||Most literature database systems use content-based technique to retrieve articles to users. However, the content-based technique relies on exact keywords provided by users to search for articles the users are interested in. On the other hand, most recommender system techniques are based on user’s long-term browsing/transaction history so as to recommend items that meet users’ long term interest. However, in literature database system, users’ information need is often short-term. Previous works in recommending articles to satisfy users’ short-term interest have utilized article content, usage log, and coauthorship network. |
In this study, we extend coauthorship network method and incorporate scholars’ collaboration topics into the coauthorship network. Specifically, we propose a LDA-coauthorship-network-based technique that integrates topic information into links of the coauthorship network using latent Dirichlet allocation (LDA), and a task-focused (short-term) technique is proposed for recommending literature articles. Experimental results show that the proposed approach is more effective than the traditional coauthorship network method under all operating regions. When compared to the content-based technique, it has better performance when each task profile contains articles that are similar in their content but is less effective otherwise. We further develop a hybrid method that switches between content-based technique and LDA-coauthorship-network-based technique based on the content coherence of a task profile. Experimental results show that the hybrid method outperforms all the other methods under all operating regions.