With the increase in popularity of e-commerce, more and more customer reviews are available online, it’s usually hard to go through each of them. Latent Dirichlet Allocation (LDA) was used to mine product aspect. Considering the weakness of standard LDA when processing review text, we defined the product aspect model, and predefined aspects for different domains, and proposed aspect model combined N-gram based on sentence, which can automate aspect clustering. Our experimental results show that the proposed model can cluster mostly aspects and recognize representative words for the aspect with more than one word, and achieve better sentence-level aspect precision than previously proposed aspect models.