Data Fusion: Boosting Performance in Keyword Extraction

In a time when volatile data is in constant growth, the importance of keyword extraction becomes particularly evident. Keywords can quickly identify, structure and reveal potentially worthwhile information. The quality of automatically extracted keywords reflects the individual characteristics of the various retrieval approaches that may be used for extraction. A combinatorial approach using multiple heuristic keyword extraction algorithms may enhance the quality of the results significantly, though it may also compound the inherent limitations. In our paper we compare different ranking aggregation and data fusion methods for single documents. Furthermore we apply principal component analysis to determine an optimal selection of retrieval algorithms for combination with respect to the use-case. To validate our approach, we provide a statistical evaluation with real-world examples.