We investigate hidden aspect mining problem that aims at automatically discovering aspect information from a collection of review texts in an unsupervised manner. The goal is to predict the user's ratings on each aspect. It does not require users to specify predefined seed terms for each aspect. We propose a generative model to tackle the hidden aspect mining problem. Our proposed model can detect the aspects of a particular domain. When predicting reviewer's ratings on each aspect, our model employs l1-regularizer to control the sparsity on the aspect and obtains more reliable predicted aspect ratings. Existing works on predicting aspect ratings fail to explore the aspect sparsity problem in the review texts leading to unreliable prediction. Experimental results on the two real-world product review corpora demonstrate that our model outperforms existing state-of-the-art models in the aspect rating prediction task.
[1]
Yue Lu,et al.
Latent aspect rating analysis without aspect keyword supervision
,
2011,
KDD.
[2]
Ivan Titov,et al.
A Joint Model of Text and Aspect Ratings for Sentiment Summarization
,
2008,
ACL.
[3]
Eric P. Xing,et al.
Sparse Topical Coding
,
2011,
UAI.
[4]
Michael I. Jordan,et al.
DiscLDA: Discriminative Learning for Dimensionality Reduction and Classification
,
2008,
NIPS.
[5]
Yue Lu,et al.
Latent aspect rating analysis on review text data: a rating regression approach
,
2010,
KDD.
[6]
Jure Leskovec,et al.
Learning Attitudes and Attributes from Multi-aspect Reviews
,
2012,
2012 IEEE 12th International Conference on Data Mining.