OAG-Bench: A Human-Curated Benchmark for Academic Graph Mining

TL;DR

OAG-Bench addresses the gap in rigorous, multi-aspect benchmarking for academic graph mining by building a human-curated, large-scale AKG-based benchmark on the Open Academic Graph. It introduces a four-stage framework (entity construction, graph completion, knowledge acquisition, and trace/prediction) across 10 tasks and 20 datasets, with data processing tools, standardized protocols, and a community-driven OAG-Challenge. Empirical results across tasks reveal that while some methods (including certain LLMs) excel on specific challenges, others like paper source tracing and scholar profiling remain difficult, underscoring the need for integrated, multi-faceted evaluation. The work provides a shared platform to accelerate algorithm development, enable fair comparisons, and promote community engagement in academic graph mining.

Abstract

With the rapid proliferation of scientific literature, versatile academic knowledge services increasingly rely on comprehensive academic graph mining. Despite the availability of public academic graphs, benchmarks, and datasets, these resources often fall short in multi-aspect and fine-grained annotations, are constrained to specific task types and domains, or lack underlying real academic graphs. In this paper, we present OAG-Bench, a comprehensive, multi-aspect, and fine-grained human-curated benchmark based on the Open Academic Graph (OAG). OAG-Bench covers 10 tasks, 20 datasets, 70+ baselines, and 120+ experimental results to date. We propose new data annotation strategies for certain tasks and offer a suite of data pre-processing codes, algorithm implementations, and standardized evaluation protocols to facilitate academic graph mining. Extensive experiments reveal that even advanced algorithms like large language models (LLMs) encounter difficulties in addressing key challenges in certain tasks, such as paper source tracing and scholar profiling. We also introduce the Open Academic Graph Challenge (OAG-Challenge) to encourage community input and sharing. We envisage that OAG-Bench can serve as a common ground for the community to evaluate and compare algorithms in academic graph mining, thereby accelerating algorithm development and advancement in this field. OAG-Bench is accessible at https://www.aminer.cn/data/.

Figures (5)

• Figure 1: OAG-Bench overview.
• Figure 2: The overall construction framework of OAG-Bench.
• Figure 3: Results of scholar interest extraction. SimCSE: SimCSE-RoBERTa-large, S-BERT: all-MiniLM-L6-v2.
• Figure 4: Results of academic question answering.
• Figure 5: Results of paper source tracing.