Community identification and word polarization.

Slides available here

Social networks are often blamed for facilitating the formation of communities of users that share the same ideology on some topic (e.g. politics) and have little exposure to alternative opinions. This renders the partecipating users prone to spreading misinformation, since the characterization of the network sustains the propagation of false rumors without barriers. These communities are often oblivious to, or openly hostile towards, alternative perspectives.

What we wanted to do is:

1. Programmatically find users' communities
2. Check whether or not they result to be polarized on a specific view for the "climate change" selected topic
3. Apply Word2Vec to detect "deviated" words, which are words that are used in different context when switching from a community to the other. Some of them are reported in the second-last slide

• Download dataset here of tweets related to climate change topic

• Build user graph, assigning a (undirected weighted) link between users following policies like number of retweets between them, comments, same links shared, etc...

• Programmatically find graph communities through Louvain's method

• Check communities polarization by looking at the tweets by users partecipating in the communities. For comparison, some tweets are reported in the second-last slide

• Check word polarization between communities using Word2Vec

This project was partially inspired by the paper "Mapping the echo-chamber: detecting and characterizing partisan networks on Twitter", by Armineh Nourbakhsh, Xiaomo Liu, Quanzhi Li, Sameema Shah.