Skip to Main Content
In recent years, the explicit semantic analysis (ESA) method has got a good performance in computing semantic relatedness (SR). However, ESA method has failed to consider the given context of the word-pair, and generates the same semantic concepts for one word in different word-pairs. It canpsilat exactly determine the intended sense of an ambiguous word. In this paper, we propose an improved method for computing semantic relatedness. Our technique, the self-adaptive explicit semantic analysis (SAESA), is unique in that it generates corresponding concepts to express the intended meaning for the word, according to the different words being compared and the different context. Experimental results on WordSimilarity-353 benchmark dataset show that the proposed method are superior to those of existing methods, the correlation of computed result with human judgment has an improvement from r = 0.74 to 0.81.