How Machine Learning Improves Reading Recommendations for Students
Reading is crucial for student development, offering various benefits such as improved language skills, academic success, and better understanding of different cultures. However, finding relevant and engaging reading material can be challenging. This is where machine learning (ML) can help. ML has been widely used in recommender systems across digital platforms to suggest content based on user preferences and engagement. In the case of educational audiobooks, Google Research collaborated with Learning Ally, an educational nonprofit, to develop a content recommender system called “STUDY.”
The STUDY algorithm takes into account the social aspect of reading. It considers the reading engagement history of students in the same classroom, leveraging the idea that what peers are reading influences a person’s interests. By analyzing anonymized data from Learning Ally’s audiobook library, the algorithm provides personalized recommendations to improve the reading experience and engagement of dyslexic students.
To build the social recommendation model, two years of anonymized audiobook consumption data was used. The data included student interactions with audiobooks, identified only by randomly generated IDs. The STUDY algorithm frames the recommendation problem as a click-through rate prediction problem and utilizes Transformer-based models developed by Google Research. However, since the sequence of books read is not temporally ordered, a standard causal decoder is not suitable. To address this, the STUDY model replaces the triangular attention matrix with a flexible attention mask based on timestamps, allowing attention across different subsequences.
Experiments were conducted using the Learning Ally dataset to evaluate the effectiveness of the STUDY model. Multiple baselines, including an autoregressive click-through rate transformer decoder, a k-nearest neighbor baseline, and a social attention memory network, were compared. The models were evaluated based on the percentage of accurate recommendations in the top n suggestions.
The results showed that the STUDY model outperformed the baselines, providing more relevant recommendations to students. This demonstrates the effectiveness of using machine learning in improving reading recommendations for students, ultimately enhancing their reading experience and engagement.
By leveraging ML algorithms like the STUDY model, educators and educational platforms can ensure that students have access to age-appropriate, engaging, and relevant reading material, fostering a love for reading and promoting academic success.