Semantic Cluster Analysis in Information Retrieval

点击翻译按钮获取中文摘要

Like its predecessor CAIR I, this proposal for a follow-up project will address research issues in the area of cluster analysis in information retrieval. In CAIR I, we started from the observation that the potential for optimizing and adapting clustering methods in IR has not been fully exploited yet, which leads to unsatisfying solutions for relevant retrieval problems. In order to fill this gap, following the general idea of semantic cluster analysis, CAIR I improved the theoretic basis for document clustering (inspired by the probabilistic IR model), investigated specialized retrieval models (based on models for indexing and representation), developed new strategies and algorithms for web applications and carried out corresponding user studies. While CAIR I worked on enriching the semantics in clustering processes via retrieval models, domain knowledge and the combined application of fusion principles, this proposal for a follow-up project aims at extracting knowledge from the clustering process. Besides the further development of the theoretic foundations, this proposal will focus on the user interface to clustering processes ("human access"). Our goal is to make the IR clustering process more understandable and ease the interpretation of its results for the user. The theoretic part will deal with soft clustering and online clustering. In the area of semantics and cluster interpretation, starting from current research results, we will address research issues such as faceted clustering, semi-supervised clustering and the evaluation of clustering results.