Research On Web Based Expert Search

Expert means someone with expertise. In the modern era of information explosion, it is a more and more important information need to search for experts within an enterprise or on the World Wide Web. However, the techniques and tools for expert search by now still cannot satisfy the need. There are already some work on expert search on the enterprise intranet, but a formal framework is still lacking in theory. Besides, these techniques are still not practical since the performance is not satisfying. While for expert search on the internet, only few researches have been done. To fill in the research gap, we propose to study the topic of Web based expert search.Particularly, we study the following sub-topics in this paper.Expert finding and identification is to automatically discover all the experts within a certain corpus, and identify all the appearances of the these experts. It is a unavoidable problem for building an expert search engine. In this paper we propose a method based on E-mail parsing to automatically obtain the expert list of a corpus, along with a highly efficient expert appearance identification algorithm.Expert Search Model is to effectively retrieve experts related to a certain query within a corpus, given the expert list of the corpus. In this paper we propose an evidence-based expert search model. In this model, the relevance of an expert and a query is dependent on the evidences extracted from the corpus. Experiments show that our expert search model can effectively retrieve relevant experts in a webpage corpus.To distinguish experts from normal people, we propose to use static ranking approaches to decide an expert's importance. We propose a ExpertRank algorithm based on link analysis. We also propose a Topic Sensitive ExpertRank algorithm to carry the idea further. Name ambiguity is very common on the World Wide Web. The search result of a person name usually involves several individuals. To perform expert search on the World Wide Web, this problem must be solved. In this paper, we propose a classification framework with feedback to approach the problem of disambiguation. Moreover, we propose two novel features for disambiguation: Key Token and Topic. Experiments show that the two features can improve the performance of disambiguation greatly. Besides, we also discuss the effects of different classification methods on disambiguation, and employed sampling methods to solve the robustness problem caused by data imbalance.