Implementing Artificial Intelligence Algorithms To Develop Recommender System For Personalized Education

Abstract

Recommender systems have been widely used in various domains, including e- commerce, social media, and entertainment. In recent times, the education domain has

also witnessed significant attention towards the use of recommender systems because personalized recommendations based on behavior patterns and preferences can enhance students' learning experience. This research focuses on developing a recommender system for education using the publicly available EdNet dataset comprising over 8 million student interactions with online learning platforms. The approach involved several steps, starting with analysing of student behavior patterns. Subsequently, students were divided into groups based on their performance as, the top performers, average and low performing. This was done by analysing the time taken to attempt all questions and how many questions they got right. However, processing the massive dataset with many users and interactions was challenging. To address this, a Min-Max scaler was utilized to scale the dataset, and principal component analysis (PCA) was applied to reduce the dimensionality of the data. The K-Means algorithm was employed to identify distinct clusters within the student population, based on their academic trajectories, with the aim of grouping together students who exhibited similar study paths. Initially, 50 clusters were utilized, resulting in a sum of squared errors (SSE) score of 6.74, which closely approached the ideal value. Subsequently, the optimal number of clusters was determined through the implementation of an elbow plot, resulting in the selection of 90 clusters, which effectively facilitated the grouping of students into cohorts based on shared characteristics. Additionally, a K-Nearest Neighbour (KNN) machine learning algorithm, in conjunction with a cosine matrix, was employed to construct a recommender system using the clusters derived from the K-Means algorithm. To validate the recommender system, students’ path was tested by inputting different user and assessing the path provided. Specifically, the system takes the input of a user ID for whom recommendations are sought and provides a list of the top 6 performing users whose learning paths match that of the input user. This yielded an SSE score of 3.15, which closely approximated the ideal scenario. As a result, tailored study paths were devised for each cohort.

Abstract

Page | 13 In conclusion, a recommender system has been developed that utilizes the KNN algorithm to generate recommendations for students based on their learning trajectories. The recommender system suggested in this study could be used by educational platforms to recommend personalized content to students based on their learning history and behaviour. This could result in a more engaging and effective learning experience for students at universities and colleges.