|
|
| import joblib, pickle, numpy as np |
| from huggingface_hub import hf_hub_download |
|
|
| class FakeNewsDetector: |
| def __init__(self): |
| print("Loading Fake News Detector from Hugging Face...") |
| repo = "ghimirewe22/Classical_Model" |
| self.rf = joblib.load(hf_hub_download(repo, "rf_classifier.joblib")) |
| self.gb = joblib.load(hf_hub_download(repo, "gb_classifier.joblib")) |
| self.lr = joblib.load(hf_hub_download(repo, "lr_classifier.joblib")) |
| self.oc = joblib.load(hf_hub_download(repo, "oneclass_svm.joblib")) |
| self.vect = joblib.load(hf_hub_download(repo, "tfidf_vectorizer.joblib")) |
| self.pt = joblib.load(hf_hub_download(repo, "power_transformer.joblib")) |
| with open(hf_hub_download(repo, "vocab.pkl"), "rb") as f: |
| self.vocab = pickle.load(f) |
| print("Model loaded! Ready to detect fake news.") |
|
|
| def predict(self, text): |
| X = self.vect.transform([text]).toarray() |
| try: |
| X = self.pt.transform(X) |
| except: pass |
| |
| oc_vote = 0 if self.oc.predict(X)[0] == -1 else 1 |
| votes = [ |
| oc_vote, |
| int(self.lr.predict(X)[0]), |
| int(self.gb.predict(X)[0]), |
| int(self.rf.predict(X)[0]) |
| ] |
| result = "REAL" if sum(votes) >= 2 else "FAKE" |
| prob = self._fake_prob(text) |
| return f"{result} ({prob:.1f}% fake)" |
|
|
| def _fake_prob(self, text): |
| X = self.vect.transform([text]).toarray() |
| try: X = self.pt.transform(X) except: pass |
| probs = [] |
| for m in [self.lr, self.gb, self.rf]: |
| if hasattr(m, "predict_proba"): |
| probs.append(m.predict_proba(X)[0][0]) |
| return np.mean(probs) * 100 if probs else 50.0 |
|
|
