Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Click it or Leave it: Detecting and Spoiling Clickbait with Informativeness Measures and Large Language Models

Wojciech Michaluk, Tymoteusz Urban, Mateusz Kubita, Soveatin Kuntur, Anna Wroblewska

TL;DR

This paper tackles clickbait detection by integrating explicit informativeness measures with transformer-based embeddings in a hybrid XGBoost model, achieving strong performance and interpretability. It constructs three task-specific datasets from multiple public sources and evaluates a wide range of representations, including OpenAI embeddings and LLM prompts. The key finding is that combining 15 handcrafted linguistic features with embeddings yields the best results (e.g., $F1$ around $0.91$ and AUC around $0.969$), outperforming strong baselines and prompting approaches. The work contributes to reproducibility and provides interpretable cues that illuminate what linguistic signals drive clickbait predictions, with potential practical impact for warning systems and content moderation tools.

Abstract

Clickbait headlines degrade the quality of online information and undermine user trust. We present a hybrid approach to clickbait detection that combines transformer-based text embeddings with linguistically motivated informativeness features. Using natural language processing techniques, we evaluate classical vectorizers, word embedding baselines, and large language model embeddings paired with tree-based classifiers. Our best-performing model, XGBoost over embeddings augmented with 15 explicit features, achieves an F1-score of 91\%, outperforming TF-IDF, Word2Vec, GloVe, LLM prompt based classification, and feature-only baselines. The proposed feature set enhances interpretability by highlighting salient linguistic cues such as second-person pronouns, superlatives, numerals, and attention-oriented punctuation, enabling transparent and well-calibrated clickbait predictions. We release code and trained models to support reproducible research.

Click it or Leave it: Detecting and Spoiling Clickbait with Informativeness Measures and Large Language Models

TL;DR

This paper tackles clickbait detection by integrating explicit informativeness measures with transformer-based embeddings in a hybrid XGBoost model, achieving strong performance and interpretability. It constructs three task-specific datasets from multiple public sources and evaluates a wide range of representations, including OpenAI embeddings and LLM prompts. The key finding is that combining 15 handcrafted linguistic features with embeddings yields the best results (e.g., around and AUC around ), outperforming strong baselines and prompting approaches. The work contributes to reproducibility and provides interpretable cues that illuminate what linguistic signals drive clickbait predictions, with potential practical impact for warning systems and content moderation tools.

Abstract

Clickbait headlines degrade the quality of online information and undermine user trust. We present a hybrid approach to clickbait detection that combines transformer-based text embeddings with linguistically motivated informativeness features. Using natural language processing techniques, we evaluate classical vectorizers, word embedding baselines, and large language model embeddings paired with tree-based classifiers. Our best-performing model, XGBoost over embeddings augmented with 15 explicit features, achieves an F1-score of 91\%, outperforming TF-IDF, Word2Vec, GloVe, LLM prompt based classification, and feature-only baselines. The proposed feature set enhances interpretability by highlighting salient linguistic cues such as second-person pronouns, superlatives, numerals, and attention-oriented punctuation, enabling transparent and well-calibrated clickbait predictions. We release code and trained models to support reproducible research.
Paper Structure (33 sections, 8 equations, 2 figures, 5 tables)

This paper contains 33 sections, 8 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Literature review
  3. Definition and characteristics of clickbait
  4. State of the art in clickbait detection
  5. Research gap
  6. The proposed model
  7. Task definition
  8. Constructed dataset
  9. Data sources
  10. Our derived datasets
  11. Our approach
  12. Informativeness measures
  13. Clickbait detection
  14. Experimentation
  15. Experimental environment setting
  16. ...and 18 more sections

Figures (2)

  • Figure 1: Clickbait vs. non-clickbait headline examples.
  • Figure 2: Overview of the experimental workflow. After dataset collection and preprocessing, embeddings and features are generated using TF-IDF, Word2Vec, TF-IDF-weighted Word2Vec, GloVe, and transformer-based models (BERT and OpenAI). These representations are then fed into clickbait detection models based on Random Forest and XGBoost to obtain the final prediction.