Dual-layer Label Perception Algorithm for POI data with long-tailed distribution

Abstract

Authentic POI datasets often exhibit a long-tailed distribution, wherein instances vary significantly across labels, rendering conventional training methods ill-equipped to handle such real-world scenarios. This paper introduces a Dual-layer Label Perception Algorithm for POI Data, which pioneers the utilization of prevalent two-layer data labels to formulate a label feature cloud, subsequently enhancing classification accuracy by broadening and refining the distribution of label features. To assess the algorithm's efficacy, extensive experimentation is conducted on genuine urban POI datasets, resulting in a notable 5% enhancement in the comprehensive F1-Score, reaching 79.41%. The experimental findings affirm the algorithm model's efficacy in enhancing its detection performance across various advanced text classification algorithms.