Search papers, labs, and topics across Lattice.
This paper introduces Profiler, a non-parametric module that captures human citation patterns to improve citation recommendation, addressing limitations of existing methods that are computationally expensive or introduce biases. They identify and address a flaw in the standard transductive evaluation setting by proposing a more realistic inductive evaluation. The proposed DAVINCI reranking model, which integrates Profiler with semantic information, achieves state-of-the-art results across multiple benchmarks.
Existing citation recommendation benchmarks overestimate real-world performance because they fail to account for the temporal constraints of recommending citations for *new* papers.
Proper citation of relevant literature is essential for contextualising and validating scientific contributions. While current citation recommendation systems leverage local and global textual information, they often overlook the nuances of the human citation behaviour. Recent methods that incorporate such patterns improve performance but incur high computational costs and introduce systematic biases into downstream rerankers. To address this, we propose Profiler, a lightweight, non-learnable module that captures human citation patterns efficiently and without bias, significantly enhancing candidate retrieval. Furthermore, we identify a critical limitation in current evaluation protocol: the systems are assessed in a transductive setting, which fails to reflect real-world scenarios. We introduce a rigorous Inductive evaluation setting that enforces strict temporal constraints, simulating the recommendation of citations for newly authored papers in the wild. Finally, we present DAVINCI, a novel reranking model that integrates profiler-derived confidence priors with semantic information via an adaptive vector-gating mechanism. Our system achieves new state-of-the-art results across multiple benchmark datasets, demonstrating superior efficiency and generalisability.
