Title: Strategies for providing recommendations

Contact person: Tim Muller

Short description:
In interactions over the internet, users often rely on other agents.
To decide whether a user wants to rely on an agent, he establishes a trust opinion.
Often, the user not only use his own information, but he also gathers recommendations of other parties.
However, there is no guarantee that these recommendations are truthful.
Thus, the user tries to establish the actual opinion of the recommender.
The relationship between recommendations and actual opinions depends on the strategy of the recommenders.
Rigourous ways of establishing trust opinions using recommendations are parameterized on these strategies.
In other words, for a user to obtain an explicit trust opinion, he needs to model the strategy of the recommender.

The project is to analyse the possible strategies of recommenders.
Possible ways to analyse these strategies are:
- A theoretical analysis of the optimal way to obtain certain goals (using game theory).
- Upper and lower bounds on the effectiveness of strategies (using information theory or formal methods).
- A practical analysis, with implementations of certain strategies and comparisons (using any programming language, or the Canephora tool).
- A statistical analysis, using data publicly available from trust systems on the web.
- An approach based on machine learning or other AI techniques.
- Adapting strategies in the literature for similar issues to this formalism.

The scope of the project is well defined, but there is lots of freedom for the student to choose his methodology.
The challenges that can be found are varied, ranging from formal and mathematical problems to implementing solutions, via reasoning about effective strategies.
The student is invited (but not obligated) to implement strategies in the Canephora tool (in Java), which provides a layer of abstraction over some mathematical details.