Overloadpy/Amharic_Sentiment_Classification-engine

# 🇪🇹 Amharic Sentiment Intelligence Studio & CLI Engine *A High-Performance, Local, CPU-Optimized Ethiopic NLP Multi-Polarity Classification Platform.* --- -41cd52.svg) -brightgreen.svg) -success.svg) --- ## 👨‍💻 Project Leadership & Authorship - **Lead Engineer & Author:** **Yordanos Andargachew** - **Contact & Inquiries:** `+251 952 190 305` (`0952190305`) - **Research Domain:** Low-Resource Semitic NLP, Ethiopic Affective Computing, and Production ML Systems. --- ## 1. System Overview & Core Capabilities **`amh-synth`** is a production-grade NLP platform engineered specifically for **Amharic (አማርኛ)** sentiment analysis, customer intelligence, and affective computing. Designed from first principles to overcome the linguistic and computational bottlenecks of Semitic languages, it delivers real-time inference on local standard CPUs without cloud dependencies, machine translation layers, or synthetic keyword shortcuts. ``` ┌─────────────────────────────────────────────────────────────────────────────┐ │ amh-synth Production Pipeline │ └─────────────────────────────────────────────────────────────────────────────┘ │ [ 1. Raw Amharic Text ] ───► "ሲስተማችሁ ሁልጊዜ አይሰራም፡ በጣም አሳፋሪ ነው!" │ [ 2. O(N) Normalizer ] ───► Homophone Unification (ሐ/ኀ->ሀ, ሠ->ሰ, ዐ->አ, ፀ->ጸ) Punctuation Transliteration (፡, ።, ፣ -> ASCII) Elongation Collapse (በጣምምም -> በጣም) │ [ 3. Clause Splitting ] ───► Syntactic & Discourse Boundary Segmentation │ [ 4. AfriBERTa Engine ] ───► XLMRobertaForSequenceClassification (Torch 4 Threads) Raw Probabilities: P(pos), P(neg), P(neu) │ [ 5. Dual-Axis Math ] ───► Decoupled Calibration: Positive / Negative / Neutral / Mixed │ [ 6. Dual Interfaces ] ───► 🖥️ Desktop Studio GUI (PySide6 / Dark-Light Mode) ⌨️ Terminal CLI Harness (Typer + Rich Live REPL) ``` ### Key Engineering Features: 1. **4-Class Continuous Sentiment Classification:** Maps input text into **`Positive` (አዎንታዊ)**, **`Negative` (አሉታዊ)**, **` …