Hybrid Knowledge-based Recommendation Approach for E-Learning Resources

Abstract

Recent years has witnessed substantial increase of learning resources available on the World Wide Web. As a result, there has been a remarkable growth in the utilization of online learning resources by learners in e-learning environments. However, despite this growth in usage of e learning resources, learners encounter difficulties of retrieval of relevant learning resources due to information overload. Recommender systems are software tools that have been largely accepted as useful solutions to alleviate the problem of information overload. They play a beneficial educational role in e-learning by assisting learners to access relevant learning resources that match their learning needs. Although conventional recommendation methods such as collaborative filtering and content based have demonstrated success in domains such as e-commerce, music and movies, there are still some challenges experienced in attempts to provide accurate and personalized recommendations of learning resources in e-learning arising from differences in learner characteristics, learner contexts and sequential access patterns among the learners. Learners possess characteristics such as learning style, background knowledge, skills and study level among others which are crucial in personalization of the learner profile and recommendations of learning resources in e-learning environments. In addition, learner’s contextual information such as knowledge level and learning goals change with time and situations. These contextual changes have an impact on learner preferences for learning resources. Similarly, different learners have different sequential access patterns for learning resources that can equally influence the learning resources that should be recommended to the learner. These additional learner information namely learner characteristics, learner context and sequential access patterns have some influence in determining the learner preferences for a learning resource, hence they should be captured during recommendation. In the context of e-learning, collaborative filtering recommendation approach recommends learning items to the target learner similar to the ones liked by other learners with similar preferences. A rating is used to measure the degree of usefulness of an item to a user. Ratings of learning resources by the learners are used to measure similarity of learners or learning resources. Content-based recommendation approach recommends learning resources to the target learner that are similar in content features to those liked by the learner in the past. Conventional recommendation methods do not incorporate additional learner information such as learner characteristics, learner context and learner’s sequential access patterns in generating recommendations for the learner. Besides, conventional recommendation approaches experience the cold-start and sparsity problems, making them unreliable in e-learning scenarios. Majority of the recommendation methods currently in use still face similar challenges due to lack of incorporation of additional learner information in their recommendation processes. As such, most of the existing recommendation methods are likely to generate recommendations with lower accuracy and poor personalization to learner preferences in e-learning environments. To overcome this problem, recommendation approaches that incorporate additional learner information into the recommendation process are required. The main goal of this thesis is to develop hybrid knowledge-based recommendation algorithms that take into account additional learner information such as learner characteristics, learner context and learner’s sequential access patterns in their recommendation processes to help improve personalization and accuracy of recommendations as well as alleviate sparsity and cold-start problems. Additionally, this thesis explores the learner and researcher related challenges of e-learning recommender systems. The major contributions of this thesis are described below. First, we explored the challenges of e-learning recommender systems by carrying out a systematic literature review of journal papers on e-learning recommender systems with a view to identifying and categorizing the challenges as either learner or researcher related challenges. In addition, we discuss the solutions for addressing each of the challenges. Secondly, we proposed a hybrid knowledge-based recommendation approach based on ontology and sequential pattern mining (SPM) algorithm for recommending relevant learning resources to learners in e-learning environments. In the proposed recommendation method, ontology was used to model as well as represent the knowledge about the learner and learning resources while SPM algorithm was used to mine the web logs and discover the learner’s sequential access patterns for filtering the recommendations according to the learner’s sequential access patterns. Experimental results over a real world dataset show improvement in performance in terms of accuracy, precision and recall metrics. Furthermore, the proposed recommendation method can alleviate the cold-start and data sparsity problems by using the ontological domain knowledge and learner’s sequential access patterns respectively before the initial ratings to work on are available in the recommender system. Lastly, we proposed a hybrid recommendation method combining context-awareness, SPM algorithm and collaborative filtering for recommending relevant learning resources to learners. In our method, context-awareness was used to incorporate the learner’s context information such as knowledge level and learning goals while SPM was used to discover the learner’s sequential access patterns. These sequential access patterns were incorporated as well into the recommendation process. Collaborative filtering was used to compute similarities, predict learner ratings and generate recommendations for the target learner taking into account contextualized data and learner’s sequential access patterns. Experimental results show that the proposed hybrid recommendation method can outperform other related recommendation approaches in terms of quality and accuracy of recommendations