Social applications for mobile phones called Social Live Stream Services (SLSS) have more and more users. These services allow sharing at any time live material in a way and on a scale without precedent.

One of the main challenges facing these services is how to detect and prevent malicious behaviour, such as, for example, publishing explicitly pornographic content. Although there are mechanisms to prevent these bad practices, in reality the users themselves in their assessments warn of some bad experiences.

By processing large amounts of content on a large scale the process for the detection of such behaviour can be easily improved and automated

Vicenç Gómez, Ramón y Cajal researcher with the Department of Information and Communication Technologies (DTIC) at UPF, together with Nikos Lykousas and Constantinos Patsakis, researchers from the University of Piraeus (Greece), are to present the results of a study at The 2018 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM 2018) to be held in Barcelona from 28 to 31 August, for which Nikos Lykousas, first author of the study, has analysed the traffic of two such SLSS servers, specifically Live.me and Loops Live, which are visited by millions of users and produce massive amounts of video content on a daily basis.

Only a very small proportion of users are “producers” of pornographic material in the services studied: Live.me and Loops.live.

“In this study we have focused on the production and consumption of pornographic content. To do so, first we filtered out user names that contained pornographic terms with the help of a large database. Then, we used computational techniques of deep learning to analyse and classify whether the images of their videos (publicly available) also contained this type of content. Finally, we analysed the network of social interactions of this community of users”, explains Vicenç Gómez.

One of the main results of the work is that this kind of behaviour is really exceptional. Only a very small proportion of users are “producers” of pornographic material in the two services studied: Live.me and Loops.live.

In one of the SLSS studied there are many users who would be classified as “producers” of pornographic material but they have not been blocked by the platform

Another important result is the relationship between the predictions of the classifier that the researchers have used and the fact that a certain user has been blocked by the platform or not. “We can see this thanks to the fact that the information on whether a user has been blocked for malpractice or not is accessible on the platform”, says Gómez. Here the authors of the study have seen that while one of the services studied is fairly consistent, the other is not. More specifically, in one of the SLSS studied there are many users who would be classified as “producers” of pornographic material but they have not been blocked by the platform.

The authors have also revealed that the “producers” are very isolated, that is to say, neither do they follow each other and nor are they followed by “other producers”, while the “consumers” have the opposite behaviour: they follow each other a lot and are followed by other “consumers”.

Through this work the researchers have demonstrated that by processing a large amount of content on a large scale, the process to detect this type of behaviour can be improved and automated easily (using public tools). In addition to presenting the work at the ASONAM congress, the authors are planning to publish the details of the work duly anonymized in a repository so that they are available to the scientific community and so that other research groups can analyse them.

Reference work:

Nikolaos Lykousas, Vicenç Gomez , Constantinos Patsakis (2018), “Adult content in Social Live Streaming Services: Characterizing deviant users and relationships”, communication at The 2018 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM 2018), Barcelona, 28-31 August.