Consumer reviews are an important data source to investigate consumer sentiment, and data mining of product reviews is an important way to help online pharmaceutical E-commerce improve business. Based on the user review text of online pharmaceutical e-commerce, a sentiment dictionary for online pharmaceutical e-commerce based on the SO-PMI（semantic orientation pointwise mutual information） algorithm is constructed , then integrate weighted word vectors and the LLM(large language model), train classifiers are used to construct a sentiment recognition model using the XGBoost（extreme gradient boosting） algorithm. Finally a sentiment index to analyze the sentiment tendency of the online pharmaceutical E-commerce consumer reviews is established from multiple dimensions. The empirical analysis shows that the validation indexes of the constructed emotion recognition model such as AUC（area under curve) are further improved than that of the LLM model, which has certain application value.