Youtube-Trending-ML
This repository contains the machine learning part of the YouTube-trending analysis. To see how the data is being collected please have a look at this repository. The amount of required filtering is calculated based on the thumbnail and the title of the video. Both parts are evaluated separately and will be combined into a single score in the extension.
Thumbnail based evaluation
In this part we train an NN to distinguish the thumbnails of trending videos from videos sampled from my subscription feed.
To do this we retrain a MobileNet that was trained on ImageNet.
The training takes place in the trainig.ipynb.
In the tf_lite_test.ipynb we can see an example prediction implementation together with the prediction of a few thumbnails (that were part of the training process).
Title based prediction
This section will contain information about the title based prediction once it is implemented. It is planned to be based on an sentiment analysis architecture.