es-no-bio-sweep-20251013-173422-t17

Slur reclamation binary classifier
Task: LGBTQ+ reclamation vs non-reclamation use of harmful words on social media text.

Trial timestamp (UTC): 2025-10-13 17:34:22

Data case: es

Configuration (trial hyperparameters)

Model: Alibaba-NLP/gte-multilingual-base

Dev set results (summary)

Thresholds

• Default: 0.5
• Best global: 0.55
• Best by language: { "es": 0.55 }

Detailed evaluation

Classification report @ 0.5

              precision    recall  f1-score   support

 no-recl (0)     0.9167    0.7857    0.8462       112
    recl (1)     0.3333    0.6000    0.4286        20

    accuracy                         0.7576       132
   macro avg     0.6250    0.6929    0.6374       132
weighted avg     0.8283    0.7576    0.7829       132

Classification report @ best global threshold (t=0.55)

              precision    recall  f1-score   support

 no-recl (0)     0.8938    0.9018    0.8978       112
    recl (1)     0.4211    0.4000    0.4103        20

    accuracy                         0.8258       132
   macro avg     0.6574    0.6509    0.6540       132
weighted avg     0.8222    0.8258    0.8239       132

Classification report @ best per-language thresholds

              precision    recall  f1-score   support

 no-recl (0)     0.8938    0.9018    0.8978       112
    recl (1)     0.4211    0.4000    0.4103        20

    accuracy                         0.8258       132
   macro avg     0.6574    0.6509    0.6540       132
weighted avg     0.8222    0.8258    0.8239       132

Per-language metrics (at best-by-lang)

Data

• Train/Dev: private multilingual splits with ~15% stratified Dev (by (lang,label)).
• Source: merged EN/IT/ES data with bios retained (ignored if unused by model).

Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification, AutoConfig
import torch, numpy as np

repo = "SimoneAstarita/es-no-bio-sweep-20251013-173422-t17"
tok = AutoTokenizer.from_pretrained(repo)
cfg = AutoConfig.from_pretrained(repo)
model = AutoModelForSequenceClassification.from_pretrained(repo)

texts = ["example text ..."]
langs = ["en"]

mode = "best_global"  # or "0.5", "by_lang"

enc = tok(texts, truncation=True, padding=True, max_length=256, return_tensors="pt")
with torch.no_grad():
    logits = model(**enc).logits
probs = torch.softmax(logits, dim=-1)[:, 1].cpu().numpy()

if mode == "0.5":
    th = 0.5
    preds = (probs >= th).astype(int)
elif mode == "best_global":
    th = getattr(cfg, "best_threshold_global", 0.5)
    preds = (probs >= th).astype(int)
elif mode == "by_lang":
    th_by_lang = getattr(cfg, "thresholds_by_lang", {})
    preds = np.zeros_like(probs, dtype=int)
    for lg in np.unique(langs):
        t = th_by_lang.get(lg, getattr(cfg, "best_threshold_global", 0.5))
        preds[np.array(langs) == lg] = (probs[np.array(langs) == lg] >= t).astype(int)
print(list(zip(texts, preds, probs)))

Additional files

reports.json: all metrics (macro/weighted/accuracy) for @0.5, @best_global, and @best_by_lang. config.json: stores thresholds: default_threshold, best_threshold_global, thresholds_by_lang. postprocessing.json: duplicate threshold info for external tools.