Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Newborn to Impact: Bias-Aware Citation Prediction

Mingfei Lu, Mengjia Wu, Jiawei Xu, Weikai Li, Feng Liu, Ying Ding, Yizhou Sun, Jie Lu, Yi Zhang

TL;DR

This work addresses bias and cold-start challenges in newborn citation prediction by combining fine-grained, agent-derived implicit factors with dynamic heterogeneous graph learning. It introduces BA-Cite, a two-stage exposure-based predictor augmented with GroupDRO and a counterfactual regularization head to reduce reliance on superficial cues like venue prestige. Empirical results on AMiner and OpenAlex show consistent improvements in both error and ranking metrics, especially under distribution shifts and for low-citation papers. The approach offers a robust, interpretable mechanism for early-impact assessment with potential for cross-domain transfer and explainability.

Abstract

As a key to accessing research impact, citation dynamics underpins research evaluation, scholarly recommendation, and the study of knowledge diffusion. Citation prediction is particularly critical for newborn papers, where early assessment must be performed without citation signals and under highly long-tailed distributions. We identify two key research gaps: (i) insufficient modeling of implicit factors of scientific impact, leading to reliance on coarse proxies; and (ii) a lack of bias-aware learning that can deliver stable predictions on lowly cited papers. We address these gaps by proposing a Bias-Aware Citation Prediction Framework, which combines multi-agent feature extraction with robust graph representation learning. First, a multi-agent x graph co-learning module derives fine-grained, interpretable signals, such as reproducibility, collaboration network, and text quality, from metadata and external resources, and fuses them with heterogeneous-network embeddings to provide rich supervision even in the absence of early citation signals. Second, we incorporate a set of robust mechanisms: a two-stage forward process that routes explicit factors through an intermediate exposure estimate, GroupDRO to optimize worst-case group risk across environments, and a regularization head that performs what-if analyses on controllable factors under monotonicity and smoothness constraints. Comprehensive experiments on two real-world datasets demonstrate the effectiveness of our proposed model. Specifically, our model achieves around a 13% reduction in error metrics (MALE and RMSLE) and a notable 5.5% improvement in the ranking metric (NDCG) over the baseline methods.

From Newborn to Impact: Bias-Aware Citation Prediction

TL;DR

This work addresses bias and cold-start challenges in newborn citation prediction by combining fine-grained, agent-derived implicit factors with dynamic heterogeneous graph learning. It introduces BA-Cite, a two-stage exposure-based predictor augmented with GroupDRO and a counterfactual regularization head to reduce reliance on superficial cues like venue prestige. Empirical results on AMiner and OpenAlex show consistent improvements in both error and ranking metrics, especially under distribution shifts and for low-citation papers. The approach offers a robust, interpretable mechanism for early-impact assessment with potential for cross-domain transfer and explainability.

Abstract

As a key to accessing research impact, citation dynamics underpins research evaluation, scholarly recommendation, and the study of knowledge diffusion. Citation prediction is particularly critical for newborn papers, where early assessment must be performed without citation signals and under highly long-tailed distributions. We identify two key research gaps: (i) insufficient modeling of implicit factors of scientific impact, leading to reliance on coarse proxies; and (ii) a lack of bias-aware learning that can deliver stable predictions on lowly cited papers. We address these gaps by proposing a Bias-Aware Citation Prediction Framework, which combines multi-agent feature extraction with robust graph representation learning. First, a multi-agent x graph co-learning module derives fine-grained, interpretable signals, such as reproducibility, collaboration network, and text quality, from metadata and external resources, and fuses them with heterogeneous-network embeddings to provide rich supervision even in the absence of early citation signals. Second, we incorporate a set of robust mechanisms: a two-stage forward process that routes explicit factors through an intermediate exposure estimate, GroupDRO to optimize worst-case group risk across environments, and a regularization head that performs what-if analyses on controllable factors under monotonicity and smoothness constraints. Comprehensive experiments on two real-world datasets demonstrate the effectiveness of our proposed model. Specifically, our model achieves around a 13% reduction in error metrics (MALE and RMSLE) and a notable 5.5% improvement in the ranking metric (NDCG) over the baseline methods.
Paper Structure (39 sections, 11 equations, 4 figures, 2 tables)

This paper contains 39 sections, 11 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Motivation
  5. Addressing Bias in Long-Tailed Citation Prediction
  6. Capturing Core Factors Driving Citation Dynamics with Agents and GNN(s)
  7. Fine-grained feature extraction
  8. Role-aware Author Reputation (A)
  9. Venue Prestige (V)
  10. Reproducibility (R)
  11. Collaboration Network (C)
  12. Topic Hotness (H)
  13. Text Quality (Q)
  14. Output
  15. Bias-Aware GNN Learning Modules
  16. ...and 24 more sections

Figures (4)

  • Figure 1: Overall architecture of the proposed BA-Cite framework.
  • Figure 2: Distribution of citation counts across venues of different prestige levels on the Aminer and OpenAlex datasets.
  • Figure 3: Parameter sensitivity results of BA-Cite on the Aminer and OpenAlex datasets.
  • Figure 4: Performance comparison of three models on Aminer dataset with different citation levels. (a) Results on lowly cited papers. (b) Results on highly cited papers.