Design of an Anomaly Detection Model Utilizing Semantic and Relational Features of Papers Assigned to Authors

18 Jul 2024 (modified: 15 Aug 2024)KDD 2024 Workshop OAGChallenge Cup SubmissionEveryoneRevisionsBibTeXCC BY 4.0
Keywords: KDD Cup, OAG, academic knowledge graph, academic graph mining
Abstract: The primary goal of academic data mining is to deepen the understanding of scientific development, nature, and trends, thereby maximizing scientific, technological, and educational value. Currently, many entity-centric applications, such as paper search, expert discovery, and venue recommendation, exist. However, the lack of appropriate public benchmarks significantly limits the progress of academic graph mining. To address this issue, the Knowledge Engineering Group (KEG) at Tsinghua University and Zhipu AI organized the OAG-Challenge in KDD Cup 2024. This paper introduces the solution presented by our DOCOMOLABS team, which achieved 6th place in the paper assignment error detection task (IND) as part of the OAG-Challenge. Our approach captures the semantic and relational features among papers assigned to each author using various methods and constructs a high-precision paper assignment error detection model by ensembling multiple binary classifier models. Our solution's source code is available on GitHub.
Submission Number: 14
Loading