Socialized Expert Finding Based On The Relationship Between Experts And The User

Expert search aims to help a user to access those experts on some specific topic with specific knowledge and techniques in intranet and Web. In the modern era of information, looking for an expert is also a key method to satisfy users'information need as well as looking for related documents which is done by document search.Traditional researches on expert finding generally focus on the degree of expertise and rank the experts according to that. However, in practice, the actual purpose is not to find the expert, but to collaborate with the expert in some way, such as communicating to get question answered or ask for advice, cooperating on work and so forth. Therefore, when we recommend experts for users, there are two aspects affecting whether users may get what they need, not only 1) the expertise, but also 2) the personal relationship with the expert. Studies show that human factors (e.g., users'social relationship, etc.) are important in the procedure of user looking for experts for collaboration. Unfortunately, existing expert finding models in information retrieval field only consider the absolute authority of expert candidates, while overlook the relationship between the user and the expert. In this paper we try to find appropriate expert for collaboration. We define two factors corresponding to the two aspects of the task that should be considered simultaneously: 1) expert authority; 2) closeness to user. We propose several models, which combine these two factors in different ways. One basic idea is to filter out expert candidates by one factor and rank them by the other; another basic model tries to combine factors using linear combination method; and friend recommendation model (FRM) is motivated by the procedure of finding experts in our daily life, which combines factors in natural friend recommendation process and formalizes it using probability and Markov process theory.Experiments are carried out in two different scenarios, one is that users look for coauthors in academic domain, while the other is that users look for answers in question and answer community (CQA).We model the social network of user in each scenario. And based on these models we then systematically test the performances of these three models. The experimental results show that different type of social network generates different influence in socialized expert search models and FRM is more efficient than the two basic models for socialized expert search.