Comparison of four approaches to term similarity measures. Semantic similarity vs. sequence similarity results using four distinct approaches to Resnik's measure: maximum (in red), average (in green), BMA (in blue) and BMA + GraSM (in violet). A - in the full dataset with the LRBS sequence similarity metric; B - in the non-electronic dataset with the LRBS metric; C - in the full dataset with the RRBS metric; D - in the non-electronic dataset with the RRBS metric. Modelling curves in A and C were composed of two additive normal cumulative distribution functions, and the curve for the average included also a negative linear component; in B and D, all curves were composed of a single normal function. It is noticeable that while all four approaches exhibit similar behaviour in the non-electronic dataset (B and D), the maximum and particularly the average approach perform poorly in the full dataset (A and C), with the former having a very low resolution and the latter showing a decreasing behaviour for high sequence similarity values. The same behaviours and the same relationships between the approaches were obtained for Lin's and Jiang and Conrath's measures.