Research On Object Level Retrieval In Enterprise Search

Enterprise search is one of the most valuable domains in Information Retrieval area, which aims at good knowledge management and information organization in a relatively small dataset to improve the leadership capacity of the managers and the efficiency of the staff. Enterprise search is a synthetically subject that gives a chance to combine various technologies, such as text retrieval, information extraction, natural language processing etc. Also, enterprise search is an advancing subject that arouses the study of more accurate information retrieval and knowledge representation in higher hierarchy.As more and more attention on enterprise search, the Text Retrieval Conference opened an enterprise track, in which a platform is provided for research on two subtasks: expert search and email search. In this dissertation, research on object level retrieval is deployed based on these two tasks to investigate the problems which include the development of processing algorithms for heterogeneous data and improving the performance of information retrieval techniques. The main contributions of this dissertation are summarized as follows:Firstly, we discussed the expert search model that applies text retrieval techniques. At the beginning, we studied the rule-based expert location method. For some experts who share same names, heuristic rules were developed for name disambiguation. As following, the language modeling techniques for expert search were discussed in a hierachical way. At last, we presented a framework of expert search using relevance feedback mechanism. In this framework, expert search is viewed as a relevance feedback process so that we can make full use of the techniques in such fruitful research field. Experimental result proves that these Easy-to-Use methods can not only get integrated with text retrieval techniques seamlessly, but also achieve high processing speed under a good accuracy.Secondly, we discussed the object-oriented modeling methods in enterprise search. In the first phase, the definition of object was given. For both expert search and email search, we defined the attributes and the relations of an object. In addition, we discussed the application of information extraction techniques for mining attributes and relations. Especially for expert objects, Textual Experience Atom was proposed as a kind of attribute, which extends the concept of Experience Atom in programming to a semantic level. Next, we discussed the models of email object. At last, we discussed the models of expert object and presented an Odds Ratio method in the text vector model of expert objects. Experimental result proves that this method can accommodate the bias of searching frequently-occurred experts and improve the accuracy of expert search significantly.At last, we investigated using object-based methods for enterprise search. We proposed an object-based retrieval model for ranking objects in object level retrieval. This model, through our theoretical analysis of ranking algorithm, is composed of similarity submodel, confidence submodel and importance submodel. In this foundation, both models for object-based expert search and object-based email search were formulated. Lastly, we extended this model to search multiple types of objects simultaneously. Experimental result proves that these methods are much fit for object level retrieval.