Presently, text mining in e-commerce reviews predominantly focus on singular sentiment analysis, yet constraints persist in sentiment score computation, semantic inclination discernment, and lexicon construction. To address these limitations, this study establishes an e-commerce user comment management system based on text mining. It performs part-of-speech tagging and dependency grammar analysis on the historical corpus of e-commerce, unveiling collocations that potentially convey users' emotive predispositions. Subsequently, a dependency grammar rule table is formulated for the extraction of emotional words. The enhanced BiGRU model is employed for bidirectional extraction of textual features, which are subsequently fused with the TextCNN model. Test results evince that the system effectively accomplishes the desired objectives, with positive comments attaining accuracy and recall rates of 93.49% and 96.98%, respectively, thereby mitigating the drawbacks associated with laborious operations and inadequate precision inherent in extant e-commerce comment analysis systems.