Description
Abstract: This work aims at combining user activity data and learning analytics as inputs on a Deep Learning system to order posts for a social e-learning platform, Fullbrain. Fullbrain allows users to learn socially and generates an unprecedented learning analytics footprint. As previous researchers have attempted to use similar data with Deep Learning modelling, we consider this an appealing scope to conduct research. This work defines the social data and learning analytics features that should partake as input in the Machine Learning system, defines a customised Deep Learning Multigated Mixture of Experts model based on the state-of-art in multi-task learning and proposes an Experimental Framework for Machine Learning models based in A/B Randomized Controlled Trials (RCTs). The proposed model is tested for offline accuracy using a set of sinusoidal functions that act as proxies for sequential user activity data. We find the first evidences that our model is not sufficient for the task at hand and propose future research to focus on a similar evaluation of LSTM based Mixtures of Experts.| Period | 15 Jun 2020 → 30 Nov 2020 |
|---|---|
| Examinee | Panagiotis Filianos |
| Degree of Recognition | International |