Construction of High-quality Feature Extension Mode Library for Chinese Short-text Classification
Xinghua Fan, Hongge Hu · 2010
One feasible way to handle the difficulties in the Chinese short-text classification is doing feature extension for its content. And a key technique of it is the construction of feature extension mode library whose quality would directly influence the effect of feature extension. This paper proposes a method of constructing a high-quality feature extension mode library. In our method, firstly, we obtained the feature extension modes from the training set, and then introduced three measures, i.e., confidence, category homoplasy and relevancy strength of the obtained feature extension mode, to improve their quality. Finally, we used the high-quality feature extension modes to construct a high-quality feature extension mode library. Experimental results show that (1) Introducing the three measures can improve the quality of feature extension mode in some degree respectively; (2)A high-quality feature extension mode library is helpful to improving the effect of Chinese short-text classification.