Submodular Approaches for Citation Recommendation

Publication Type:
Issue Date:
Full metadata record
With the rapid growth of the scientific literature, manually selecting appropriate citations for a paper is becoming increasingly challenging and time-consuming. Automated citation recommendation can help ease this challenge by suggesting the most appropriate citations for a query document, e.g., a thesis draft. While several approaches for automated citation recommendation have been proposed in the recent years, effective improvements in content-based citation recommendation are still elusive to a large extent. In this thesis, we aim to find a novel approach for recommending global citations for an academic paper draft based on deep representation learning and submodular inference. The current state-of-the-art systems for this task are based on deep learning and graph representations and have already achieved im- pressive results. However, their results are only based on the ranking of matching scores and do not fully answer the question: “is the recommended list of references adequate?”. In this thesis, we aim to provide an answer to this question by applying citation selection methods instead of matching score rankings only. For this, we rely on submodular inference in combination with a deep sequential representation of the query and the candidate references. The results we have obtained show that the proposed approach has been able to outperform a range of existing, relevant approaches over a number of citation datasets and evaluation measures.
Please use this identifier to cite or link to this item: