Intelligent clustering of authentic islamic texts based on contextual similarity using deep learning techniques

While issuing the rulings regarding some social or religious affair, the Islamic Jurists (legal experts) has to go into all the basic but authentic texts to find the stuff which is directly related or having the contextual analogy or has an association with it. It is obvious that the verdict or ruling of a jury or jurist will be strong if he takes into account all the contextually related stuff. If the decision is taken just at the shallower level, it potentially leads to the rift and conflict in the society and has been historically observed. In this thesis, we propose various techniques based on deep learning, which upon query, infers a context, and produces before the jury all the related corpus. Deep learning techniques especially Deep Neural Networks and unsupervised clustering algorithms (K-Means, BIRCH, Deep Embedded Clustering (DEC) and Spectral ) were used to develop clusters based on the contextual similarity and analogy. The embeddings were calculated in itself from the text corpus trained on embedding models Word2Vec, Doc2Vec, Facebook's fastText, and Stanford's GloVe in the multidimensional context. Google's Tensorflow along with Keras framework have been deployed to carry out the training of the autoencoder.