Imam: Word Embedding Model for Islamic Arabic NLP

This paper can be considered one of the first works to introduce an efficient distributed word representation model for different NLP tasks in the islamic domain. The Word Embedding Model and the algorithm on top of it is implemented in Imam application where user can ask the application to search for any data related to Isalmic domain and get an answer. The data is gathered from different resources (Maliks muwataa, Musnad Ahmad Ibn-hanbal, Sahih Muslim ahadith, Sahih Al-bukhari, Sunan Al-darimi, and more). The amount of records gathered was more than ninety thousand documents (Text Blocks) from 10 different books.After several sequential pipeline processes of Data cleaning, preprocessing and Normalization, Skip-gram technique was used to built the word2vec model and then At last tested with different methods, first by using the K-means clustering and then nonlinear dimensionality reduction technique to represent the data in 2D dimension, secondly by using word similarity to test model ability to understand the Quranic language. The tests clearly show that the model can be used effectively in different NLP Arabic Islamic tasks.