MLPRank: Importance ranking model from multiple perspectives for unsupervised keyphrase extraction

2026-08-07

PRE

Zhihao Zhang

Chuanqian Zeng

Shi Yuan *

Jihaochen Zhang

DOI：10.1016/j.eswa.2026.133955

Original

Original request for help

Save

Abstract

En 中文

• MLPRank integrates topic, structure, and saliency cues for ranking. • Phrase integrity is measured via PMI-based compositional cohesiveness. • Topic consistency is captured by clustering phrase-level embeddings. • Phrase saliency is estimated through a graph-based positional ranking. • It outperforms existing unsupervised methods on three benchmark datasets.

Keywords:

Unsupervised keyphrase extraction

Multi-perspective information aggregation

Pre-trained language model

Pointwise mutual information

MLPRank: Importance ranking model from multiple perspectives for unsupervised keyphrase extraction

Abstract

Journal

Organization

Cited Papers

Citing Papers