💬 Text2Moji 😏

😏 Feature Overview

• 😄 Text → Vector

  • TF–IDF on unigrams + bigrams
  • Configurable vocab size, min_df, max_df

• 😇 Text cleaning

  • Unicode fixes, lowercasing
  • Optional removal of URLs / @mentions / #hashtags
  • Caching of cleaned text for fast iteration

• 🤖 Models (v1 baselines)

  • Keyword / Bag-of-Words Weighted Classifier
  • Nearest-Centroid (cosine) Classifier

• 🤓 Evaluation

  • Top-1 / Top-3 / Top-5 accuracy
  • Macro + weighted precision / recall / F1
  • Per-class reports and qualitative top-k examples

• 😬 Trained models

  • Logistic Regression (OvR) on TF–IDF
  • Linear SVM & Multinomial Naive Bayes

• 🤨 Better UX metrics

  • Confusion matrices
  • Per-emoji “failure stories” (where the model gets the vibe wrong)

• 😃 Integration experiments

  • Minimal REST API (FastAPI/Flask) for /predict calls
  • Tiny web demo: type a message, see top-5 emojis live

• 😈 Stretch goals

  • fastText-style baseline
  • Tiny transformer/embedding model
  • Browser / VS Code prototype extension for emoji suggestion

---

😌 Architecture at a Glance

          ┌───────────────────────┐
          │       CSV Data        │
          │  (TEXT, Label, Map)   │
          └─────────┬─────────────┘
                    │
          ┌─────────▼─────────────┐
          │     Data Layer        │
          │  load + clean + cache │
          └─────────┬─────────────┘
                    │
          ┌─────────▼─────────────┐
          │   Features Layer      │
          │  TF–IDF (uni/bi-gram) │
          └─────────┬─────────────┘
                    │
      ┌─────────────▼──────────────┐
      │       Model Layer          │
      │ Keyword / Centroid / LR    │
      └─────────────┬──────────────┘
                    │
          ┌─────────▼─────────────┐
          │ Evaluation & Reports  │
          │  top-k, F1, plots, ex │
          └───────────────────────┘