Trust, Context and Sentiment model for Social Networks

My current research is constructing Trust-based model that uses context evaluation for preserving privacy in OSN (Online Social Networks).

The model has a main purpose: To every data instance we analyze its proper reliable audience, and users that are less trustworthy in different contexts will be allotted from the data cycle.
The need for such a model can be seen in this figure, where the user would have not wanted that the vice-principal of his workplace would see this post:

We can see an example for such a set of access granting for certain data instances in this figure, where there is a Minimum Trust Value of a certain category of a data instance is presented as MTVk.
Three out of four users hold the necessary trust value (UTVk) for Politics, thus have access to it, while only one user hold the necessary UTVk for Sales and has access to it:

We have conducted an experimental evaluation that gives us a comprehensive view of the user's subjective trust value, for the purpose of privacy preservation in the Ego network. The less trusted users in different contexts are the ones that the Ego user will most likely to prefer not to show his data in certain categories.

In our previous work, we have created an OSN security model that is composed of three main phases addressing three of its major aspects: trust, role-based access control and information flow, by creating an Information Flow-Control model for adversary detection , or a trustworthy network.
This research extends the basic Trust model and make an important separation for different types of data instances, that differ by their subject's category.
For example, a political post might be more sensitive for its publisher than a simple "Good morning everyone". Another aspect that is affected from this extension is the users themselves.
The OSN user's friends are not homogenic by nature and accommodate different perspectives and views, and accordingly, to the user himself, are trustworthy in several levels. Some are considered close friends, some just acquaintances, or even less.
We use Sentiment Analysis for assessing the effect of different sentiments of posts in Social Media - on the user's trust in a specific context.

The full report for the project is here: Project report
The datasets are here: Dataset Part 1 ; Dataset A Part 2 ; Dataset B Part 2
The Python software is here: GitHub link
Our previous research papers on the model are here: [1] ; [2] ; [3] ; [4]