Quantitative analysis and mining of product reviews posted by users are helpful for both manufacturers and consumers. During the work of fine-granularity product review mining, extracting feature-opinion pair is one of the core works. The corresponding corpus construction is of great significance for the application of high performance machine learning methods, improving the performance of feature-opinion extraction algorithm and automatic evaluation. This article introduces corpus constructing technology for fine-granularity product review mining and the initial corpus labeling work, thus realizing non-structured to structural changes. The corpus can be stored in database and thus provide great convenience for automatic query processing. Although current labeling work was performed in mobile phone products, it can be applied also to other product types for fine granularity corpus construction. So our work has good transplantation ability. 