Recommender Systems for Audiobooks: Boosting Student Reading with Machine Learning
Reading is crucial for young students, as it improves their linguistic skills, life skills, and academic success. However, finding relevant and engaging reading material can be challenging. That’s where machine learning comes in. Machine learning has been used to build recommender systems for various types of content, including books. These systems suggest items to users based on their preferences and engagement.
In a recent study, titled “STUDY: Socially Aware Temporally Causal Decoder Recommender Systems,” Google Research collaborated with Learning Ally, an educational nonprofit that provides audiobooks to dyslexic students, to develop a content recommender system for audiobooks in an educational setting. The aim is to help students find the right content to enhance their reading experience.
The STUDY algorithm takes into account the social aspect of reading. It leverages the reading engagement history of students in the same classroom, allowing the model to benefit from what is currently trending within their localized social group. Learning Ally’s large digital library of curated audiobooks provides the perfect dataset for this social recommendation model.
To build the recommendation system, the researchers framed the problem as a click-through rate prediction problem. They used a Transformer-based model, a popular model class in machine learning developed by Google Research. However, they faced a challenge: the data representation required careful attention. Traditional Transformer models assume temporal order, but the STUDY model needed to consider non-ordered sequences. To address this, the researchers introduced a flexible attention mask with values based on timestamps to allow attention across different subsequences.
The research team conducted experiments using the Learning Ally dataset. They compared the performance of the STUDY model with other baselines, such as an individual autoregressive model, a k-nearest neighbor baseline, and a social attention memory network. The evaluation metric used was the percentage of times the model’s top recommendations matched the items the users actually interacted with.
The results showed that the STUDY model outperformed other models in recommending audiobooks, demonstrating the effectiveness of the socially aware temporally causal decoder recommender system. With the help of machine learning, students can have access to age-appropriate and engaging reading material, ultimately improving their reading experience and engagement.
In conclusion, machine learning has the power to enhance student reading by providing effective recommendations for relevant and engaging audiobooks. The STUDY algorithm, developed in collaboration with Learning Ally, demonstrates the potential of recommender systems in promoting reading and improving student learning outcomes. It’s a valuable tool for educators and students alike in the digital age.