5.2 Expert Search Models All systems mentioned up to now use different ad-hoc techniques but do not formally define retrieval models for experts. Some first steps in this direction have been made: probabilistic models [18 and language models [1-3 have been proposed. Another model for expert search proposed in 23 views this task as a voting problem. The documents associated to a candidate are viewed as votes for this candidate's expertise. In 24 the same authors extended the model including elevance feedback techniques, which is an orthogonal issue. More recently, focu has been put on finding high quality relationships between documents and people and evidence of expertise [ 28, 22, 6 5.3 Personal Information Management Systems a lot of research have been also done in the field of pim. the most relevant is the one of desktop search. Finding items desktop is not the sam task as finding documents on the web. Several commercial systems have been proposed(e. g, Google, Yahoo!, Microsoft ) Our expert finding system builds on top of the Beagle++ system: a semantic desktop search engine 8. Beagle++ exploits the implicit semantic information residing at the desktop level in order to enhance desktop search. Moreover, it creates metadata annotations, thanks to its extractors, that can be reused by our expert finding system One important issue in the field of PIM is the evaluation of retrieval effe tiveness. Retrieval systems are usually evaluated using standard testbeds(e. g. TREC). In PIM such testbeds are not available mainly because of the privacy issues of sharing personal data. A way to overcome this problem is to create small collections internally to each research group 11] The Nepomuk project aims at developing a framework for the Social Seman tic Desktop. Our expert finding system is integrated in the Nepomuk system providing the user of the semantic desktop this additional search functionality If we want to find experts on the desktop then a crucial task is to extract people names out of full text. Many techniques have been proposed and can be reused for this step. Possible solutions to the problem of measuring similarity between two named entities are presented in 14, how to pre-process a document collection in order to extract names from documents such as e-mail has been proposed in [10 6 Conclusions and Future work In this paper we presented a system for finding experts on the semantic desktop The approach works as follow. The desktop content is first indexed: metadata extracted and an RDF repository is built with information about persons and documents. Then, a vector space containing candidate experts and documents is created by exploiting the relations existing between them. Once the documents as well as the candidates are placed into the vector space, a query vector can be placed into the space and a ranked list of experts can be obtained using a http://trec.nist.gov5.2 Expert Search Models All systems mentioned up to now use different ad-hoc techniques but do not formally define retrieval models for experts. Some first steps in this direction have been made: probabilistic models [18] and language models [1–3] have been proposed. Another model for expert search proposed in [23] views this task as a voting problem. The documents associated to a candidate are viewed as votes for this candidate’s expertise. In [24] the same authors extended the model including relevance feedback techniques, which is an orthogonal issue. More recently, focus has been put on finding high quality relationships between documents and people and evidence of expertise [28, 22, 6]. 5.3 Personal Information Management Systems A lot of research have been also done in the field of PIM. The most relevant area is the one of desktop search. Finding items on the desktop is not the same task as finding documents on the web. Several commercial systems have been proposed (e.g., Google, Yahoo!, Microsoft). Our expert finding system builds on top of the Beagle++ system: a semantic desktop search engine [8]. Beagle++ exploits the implicit semantic information residing at the desktop level in order to enhance desktop search. Moreover, it creates metadata annotations, thanks to its extractors, that can be reused by our expert finding system. One important issue in the field of PIM is the evaluation of retrieval effec￾tiveness. Retrieval systems are usually evaluated using standard testbeds (e.g., TREC10). In PIM such testbeds are not available mainly because of the privacy issues of sharing personal data. A way to overcome this problem is to create small collections internally to each research group [11]. The Nepomuk project aims at developing a framework for the Social Seman￾tic Desktop. Our expert finding system is integrated in the Nepomuk system providing the user of the semantic desktop this additional search functionality. If we want to find experts on the desktop, then a crucial task is to extract people names out of full text. Many techniques have been proposed and can be reused for this step. Possible solutions to the problem of measuring similarity between two named entities are presented in [14], how to pre-process a document collection in order to extract names from documents such as e-mail has been proposed in [10]. 6 Conclusions and Future Work In this paper we presented a system for finding experts on the semantic desktop. The approach works as follow. The desktop content is first indexed: metadata is extracted and an RDF repository is built with information about persons and documents. Then, a vector space containing candidate experts and documents is created by exploiting the relations existing between them. Once the documents as well as the candidates are placed into the vector space, a query vector can be placed into the space and a ranked list of experts can be obtained using a 10 http://trec.nist.gov