Predicting Video Memorability Scores Using various Machine Learning Models. This project is an attempt on solving "Predicting Media Memorability Task at MediaEval 2018" challege. I have used the dataset provideed by the organisers to train and build the model. The code for this challege was written using python.
Repository organization:
-
Dataset : Contains two files namely : dev-set_ground-truth.csv : This file contains the ground truth of memorability scores given to each video. dev-set_video-captions.txt: Contains the captions assigned to each video.
-
Models_Exploration.ipynb In this file, various features have been extracted and explored from the videos dataset. I have extensively used Natural Language Processing techniques for retrieving the text features from captions. After evaluting all models, the final model was built.
-
Memorability_Predictor.ipynb This is the final predictor, that can predict the short-term and long-term memorability scores of any video with captions. The best result for this predictor for predicting the short-term memorability scores is 𝟬.𝟰𝟬𝟰 (which is promising, as the highest scores achieved in this challange is 0.49)
Additional details and some research on this challenge is described in Model_Report.pdf file.