It is generally believed that the degree of a relation between two entities is likely to be stronger if they co-occur more often in the literature. Based on this assumption, several methods are used in biomedical text mining such as support, confidence, chi-square, odds ratio, lift, all-confidence, coherence, and pof. Comparing these eight methods, our work aims to find the best one. Also, we present a consensus approach that can further improve the performance. Experimental results on prioritising drug targets have shown that pof, coherence, and all-confidence in sequence are the top three. By integrating coherence into pof, the consensus method is the best one among all compared methods.
|Number of pages||18|
|Journal||International Journal of Data Mining and Bioinformatics|
|Publication status||Published - 01 Jul 2012|