All the feature extraction from the two questions are saved into a csv file as a training set and are made ready for the training. Google news vector of 300 dimension is used for the vector conversion of the question pair. Own artificial neural network is yet to be made.
Testing with the supervised machine learning models, the accuracy was highest obtained to be 0.76 This is aimed to be increased through our own neural network . Stay tuned ;)
**Update: **
Relying on the project's idea, new workflow of architecture of was designed and a research was proposed. You can have a full read of the paper here:
**"Exploring Deep Learning in Semantic Question Matching"**
https://ieeexplore.ieee.org/document/8586832
Thanks!