Collaborative Filtering
- User/Item-based CF
- Matrix Factorization techniques
- Deep Neural Network-based CF approaches.
Input data
historical, long-duration engagement data, like course watch history, as input to the model. To reduce the noise, we curate the data in a pre-processing step by filtering the learner’s course watch history using the recency of the course watch and the depth of the engagement, like total course watch time. This means that if a learner only watches the first three seconds of a course, that engagement does not impact our model in the same way as viewing a full course session.
Architecture
