Multimedia sentiment analysis using modified CNN and rnn models /
Youssef Saad Ghatas
Multimedia sentiment analysis using modified CNN and rnn models / دراسة لتصنيف الآراء ذات الوسائط المتعددة عن طريق نموذج يتضمن شبكة عصبية التفافية و شبكة عصبية متكررة Youssef Saad Ghatas ; Supervised Elsayed Hemayed - Cairo : Youssef Saad Ghatas , 2017 - 70 P. : charts , facsimiles ; 30cm
Thesis (M.Sc.) - Cairo University - Faculty of Engineering - Department of Computer Engineering
Multimedia Sentiment Analysis is considered a great challenge, given the diversity of data it combines and the informality challenges it produces. Firstly, this study divides the problem into textual and visual fields and study the effect of using different CNN and RNN models. Secondly, the proposed model is compared to the top candidate and the state-of-the-art model. The proposed model -which uses deep CNN model for images and RCNN model for texts- outperforms the other tested models and the state-of-the-art model on the same dataset in both accuracy and F1 score with absolute improvement of about 5% and relative error improvement of more than 25%. Finally, a sensitivity test is conducted to test the effect of different values for some important parameters of the proposed model
CNN Multimedia Sentiment Analysis RNN
Multimedia sentiment analysis using modified CNN and rnn models / دراسة لتصنيف الآراء ذات الوسائط المتعددة عن طريق نموذج يتضمن شبكة عصبية التفافية و شبكة عصبية متكررة Youssef Saad Ghatas ; Supervised Elsayed Hemayed - Cairo : Youssef Saad Ghatas , 2017 - 70 P. : charts , facsimiles ; 30cm
Thesis (M.Sc.) - Cairo University - Faculty of Engineering - Department of Computer Engineering
Multimedia Sentiment Analysis is considered a great challenge, given the diversity of data it combines and the informality challenges it produces. Firstly, this study divides the problem into textual and visual fields and study the effect of using different CNN and RNN models. Secondly, the proposed model is compared to the top candidate and the state-of-the-art model. The proposed model -which uses deep CNN model for images and RCNN model for texts- outperforms the other tested models and the state-of-the-art model on the same dataset in both accuracy and F1 score with absolute improvement of about 5% and relative error improvement of more than 25%. Finally, a sensitivity test is conducted to test the effect of different values for some important parameters of the proposed model
CNN Multimedia Sentiment Analysis RNN