Mining opinions in Arabic text using an improved “Semantic Orientation using Pointwise Mutual Information” Algorithm

This paper presents an optimized approach for mining opinions in Arabic Religious Decrees using an improved “Semantic Orientation using Pointwise Mutual Information” Algorithm. The original approach executed a number of steps to classify a religious decree into either Halal (Allowed) or Haraam (Prohibited). Those steps included Data Collection, Simple Text Preprocessing, Manual Data labeling, Advanced Text Preprocessing, Weight Calculation and experimentation using Supervised and Unsupervised Learning Algorithms. Weight Calculation process utilized SO-PMI Algorithm proposed by Wang and Araki in 2008. Results obtained by original approach gave an accuracy rate of 73.08%. The new approach utilizes an improved SO-PMI Algorithm that executes a series of advanced steps to improve the calculation of the weights. The improved algorithm increased the accuracy rate of the Unsupervised Learning Algorithm up to 20% but produced poor results for the Supervised Learning Algorithm.