Sentimental Analysis App with MLOps

Abstract

An application-oriented sentiment classification framework is presented to process short English-language text through a lightweight natural-language pipeline and an APImediated web interface. The implementation uses TextBlob polarity estimation, FastAPI for request validation and inference delivery, and HTML, CSS, and JavaScript for interactive feedback. A polarity-based decision rule maps analyzed text to positive, negative, and neutral classes, while presentation components provide visual cues and a compact response workflow. The associated engineering design addresses version control, container packaging, reproducibility, automated testing, and staged CI/CD adoption. The project report describes an evaluation involving 200 human-labeled sentences and reports 95% classification accuracy, 93.5% F1-score, and 0.35 s average response time; these figures are presented as reported project observations because the underlying per-class confusion matrix and individual timing traces are not supplied. The findings illustrate the feasibility of lightweight text opinion mining alongside a maintainable web-service architecture, while identifying limitations for sarcasm, contextual semantics, and verified deployment automation.