An Ensemble Model with Multi-Scale Features for Incorrect Assignment Detection
Keywords: Name Ambiguity, Machine Learning, Text Embedding, Feature Extraction, Tree Model, Graph Neural Network
Abstract: With the number of the publication increasing, the name ambiguity problem is becoming increasingly complex. To improve this research, OGA-Challenge Team published a large-scale dataset and hosted KDD Cup 2024 Challenge for detecting paper assignment errors based on each author and their paper matadata. This paper presents an effective and resource-efficient solution to the aforementioned challenge. Rather than utilising LLM, we have elected to employ an embedding model for the representation of text information. Furthermore, we have implemented multi-scale feature extraction and a graph neural network for the extraction of relationships between papers. Finally, with our solution, our team LoveFishO won 2nd place in task1(WhoIsWho) among 400+ participants.
Submission Number: 12
Loading