Providing personalized online learning services has become a hot research topic. Online knowledge-sharing services represents a popular approach to enable learners to use fragmented spare time. User asks and answers questions in the platform, and the platform also recommends relevant questions to users based on their learning interested and context. However, in the big data era, information overload is a challenge, as both online learners and learning resources are embedded in data rich environment. Offering such web services requires an intelligent recommender system to automatically filter out irrelevant information, mine underling user preference, and distil latent information. Such a recommender system needs to be able to mine complex latent information, distinguish differences between users efficiently. In this study, we refine a recommender system of a prior work for web-based knowledge sharing. The system utilizes attention-based mechanisms and involves high-order feature interactions. Our experimental results show that the system outperforms known benchmarks and has great potential to be used for the web-based learning service.