Integrating Knowledge Graphs with Natural Language Processing for Context-Aware Educational Content Recommendations

Abstract

Personalization of content recommendation for education has resulted in a combination of knowledge graphs and natural language processing in order to develop context-aware recommendation systems. The proposed paper presents the KG-NLP-CAR (Knowledge Graph and Natural Language Processing for Context-Aware Recommendations) model, that allows for dynamic personalization of recommendations based on the power of reasoning inherent in knowledge graphs and the contextualization provided by natural language processing technologies. Specifically, the uniqueness of the proposed model is associated with using specific information about each individual learner such as preferences, goals, and interactions with the platform, all of which are represented within a knowledge graph. Preprocessing of the educational content is performed using NLP tools such as tokenization, entity recognition, and generation of semantic embeddings. Thus, the model is able to give recommendations for content that will be most useful for the user according to his context. The performance of the model was evaluated on Last.FM, Book-Crossing, and MovieLens-1M datasets, where the KG-NLP-CAR model performed better than the several state-of-the-art models. It gave a very high AUC value (95.65%) and an F1 score of 88.45% when evaluated on the MovieLens-1M dataset, easily beating Ripple Net and KGAT in all these categories. The ablation studies conducted showed that KGs and NLP were key factors in making the recommendations accurate and relevant. Thus, the KG-NLP-CAR model shows that the combination of KGs and NLP can be effectively used to give educational recommendations.