OSN Privacy 2.0: Group Decision Making via Trust-Weighted EIOWA

    ## TL;DR
    Managing privacy for co-owned data (like a photo featuring multiple people) is a major pain point in OSNs. This paper proposes a consensus-based framework using the **Extended Induced Ordered Weighted Average (EIOWA)** operator. By leveraging **dynamic trust values** as weights for users' linguistic preferences, the system helps groups reach a collective decision on whether to share sensitive content, effectively reducing privacy leakage.

    ## The Motivation: When "My" Photo is Also "Yours"
    In the current landscape of Online Social Networks (OSNs), data ownership is often treated as a solo act. If you upload a photo, you decide the privacy settings—even if five other friends are tagged in it. This "uploader-takes-all" approach is a breeding ground for privacy disputes and broken friendships.

    The authors argue that co-owned data requires a **Group Decision Making (GDM)** approach. However, the challenge lies in how to weight each person's opinion. Should a stranger's vote count as much as a best friend's? Probably not. This is where **Trust-based SNGDM** enters the picture.

    ## Methodology: EIOWA Meets Interpersonal Trust
    The core of the proposed solution is the **EIOWA operator**. Unlike simple averaging, EIOWA allows for an "inducing" variable—in this case, **Trust ($\lambda$)**—to dictate the importance of each user's linguistic preference.

    ### The Workflow
    1. **Preference Collection**: Co-owners provide preferences in linguistic terms (e.g., "Very Worried" to "Extremely Agree").
    2. **Trust Weighting**: The system retrieves trust values between the owner and DMs from a dynamic database.
    3. **EIOWA Aggregation**: The operator fuses individual preferences into a collective preference degree.
    4. **Consensus Check**: If the distance between ranked alternatives is too large or inconsistent, the system triggers a **Consensus Reaching Process (CRP)**, asking users to revise their opinions.

    ![Structure of the round for consensus reached decision](https://cdn.atominnolab.com/wisdoc/images/20260521-0a33fd5f-fb76-4df5-85b9-59ce40beceb8/page_001_block_008.png)
    *Fig 1: The iterative loop ensuring all stakeholders reach a stable agreement.*

    ## Case Study: The "Trusty" Social Network
    The authors tested their framework on **Trusty**, a prototype OSN. They modeled a scenario with four sharing alternatives:
    - $x_1$: Share with full permission
    - $x_2$: Share with restrictions
    - $x_3$: Share with no permission
    - $x_4$: Do not share

    In the first round of the experiment, the results were inconsistent. The distance between "Share with permission" and "Share with restrictions" was high, indicating a lack of consensus.

    ![Ranked aggregated alternative choices](https://cdn.atominnolab.com/wisdoc/images/20260521-0a33fd5f-fb76-4df5-85b9-59ce40beceb8/page_004_block_013.png)
    *Fig 2: Initial round results showing a lack of clear group consensus.*

    After the system prompted co-owners to reconsider in Round 2, the preference shifted significantly toward **"Do Not Share" ($x_4$)** with a much higher confidence value (1.02), showcasing how the CRP mechanism successfully protects sensitive data when users realize the collective concern.

    ## Critical Analysis & Conclusion
    The brilliance of this work lies in the **dynamic nature of trust**. Trust values are not static; they increase if the data owner respects the group's decision and decrease if they ignore it. This creates a self-regulating ecosystem that incentivizes "good digital citizenship."

    ### Limitations
    - **Scalability**: The paper demonstrates the model with 5-6 users. In a massive group chat/tagging scenario, reaching consensus through multiple rounds might lead to "notification fatigue."
    - **Incomplete Information**: The model assumes all co-owners provide feedback. Future work should address how to handle non-responsive users (e.g., through trust-based delegation).

    ### Final Takeaway
    This paper successfully bridges the gap between abstract fuzzy logic (EIOWA) and practical social problems. By making privacy a collaborative, trust-weighted conversation rather than a unilateral checkbox, we can build social platforms that actually respect the complex web of human relationships.