A Chinese student's thesis: sentiment analysis with SVM, Bayes, and AdaBoost

What it does

This repo implements a full text-classification pipeline for Chinese Weibo posts: fetch posts via a companion scraper, run SVM for initial classification, apply Naive Bayes for sentiment scoring, then boost everything with AdaBoost (both binary and multiclass SAMME variants). It is explicitly framed as a graduation thesis project.

The interesting bit

The author is refreshingly honest: “the article and code have many errors, just borrow the ideas.” That candor is rarer than the algorithms themselves. The project also serves as a time capsule of 2017-era ML pedagogy — Anaconda 3.5+, hand-rolled ensemble methods, and no deep learning in sight.

Key highlights

• Three-classifier stack: SVM → Naive Bayes → AdaBoost
• Includes both binary and multiclass AdaBoost (SAMME, SAMME.R)
• Companion repo weibo_get handles data collection
• Full Chinese-language documentation in /doc
• 1,085 stars despite the author’s self-deprecating warnings

Caveats

• Author explicitly states code contains errors and is no longer maintained
• “Mostly referenced from others” — unclear how much is original implementation vs. glue code
• No benchmarks, tests, or reproducibility instructions provided

Verdict

Worth a quick browse if you’re teaching or learning classic ensemble methods and want a realistic example of undergraduate ML work. Skip it if you need production sentiment analysis or modern tooling.