Machine Learning and Information Retrievel

摘要

From the early days of information retrieval (IR), it was realized that to be effective in terms of locating the relevant text, systems had to be designed to be responsive to individual requirements and interpretations of topics. In addition, the emphasis on statistical models of retrieval led to approaches that were domain-independent but collection-dependent. The concept of relevance feedback, where user judgments of the relevance of retrieved documents are used to update the original query, is a simple, collection-dependent and statistical approach to learning that has been studied for more than 30 years. We still, however, have much to learn about how to use machine learning effectively for IR, and the challenges of new applications such as distributed IR (i.e. the World Wide Web), routing, extraction, visualization, catego-rization, and discovery provide exciting opportunities to apply what has been learned in this field.