Abstract
Although taste and trust are concepts on clearly distinct ontological levels, they are strongly interrelated in several contexts. For instance, when assessing trust through a trust network, it is important to understand the role that personal taste plays in order to correctly interpret potential value-dependent trust recommendations. This paper aims at exploring the relationship between taste and trust in the analysis of semantic trust networks.
Introduction
Most trust management systems treat trust as a monolithic, objective concept. However, in real-world social interactions, trust is often colored by personal preference. If a friend recommends a movie you dislike, your trust in their character remains intact; you simply recognize a difference in taste. Conversely, if they recommend a demonstrably incompetent mechanic, you may begin to question their judgment or integrity. This paper addresses the challenge of modeling these nuances by distinguishing between "taste-dependent" opinions and "taste-independent" facts.
The Taxonomy of Trust and Taste
A fundamental contribution of this work is the categorization of trust into functional and referral types, further distinguished by the domain of application.
- Functional Trust: The subjective probability that an entity will perform a specific, objective function (e.g., fixing a car).
- Functional Taste: A subjective preference or evaluation of an entity's output in a value-dependent domain (e.g., music or art).
- Referral Trust: The willingness to accept a recommendation from another party regarding their functional trust or taste in a third entity.
By separating these concepts, we can model transitive trust paths more accurately. For example, Alice may trust Bob's referral, which leads her to consider Bob's functional taste in music or his functional trust in a car service.
Semantic Trust Networks
The authors propose generalizing trust networks into "Semantic Opinion Networks." In this framework, "trust" and "taste" are treated as subcategories of a more general cognitive concept: the Opinion. This allows for the application of subjective logic, representing opinions through parameters of belief, disbelief, and uncertainty.
This semantic approach enables Knowledge Federation, where a system can maintain a shared view of entities while allowing for dissenting subjective opinions. Instead of forcing a single "unified" reputation score, the network can account for "preference cliques"—groups of users who share similar values—and weight recommendations accordingly.
Key Insights: The "Agree to Disagree" Rule
The paper's most intuitive insight is the "Agree to Disagree" rule. People can maintain mutual acceptance despite differing opinions on taste-dependent aspects (like art). However, they cannot "agree to disagree" on taste-independent aspects (like whether a service was performed correctly) without a breakdown in the trust relationship or a perception of irrationality.
To support this, the authors suggest a Value Matrix attached to resources, storing evaluative data across various criteria without pre-judging them, allowing users to apply their own tastes to the raw data.
Key Results and Formalisms
The paper provides formal definitions for:
- Functional Opinion Derivation: A rule stating that for a recommendation chain to be valid, the final edge must be functional while all preceding edges must be referrals.
- Opinion Scope Consistency: A criterion requiring that transitive paths share a common "subset" scope (e.g., trust in "car mechanics" is a valid subset of trust in "changing tires").
Limitations and Future Work
While the theoretical framework is robust, the paper lacks empirical validation through simulations or real-world datasets. Furthermore, the computational challenge of determining the "largest common subset" of scopes in massive, automated networks remains an open problem. Future work will focus on developing the technical tools for Knowledge Federation and the practical implementation of the Value Matrix.
Summary updated