An online language-learning platform.
When a student is taught a word in a foreign language, the likelihood of them recalling the meaning when tested decreases with time. It is well known that reinforcement is key to committing the word to long-term memory. The question is: at what points in the future should this be done? Wait too long before reinforcing and the student will never learn; reinforce too soon and you waste valuable learning time and risk boring the student with material they already know.
This online language-learning platform already used memory retention curves from academic literature in its product, but wanted to develop models that more accurately matched the data it had. The data available for the project was rich, with learning events from over 300,000 users. In addition, the client wanted to classify the users into different groups in order to tailor the learning experience for individuals.
Memory retention curves from academic literature were quickly shown to be poor matches for the data provided. We used a random forest machine learning algorithm, which produced a model with a much stronger fit, allowing us to begin classifying users into groups. This helped us to break down the users into fast-learning and slow-learning groups.
While the ‘actual data’ is taken from the learning of one word, ‘étude’, the random forest model is generalisable to any word.
The new models we provided have significantly more predictive power than those used before. Accurate identification of a student’s group will enable them to learn a language faster.