Text clustering is a big challenge in the text mining field; traditional algorithms are powerless when dealing with short texts. Short messages are a much more flexible form of data in social media, containing not only textual information, but also comment, time and regional information. We propose an algorithm to extract semantic and multidimensional feature representation from such texts. In particular, by using the fact that comments are semantically related to the short message, we can get the supervised information and train the text representation, with which we transform the problem into a semi-supervised problem. We use a convolutional-pooling structure that aims at mapping the text into a semantic representation. What’s more, we expand the semantic representation with time- and region-related features, leading to a much more flexible and strong representation for short messages. Our approach shows great advantages in labelled data over traditional feature representation methods and performs better than other clustering methods via deep neural network representation.
CITATION STYLE
Wu, S., Zhang, H., Xu, C., & Guo, T. (2019). Text clustering on short message by using deep semantic representation. In Advances in Intelligent Systems and Computing (Vol. 760, pp. 133–145). Springer Verlag. https://doi.org/10.1007/978-981-13-0344-9_11
Mendeley helps you to discover research relevant for your work.