Short Text Representation via Adaptive Weighting with Two Views

This paper proposes a short text representation method via adaptive weighting with two views. Two views are established and integrated for training texts for representation. With two representations of the training set, the improved adaptive two-view weighting clustering algorithm is designed to cluster texts, view weights and the attribute weights can be obtained respectively, based on which the obtained cluster centers are utilized to represent the feature space. A weighted similarity calculation method utilizing two types of weights of clustered word vectors is established to calculate the similarity between the terms of the short text to be represented and the feature terms in the feature space. Thereafter the text mapping matrix is constructed for short text representation. The experimental results reveal that our method has a remarkable effect on representing short text.