Our experiments use two public datasets crawled from Instagram and Vine. The datasets were respectively introduced and released in [1] and [2]. Please reach out to the wonderful CU Boulder team for the datasets.
The CU Boulder team provides 3 csv files for the Instagram dataset -- sessions_0plus_to_10_metadata.csv, sessions_10plus_to_40_metadata.csv, and sessions_40plus_metadata.csv.
The CU Boulder team also provide the user "follows" and "followed" list in files Instagram_common_users.zip and Instagram_normal_users.zip.
Extract the network statistics and detailed network into the following 2 files, and place them in the data/ins/ folder.
Similarly, extract the session data vine_labeled_cyberbullying_data.csv from the Vine dataset, and place them in the data/vine/ folder.
We use word2vec-twitter for word embeddings.
We also provide an off-the-shelf pickle version in the follwing url:
'https://drive.google.com/file/d/1PqHeFg8QvXX_vGv54kgjRZK89XO6Gde0/view?usp=sharing', 'https://drive.google.com/file/d/1onEQw6yFAslUNlZeuviWp6_UyVyoB3KM/view?usp=sharing',
download them and place their path as w2vVocab_file and w2vVector_file respectively.
[1] Homa Hosseinmardi, Sabrina Arredondo Mattson, Rahat Ibn Rafiq, Richard Han, Qin Lv, and Shivakant Mishra. 2015. Analyzing labeled cyberbullying incidents on the instagram social network. In Socinfo. Springer, 49–66.
[2] Rahat Ibn Rafiq, Homa Hosseinmardi, Richard Han, Qin Lv, Shivakant Mishra, and Sabrina Arredondo Mattson. 2015. Careful what you share in six seconds: Detecting cyberbullying instances in Vine. In ASONAM. ACM, 617–622.