This paper constructs a defect evaluation model of express service, uses the text mining methods of web crawler, SVM (Support Vector Machine) emotion analysis and LDA (Linear Discriminant Analysis) topic model to capture and clean up the online negative comment data of express service, establishes a semantic network diagram, and uses LDA topic model to extract the characteristic words of defect topic. Based on SERVQUAL model, it can classify the subject characteristic words of express service defects from the dimensions of tangibility, reliability, responsiveness, assurance, empathy and economy, etc., calculate the degree value and attention value of express service defects, and establish IPA model for defect mapping and identify the improvement direction. The evaluation model constructed in this paper has reference value for evaluating the defects of service industry and improving service quality. It is found that the “responsiveness” defect is the primary improvement direction, and the reliability, assurance and economy are the secondary improvement defects. Among them, the “responsiveness” defect has five improvement detail defects. The evaluation model constructed in this paper has reference value for evaluating the defects of service industry and improving service quality.