Research On Topic Detection And Classification In Short Text

With the rapid development of social network and the prevalence of smartphones,an increasingly large volumn of data are produced throughout the social network plat-forms like Sina Microblog,qq as well as speech assistant like Cortana and Siri.There are many similarities which they share in common.For example,the text lengths are short.Natural language processing algorithms has been more effctive on the long text such as articles and news on the task of topic word found and document classification.Because of the texts length are short and have less information,the texts update very fast,the challenges are:1)the traditional feature extraction methods such as one-hot,bag of words has got relatively sparse features,2)the large-scale data set always update,training model from scratch need a lot of time.This paper includes comprehensive and in-depth research in topic word found,domain classification and incremental learning in an anttempt to address the issues in short text processing and also we conducted experiments includeing short text topic word found,domain classification and incre-mental learning on domain classification using our dataset to validate its effctiveness.For short text topic word found using LDA topic decomposition to obtain hidden infor-mation which can rich feature,for domain classification to introduce slot entity which can rich feature too,for large-scale data training using incremental learning methods to reduce training time.Compared to traditional methods,our approach on three tasks are achieved better results...