On the effectiveness of feature set augmentation using clusters of word embeddings

Word clusters have been empirically shown to o‚er important performance improvements on various tasks. Despite their importance, their incorporation in the standard pipeline of feature engineering relies more on a trial-and-error procedure where one evaluates several hyper-parameters, like the number of clusters to be used. In order to beŠer understand the role of such features we systematically evaluate their e‚ect on four tasks, those of named entity segmentation and classi€cation as well as, those of €ve-point sentiment classi€cation and quanti€cation. Our results strongly suggest that cluster membership features improve the performance.