Effectively Classifying Short Texts
via Improved Lexical Category
and Semantic Features
Huifang Ma
(&)
, Runan Zhou, Fang Liu, and Xiaoyong Lu
College of Computer Science and Engineering,
Northwest Normal University, Lanzhou, China
mahuifang@yeah.net
Abstract. Classification of short text is challenging due to its severe sparseness
and high dimension, which are typical characteristics of short text. In this paper,
we propose a novel approach to classify short texts based on both lexical and
semantic features. Firstly, the term dictionary is constructed by selecting lexical
features that are most representative words of a certain category, and then the
optimal topic distribution from the background knowledge repository is
extracted via Latent Dirichlet Allocation. The new feature for short text is
thereafter constructed. The experimental results show that our method achieved
significant quality enhancement in terms of short text classification.
Keywords: Short text classification
Latent Dirichlet allocation Lexical
features
Semantic features Optimal topic distribution
1 Introduction
With the rapid development of the social network, we are now dealing with much more
short texts in various applications. Examples are snippets in search results, tweets,
status updates, news comments, and reviews from various social platforms. There is an
urgent demand to interpret these short texts efficiently and effectively. However, short
texts do not provide sufficient word occurrences, traditional methods [1] such as
“Bag-Of-Words” fail to represent these short texts accurat ely due to their
high-dimension and severe sparsity. One crucial question, for conventional classifying
methods like support vector machine (SVM) [2]andk nearest neighbors (k-NN) [3], is
how to select representative features in such short documents. The characteristics of
short text will hinder the application of conventional machine learning and text mining
algorithms.
Most existing approaches have attempted to enrich short text to get more features,
which combine more semantic, contexts and associations. Existing work in the liter-
ature tried to address the aforementioned challenges from two directions. One was to
employ search engines and utilize the search results to expand related contextual
content [4, 5] while the other was to utilize external repositories as background
knowledge [6, 7]. These two types of methods can solve somewhat the problem, but
still leave much space for improvement. Both methods tried to obtain more sema ntic
© Springer International Publishing Switzerland 2016
D.-S. Huang et al. (Eds.): ICIC 2016, Part I, LNCS 9771, pp. 163–174, 2016.
DOI: 10.1007/978-3-319-42291-6_16
剩余11页未读，继续阅读
资源评论
