Using Relevance Feedback in XML Retrieval

Abstract

Information retrieval has a long tradition: in the early days, the main focus was on the retrieval of plain text documents and on search systems for books and structured documents in (digital) libraries. Often, users were assisted by well-trained librarians or specialists to retrieve documents fitting their information need. With the proliferation of the internet, retrieval systems for further media types like images, video, audio and semi-structured documents have emerged. But more importantly, an ever increasing number of untrained users deploy retrieval systems to seek for information. Since most users lack a profound understanding of how retrieval engines work and of how to properly describe an information need, the retrieval quality is often not satisfactory due to bad query formulations. As an illustration of this, Jansen et al. [177] reported that 62% of queries submitted to the Excite web search engine consisted of less than three query terms. Obviously, this is by far insufficient to accurately describe an information need. But search systems often do not support users (or only rudimentary) to adjust their queries to improve retrieval effectiveness.