot architecture.

Methodology / Approach

We will be investingating the baseline models from Stanford and the CoQA training datasets and build models for comprehension and extraction using both hybrid and deep learning techniques.

Typically the models will use sequence to sequence, LSTM encoder, decoder, DrQA, OpenNMT and traditional NLP techniques.

Technologies Used

Stack:
Intel Distribution for Python - Optimision for Intel Architectures
Intel® Optimization for TensorFlow

Software Architecture:
Conversational Model:

Sequence to Sequence with attention model
LSTM encoder, decoder

Reading Comprehension Model
Document Reader (DrQA)

References:
https://github.com/facebookresearch/DrQA
https://stanfordnlp.github.io/coqa/