Political-Advertisement Video Classification using Deep Learning Methods
Today’s digital world consists of vast multimedia contents: images, audios and videos. Thus, the availability of huge video datasets have encouraged researchers to design video classification techniques to group videos into categories of interest. One of the topics of interest to political scientists is automated classification of a video advertisement into a political campaign ad category or others. Recent years have seen a plethora of deep learning-based methods for image and video classification. These methods learn feature representation from the training data along with the classification model. We investigate the effectiveness of three recent deep-learning based video classification techniques for the political video advertisement classification. The best technique among the three yields an accuracy of 80%. In this thesis, we further improve the classification accuracy by combining the results of classification of text features with that of the best deep learning method. Our method achieves the classification accuracy of 91%.
Committee: Adisak Sukul (major professor), Wallapak Tavanapong (major professor), & David Peterson