Politecnico di Torino (logo)

Exploring Abstract Concepts for Images Privacy Classification in Social Media

Gabriele Galfre'

Exploring Abstract Concepts for Images Privacy Classification in Social Media.

Rel. Enrico Magli, Cornelia Caragea. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Informatica (Computer Engineering), 2019

PDF (Tesi_di_laurea) - Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives.

Download (2MB) | Preview

The work presented in this dissertation focuses on the task of predicting the privacy class of images posted on social media. In particular, the results we are going to show aim at supporting the hypothesis that abstract concepts are better suited for capturing the private nature of content shared online. The definition of abstractness we adopted along this work refers to the idea of something that is elevated from anything concerning the sphere of perceptions, difficult to be appreciated through our senses or impossible to conceptualize as something even remotely physical. We developed a novel approach to investigate this hypothesis about abstractness in the context of a specific task. Specifically we applied this type of analysis on the textual user tags associated to a total of around 3 thousands selected images recently posted on Flickr. The privacy classification task we target is binary and consists in labeling posts as "public" or "private". In order to provide a solid foundation to our experimental setup, we evaluated the performances of different types of classification models, achieving results following a similar pattern. To the best of our knowledge we are the first facing this kind of investigation, trying to define some guidelines for the development of a methodology that could be applied to many different topics and used as proof for intensifying the focus of researchers toward concepts' abstractness. In the effort of expanding the initial resources about words' abstractness, our contribution dealt with the task of scoring terms by abstractness, evaluating several techniques. Our approach made use of a dataset and samples' representations never tried before. After extensive analysis the results have been used for scoring, as precisely as possible, a set of unlabeled words, exploited for further experimentation in the privacy prediction task. The results of this thesis' work are demonstrating the truthfulness of the hypothesis introduced, supporting it from different points of view. We conclude our analysis providing some insights about the directions the future works could follow starting from our conclusions.

Relators: Enrico Magli, Cornelia Caragea
Academic year: 2019/20
Publication type: Electronic
Number of Pages: 111
Corso di laurea: Corso di laurea magistrale in Ingegneria Informatica (Computer Engineering)
Classe di laurea: New organization > Master science > LM-32 - COMPUTER SYSTEMS ENGINEERING
Aziende collaboratrici: UNSPECIFIED
URI: http://webthesis.biblio.polito.it/id/eprint/12415
Modify record (reserved for operators) Modify record (reserved for operators)