Graph Embedding for Citation Recommendation
This work addresses the problem of efficiently finding relevant papers for researchers, though it is incremental as it builds on existing graph embedding methods.
The paper tackles citation recommendation by proposing a task-specific neighborhood construction strategy for graph embedding, which outperforms random walks-based sampling across ranking schemes and shows robustness when the set of seed papers is small, with performance drops in classic methods being more significant.
As science advances, the academic community has published millions of research papers. Researchers devote time and effort to search relevant manuscripts when writing a paper or simply to keep up with current research. In this paper, we consider the problem of citation recommendation on graph and propose a task-specific neighborhood construction strategy to learn the distributed representations of papers. In addition, given the learned representations, we investigate various schemes to rank the candidate papers for citation recommendation. The experimental results show our proposed neighborhood construction strategy outperforms the widely-used random walks based sampling strategy on all ranking schemes, and the model based ranking scheme outperforms embedding based rankings for both neighborhood construction strategies. We also demonstrated that graph embedding is a robust approach for citation recommendation when hidden ratio changes, while the performance of classic methods drop significantly when the set of seed papers is becoming small.