"""Inference for the fediverse bot classifier. from huggingface_hub import snapshot_download import sys path = snapshot_download("FranUPM/fediverse-bot-detector") sys.path.insert(0, path) from predict import BotClassifier clf = BotClassifier() clf.predict([{"username": "news_bot_2024", "note": "
Automated headlines every hour
", "statuses_count": 52000, "followers_count": 120, "following_count": 0}]) """ import numpy as np import pandas as pd import torch from bs4 import BeautifulSoup from transformers import BertModel, BertTokenizer from bot_features import add_account_features, build_feature_matrix, load_artifacts EMBEDDING_MODEL = 'bert-base-multilingual-cased' REQUIRED_FIELDS = ('username', 'note', 'statuses_count', 'followers_count', 'following_count') def remove_html(text): if not isinstance(text, str): return '' return BeautifulSoup(text, 'html.parser').get_text(separator=' ', strip=True) class BotClassifier: def __init__(self, device=None, threshold=0.5): self.model, self.scaler, self.vocab = load_artifacts() self.device = torch.device(device or ('cuda' if torch.cuda.is_available() else 'cpu')) self.tokenizer = BertTokenizer.from_pretrained(EMBEDDING_MODEL) self.bert = BertModel.from_pretrained(EMBEDDING_MODEL).to(self.device).eval() self.threshold = threshold @torch.no_grad() def embed(self, texts, max_length=128, batch_size=32): """[CLS] embedding of each text (768-dim), same as used in training.""" out = [] for i in range(0, len(texts), batch_size): inputs = self.tokenizer(texts[i:i + batch_size], return_tensors='pt', truncation=True, padding=True, max_length=max_length) inputs = {k: v.to(self.device) for k, v in inputs.items()} out.append(self.bert(**inputs).last_hidden_state[:, 0, :].cpu().numpy()) return np.vstack(out) def predict(self, accounts): """accounts: list of dicts with username, note (HTML bio), statuses_count, followers_count, following_count. Optional 'language' skips fastText detection.""" if isinstance(accounts, dict): accounts = [accounts] for a in accounts: missing = [f for f in REQUIRED_FIELDS if f not in a] if missing: raise ValueError(f"account {a.get('username')!r} is missing {missing}") df = pd.DataFrame(accounts) df['clean_note'] = df['note'].apply(remove_html) df = add_account_features(df) emb = self.embed(df['clean_note'].fillna('').astype(str).tolist()) X = build_feature_matrix(df, emb, self.scaler, self.vocab) proba = self.model.predict_proba(X)[:, 1] return [{'username': u, 'prediction': bool(p >= self.threshold), 'probability': float(p)} for u, p in zip(df['username'], proba)] if __name__ == '__main__': clf = BotClassifier() print(clf.predict([ {'username': 'news_bot_2024', 'note': 'Automated headlines posted every hour.
', 'statuses_count': 52000, 'followers_count': 120, 'following_count': 0}, {'username': 'maria', 'note': 'Biologist, cyclist, coffee lover.
', 'statuses_count': 830, 'followers_count': 310, 'following_count': 280}, ]))