SVM-BASED RELEVANCE FEEDBACK DOCUMENT RETRIEVAL IN DIFFERENT REPRESENTATIONS OF DOCUMENT VECTORS

摘要

We propose an interactive method of document retrieval using active learning based on Support Vector Machines (SVMs). It uses heuristics to bias documents based on user assessments according to the distribution of examples in documents being retrieved. This heuristics is executed by selecting examples to find neighbors in positive support vectors, and it improves learning efficiency. The performance of SVMs depends on the vector representation of documents. We compare our proposed method with a Rocchio-based and an SVM-based system without heuristics in several representations of document vectors, and explain the difference in performances between vector representations.