Comparing Different Properties Involved in Word Similarity Extraction

In this paper, we will analyze the behavior of several parameters, namely type of contexts, similarity measures, and word space models, in the task of word similarity extraction from large corpora. The main objective of the paper will be to describe experiments comparing different extraction systems based on all possible combinations of these parameters. Special attention will be paid to the comparison between syntax-based contexts and windowing techniques, binary similarity metrics and more elaborate coefficients, as well as baseline word space models and Singular Value Decomposition strategies. The evaluation leads us to conclude that the combination of syntax-based contexts, binary similarity metrics, and a baseline word space model makes the extraction much more precise than other combinations with more elaborate metrics and complex models.