Title: Lexicon for sentiment analysis based on ensemble deep learning

Abstract

In this work, lexicon developing will be presented as an optimization problem. The proposed solution will depend on the concept of ensemble learning. Most of current leaning methods tends to be ensemble in nature. In this work, a new joining method will be defined, namely, gradually joining method. Gradually here means passing the output lexicon of the current model as an initial data for the next model. All the used models will be a kind of a deep net of the same type with different hyper parameters that will optimized based on the results of the previous model. Since LSTM is the most proper deep net for sequences processing, all models will be a variation of LSTM nets. A feature will be added to represent the corresponding values of each term in the lexicon. For the next generations of the model, if the resulting accuracy is less than the previous one the lexicon will be modified according to the classification results. Also, the hyper parameters will be modified based on internal optimization protocol to achieve a better model. Then the training is repeated, and the adjusting process continue till reaching a termination condition. A shortcut will be adopted to slightly modify the structure of the training model and a random modification of the current lexicon will be done to avoid local optimization trap. This manner will be imposed when a non-improvement in the accuracy is noticed especially during early training generations. Since the process proceed to produce a better model as well as an optimized lexicon, it is expected to achieve accuracy values that compete the current art of state values for the resulting model and the output lexicon.

Biography

MAHMOUD B. ROKAYA was born in Tanta city, Egypt in 1971. He received the B. S and M.S. degrees in mathematics from Tanta University, Egypt in 2003 and the Doctor of Engineering in information science From Tokushima University, Japan in 2009. From 1997 to 2003, he was assistant of teaching in department of mathematics, Tanta University, Egypt. From 2003 to 2009, he was a researcher in the advanced engineering institute, Tokushima University, Japan. Since 2009-2020, he was assistant professor in information technology, Taif University, KSA. Currently, currently, he works as associate professor in informatics, Taif University. His research interests related to AI, information retrieval, natural language processing and data science.