文本表示方法对微博Hashtag推荐影响研究——以Twitter上H7N9微博为例

The Impact of Document Representation on Hashtag Recommendation for Microblog——Use H7N9 Corpus on Twitter as Test Dataset

在总结国内外Hashtag推荐方法和短文本表示方法的基础上,文章利用基于K最近邻(KNN)的Hashtag推荐方法,将微博文本表示为向量然后计算相似度,从语料中选出与目标微博最相似的微博文本,然后抽取候选Hashtag。文章比较了向量空间模型(VSM)、潜在语义分析模型(LSA)、隐含狄利克雷分布模型(LDA)、深度学习(DL)等四种文本表示方法对基于KNN的Hashtag推荐效果的影响。以Twitter上H7N9微博为测试数据,实验结果表明深度学习的文本表示方法在基于KNN的Hashtag推荐中取得最好的效果。

According to the summary of various Hashtag recommendation technologies and short text representation methods, this paper uses a Hashtag recommendation method based on K-Nearest Neighbor. Firstly, we represent the texts of microblog into vectors, calculate similarities between user＇s text and training text. Then we extract the most similar blogs from the corpora. The results of four text representation methods named Vector space model, Latent semantic analysis, Latent Dirichlet allocation, Deep Learning for Hashtag recommendation are compared with each other. We use H7N9 Corpus on Twitter as our test dataset. Experimental results show that deep learning text representation method has achieved the best performance among all the methods.