Clickbait detection: quick inference with maximum impact

Soveatin Kuntur; Panggih Kusuma Ningrum; Anna Wróblewska; Maria Ganzha; Marcin Paprzycki

cs.CLApr 9, 2026

Clickbait detection: quick inference with maximum impact

Authors: Soveatin Kuntur, Panggih Kusuma Ningrum, Anna Wróblewska, Maria Ganzha, Marcin Paprzycki

Organizations: Faculty of Mathematics and Information Science, Warsaw University of Technology, plac Politechniki 1, 00-661 Warsaw, Poland · Université Marie et Louis Pasteur, CRIT, F-25000 Besançon, France · Systems Research Institute, Polish Academy of Sciences, Newelska 6, 01-447 Warsaw, Poland

Abstract

We propose a lightweight hybrid approach to clickbait detection that combines OpenAI semantic embeddings with six compact heuristic features capturing stylistic and informational cues. To improve efficiency, embeddings are reduced using PCA and evaluated with XGBoost, GraphSAGE, and GCN classifiers. While the simplified feature design yields slightly lower F1-scores, graph-based models achieve competitive performance with substantially reduced inference time. High ROC--AUC values further indicate strong discrimination capability, supporting reliable detection of clickbait headlines under varying decision thresholds.

Figures & tables

Clickbait headline	Non-clickbait headline
“6 Things Most Doctors Won’t Tell You About Dieting” “This Simple Investment Trick Could Make You Rich Overnight”	“Aryna Sabalenka survives the Wimbledon heat: World No. 1 into semi-finals with rollercoaster comeback to beat Laura Siegemund, 37, despite 23C temperatures”

Table 1: Examples of clickbait and non-clickbait headlines. Clickbait headlines are typically short and designed to induce a curiosity gap, whereas non-clickbait headlines tend to be longer and provide explicit informational content.

Headline	Baitness	Informativeness
“You Won’t Believe What This Celebrity Did!”	0.82	0.18
“Study Finds 23% Increase in Solar Adoption Across Europe”	0.21	0.79

Table 2: Illustrative examples of baitness and informativeness scores. Clickbait headlines exhibit high baitness and low informativeness, while factual headlines show the opposite pattern.

Feature Representation	Classifier	F1-score	ROC-AUC	Inference
				time (ms)
Our proposed hybrid approach
OpenAI Emb. (1000d) + Baitness + Informativeness	XGBoost	0.8465	0.9330	236.65
OpenAI Emb. (1000d) + Baitness + Informativeness	GraphSAGE	0.8572	0.9356	177.79
OpenAI Emb. (1000d) + Baitness + Informativeness	GCN	0.8382	0.9219	98.47

Table 3: Performance and inference time comparison of clickbait detection models. Inference time is reported per sample.

Figure 1: ROC curves for hybrid clickbait detection models using semantic embeddings and baitness features. Graph-based models achieve competitive discrimination performance compared to XGBoost.

Click it or Leave it: Detecting and Spoiling Clickbait with Informativeness Measures and Large Language Models
Feb 20, 2026Wojciech Michaluk, Tymoteusz Urban, Mateusz Kubita +2
Faculty of Mathematics and Information Science, Warsaw University of Technology
ClickGuard: A Trustworthy Adaptive Fusion Framework for Clickbait Detection
Apr 8, 2026Chhavi Dhiman, Naman Chawla, Riya Dhami +2
Department of Electronics and Communication Engineering, Delhi Technological University, Delhi, India · Department of Computer Science & Engineering, Bhagwan Parshuram Institute of Technology, Delhi, India · Department of Electronics & Communication Engineering, Bhagwan Parshuram Institute of Technology, Delhi, India +1
BanClickThumb: A Multimodal Dataset and Transformer Fusion Benchmarks for Clickbait Detection in Bengali YouTube Videos
Jul 19, 2026Md. Ariful Islam, Md Tanvirul Islam, Md. Maruf Hossain Miru +1Bengali NLP Multimodal Classification
Department of Computer Science and Engineering, Bangladesh Army University of Science and Technology, Saidpur Cantonment, Saidpur, Bangladesh

Feb 20, 2026J/K move · Enter open · S save

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

Wojciech Michaluk, Tymoteusz Urban, Mateusz Kubita, Soveatin Kuntur+1

Faculty of Mathematics and Information Science, Warsaw University of Technology

Open paper AskPDF arXiv

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.

Abstract

Explore similar work

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

ClickGuard: A Trustworthy Adaptive Fusion Framework for Clickbait Detection

BanClickThumb: A Multimodal Dataset and Transformer Fusion Benchmarks for Clickbait Detection in Bengali YouTube Videos