In order to solve the problem of the emotional orientation of each evaluation object in the massive e-commerce evaluation information and the mismatch between the evaluation object and the evaluation word, a multi-granularity conditional random field model extraction evaluation unit method SSMCRFs (syntactic semantic and multi-grained conditional random fields, SSMCRFs) is proposed. First, we crawl the comment data of Jingdong Mall as the basic data, and process the comment text in syntactic relationship and semantic relationship; then, we use the TF-IDF algorithm to perform statistical analysis on the preprocessed data set to determine the user's attention; Finally, the conditional random field model is used for evaluation unit identification. The experimental results show that the accuracy rate of SSMCRFs in the identification and evaluation unit is 92.92\%, the recall rate is 93.25\%, and the F value is 93.08\%. Compared with the method by Ma, et al. (2017), the SSMCRFs method has a better improvement in accuracy, recall rate and F value.