Detection of Clickbait posts

The snapper clickbait detector. A framework for detecting clickbait posts using a two-level classification approach. A set of text-based features are extracted and combined for training 65 classifiers. The outputs of these classifiers are used as a second-level feature vector.

Getting Started

Requirements

1. JDK 1.7 or greater
2. Python modules
   • numpy
   • scipy
   • SocketServer
   • Scikit Learn
   • networkx
3. mongoDB

Usage

---

The framework performs feature extraction on a new post and then classify it as clickbait or not.

Define variables in src/main/resources/local.properties:

- resources: path where the resources are stored.
- stopClickbaitPython: path to python script executing the external feature extraction. (/resources/stop_clickbait_paper/mass_extractor.py)

A set of features is extracted using an external framework. Refer to https://github.com/bhargaviparanjape/clickbait for installation instructions. Replace the utility.py with /resources/stop_clickbait_paper/utility.py

Initialize parameters

- auth = true or false. Whether to use authentication for connecting to mongoDB. Default value = false;
	if auth = true define credentials in mongo.properties file (/src/main/resources)
- mainDB =  name of the main database where intermediate data will be stored.
- ProbCollection1stLevel = collection name where the probabilities that form the feature vector for the second level classification are stored.

Main class ClickbaitChallenge in gr.iti.mklab package. Provide command line arguments:

Input arguments:

	-i input_path: path of a text file (instances.jsonl) containing the post objects. Post file format (1 line per post):
		{
			"id": "608999590243741697",
			"postText": ["Some people are such food snobs"],
			"postMedia": ["608999590243741697.png"],
			"targetTitle": "Some people are such food snobs",
		}     

 	-o output_path: path where a file results.jsonl will be stored containing clickbait scores in the range [0,1], where a value of 1.0 denotes that a post is heavily click baiting. Output file format:
		{"id": "608999590243741697", "clickbaitScore": 1.0}

Citations

@article{papadopoulou2017two,
	title={A Two-Level Classification Approach for Detecting Clickbait Posts using Text-Based Features},
	author={Papadopoulou, Olga and Zampoglou, Markos and Papadopoulos, Symeon and Kompatsiaris, Ioannis},
	journal={arXiv preprint arXiv:1710.08528},
	year={2017}
}

Contact for further details

---

Olga Papadopoulou (olgapapa@iti.gr)
Symeon Papadopoulos (papadop@iti.gr)