Relevance Feedback is an important technique to improve result quality in text retrieval and has been recently successfully applied for XML retrieval, and several feedback algorithms have been proposed. The comparison of them, however, is still an unsolved problem. Even though some evaluation methods have been proposed in the literature, it is still unclear which of them can be applied in the XML context, and which of them can be combined with the plethora of metrics proposed to assess the quality of retrieval algorithms. Within the INEX evaluation initiative for XML retrieval, the Relevance Feedback Task aims at comparing different feedback algorithms.

Within the context of the INEX 2006 Relevance Feedback Task, this thesis should examine at least the following aspects of the evaluation of relevance feedback algorithms for XML:

1. Definition of different evaluation methods, formal analysis of the methods (e.g., determine "optimal" results for each method, examine the interaction of the methods with different evaluation metrics, etc.)
2. Implementation of an evaluation framework for feedback algorithms that supports different evaluation methods and metrics (especially those provided by INEX)
3. Experimental comparison of the performance of several feedback algorithms for keyword-based and structural XML queries, using several evaluation methods and metrics, based (among others) the submissions of the participants of the INEX 2006 Relevance Feedback Task
4. Statistical analysis of the results from the INEX 2006 Relevance Feedback Task (e.g., statistical significance of results, characteristics of topics where feedback was very efficient or very inefficient, etc.)
5. Optionally, engine-independent comparison of feedback algorithms
   

This thesis requires a good knowledge of Java. A thorough knowledge of IR in general and especially XML retrieval and feedback techniques is helpful.