Assessing Semantic Similarity Between Concepts Using Wikipedia Based on Nonlinear Fitting

Feature-based methods of semantic similarity with Wikipedia achieve fruitful performances on measuring the "likeness" between objects in many research fields. However, since Wikipedia is created and edited by volunteers around the world, the preciseness of these methods more or less are influenced by the incompleteness, invalidity and inconsistency of the knowledge in Wikipedia. Unfortunately, this problem has not got enough attention in the existing work. To address this issue, this paper proposes a novel feature-based method for semantic similarity, which has three parts: low frequency features removal, the similarities of generalized synonyms computing, and weighted feature-based methods based on nonlinear fitting. Moreover, we show that our new method can always get a better Pearson correlation coefficient on one or more benchmarks through a set of experimental evaluations.