This November, our team had the great opportunity of attending the 5th Symposium on Multidisciplinary International Symposium on Disinformation in Open Online Media or MISDOOM 2023. This occasion brought together a number of researchers from many differentiated areas which allows for the ideas to leave the sometimes closed environment of one research group, where we are dead focused on the latest specific method we are using for this very specific task. At least for me, coming from a computer science background, the open conversation with experts in anthropology or communication science can ground the research I do into real applications against disinformation.

Our attendance came with the acceptance of the extended abstract Controversy detection and automated characterization of polarized communities by compression distances, available on the booklet published by the organization. In this work we presented an approach to quantify how controversial a topic is on social media. We contrasted the use of two methods for the construction of an interaction graph. First, following the work done by Garimella et al, we used the use of the co-occurrence of hashtag by two users as a way of connecting them. Second, a connection between users based on the closeness of their speech by Normalized Compression Distance, a distance we had previously applied to other social media tasks.

Due to the elimination of the academic access to X (Previously Twitter) API, the method described by this abstract has not been tested on a corpus of broad common twitter users. However, for the purpose of presenting the work on the Symposium, we used a previously recorded corpus of Spanish politicians timelines. This corpus gave us a surprising hidden benefit: due to the tweets being associated with a politician we were able to divide it by their political party association.

What we found out with this small experiment is not far off from what we already knew. The use of NCD to categorize users from one same organization work amazingly. In any topic, controversial or not, the NCD method has a high overlap between the binary divided interaction graph, and the association of the twitter authors to a right leaning or left leaning political party.

However, as a measure of controversial topics, this measure seems to perform worse than the co-occurrence of hashtag use. The measure we use to quantify controversy gives less differentiated values between common hashtags about breaking news and controversial policies when measured on the NCD constructed graph.

Here we let public the slides used for the presentation done by our researcher Luis Pérez on MISDOOM 2023.