PerSent 0.1.2__tar.gz → 0.1.3__tar.gz

{persent-0.1.2 → persent-0.1.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: PerSent
-Version: 0.1.2
+Version: 0.1.3
 Summary: Persian Sentiment Analysis Toolkit
 Home-page: https://github.com/RezaGooner/PerSent
 Author: RezaGooner

{persent-0.1.2 → persent-0.1.3}/PerSent/__init__.py

@@ -1,4 +1,4 @@
 from .analyzer import PersianSentimentAnalyzer
 
-__version__ = "0.1.2"
+__version__ = "0.1.3"
 __all__ = ['PersianSentimentAnalyzer']

{persent-0.1.2 → persent-0.1.3}/PerSent/analyzer.py

@@ -1,13 +1,16 @@
 import pandas as pd
 from hazm import Normalizer, word_tokenize, Stemmer, stopwords_list
 import re
-from tqdm import tqdm  
+from tqdm import tqdm
 from gensim.models import Word2Vec
 import numpy as np
 from sklearn.linear_model import LogisticRegression
 from sklearn.model_selection import train_test_split
 import os
 import joblib
+from transformers import AutoModelForSequenceClassification, AutoTokenizer
+import torch
+from torch.nn.functional import softmax
 
 class PersianSentimentAnalyzer:
     def __init__(self, model_dir='persian_sentiment/model'):
@@ -17,19 +20,18 @@ class PersianSentimentAnalyzer:
         self.model_dir = model_dir
         self.vectorizer = None
         self.classifier = None
-        self.emotion_keywords = {
-            'Fear': ['ترس', 'وحشت', 'هراس'],
-            'Surprise': ['تعجب', 'شگفت', 'حیرت'],
-            'Anger': ['خشم', 'عصبانیت', 'خشونت'],
-            'Contempt': ['تحقیر', 'توهین', 'حقارت'],
-            'Joy': ['شادی', 'خوشحالی', 'خنده'],
-            'Sadness': ['غم', 'اندوه', 'ناراحتی'],
-            'Disgust': ['انزجار', 'تهوع', 'نفرت']
-        }
         
-        os.makedirs(self.model_dir, exist_ok=True)
+        # مدل تحلیل احساسات
+        self.emotion_tokenizer = AutoTokenizer.from_pretrained(
+            "HooshvareLab/bert-fa-base-uncased-emotion")
+        self.emotion_model = AutoModelForSequenceClassification.from_pretrained(
+            "HooshvareLab/bert-fa-base-uncased-emotion")
+        self.emotion_labels = ['ترس', 'تعجب', 'خشم', 'تحقیر', 'شادی', 'غم', 'انزجار']
         
+        os.makedirs(self.model_dir, exist_ok=True)
+
     def _preprocess_text(self, text):
+        """پیش‌پردازش متن فارسی"""
         text = self.normalizer.normalize(str(text))
         text = re.sub(r'[!()-\[\]{};:\'",؟<>./?@#$%^&*_~۰-۹\d]+', ' ', text)
         text = re.sub(r'\s+', ' ', text).strip()
@@ -40,8 +42,9 @@ class PersianSentimentAnalyzer:
             if token not in self.stopwords and len(token) > 1
         ]
         return processed_tokens
-    
+
     def _sentence_vector(self, sentence, model):
+        """تبدیل جمله به بردار با میانگین بردارهای کلمات"""
         vectors = []
         for word in sentence:
             try:
@@ -49,20 +52,34 @@ class PersianSentimentAnalyzer:
             except KeyError:
                 vectors.append(np.zeros(100))
         return np.mean(vectors, axis=0) if vectors else np.zeros(100)
-    
-    def _emotion_analysis(self, tokens):
-        """تحلیل احساسات بر اساس کلیدواژه‌های از پیش تعریف شده"""
-        emotions = {emotion: 0 for emotion in self.emotion_keywords.keys()}
-        for word in tokens:
-            for emotion, keywords in self.emotion_keywords.items():
-                if word in keywords:
-                    emotions[emotion] += 1
-        return emotions
-    
+
+    def _emotion_analysis(self, text):
+        """تحلیل احساسات با مدل BERT فارسی"""
+        inputs = self.emotion_tokenizer(
+            text,
+            return_tensors="pt",
+            truncation=True,
+            padding=True,
+            max_length=128
+        )
+        
+        with torch.no_grad():
+            outputs = self.emotion_model(**inputs)
+        
+        probs = softmax(outputs.logits, dim=1)
+        scores = probs.numpy()[0]
+        
+        return {
+            label: round(float(score), 4)
+            for label, score in zip(self.emotion_labels, scores)
+        }
+
     def train(self, train_csv, test_size=0.2, vector_size=100, window=5):
+        """آموزش مدل تحلیل توصیه"""
         df = pd.read_csv(train_csv)
         df['tokens'] = df['body'].apply(self._preprocess_text)
         
+        # آموزش مدل Word2Vec
         self.vectorizer = Word2Vec(
             sentences=df['tokens'],
             vector_size=vector_size,
@@ -71,6 +88,7 @@ class PersianSentimentAnalyzer:
             workers=4
         )
         
+        # تبدیل جملات به بردار
         X = np.array([self._sentence_vector(s, self.vectorizer) for s in df['tokens']])
         y = df['recommendation_status'].map({
             "no_idea": 2,
@@ -78,29 +96,34 @@ class PersianSentimentAnalyzer:
             "not_recommended": 0
         }).values
         
+        # تقسیم داده‌ها و آموزش مدل
         X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=test_size)
-        
         self.classifier = LogisticRegression(max_iter=1000)
         self.classifier.fit(X_train, y_train)
         
+        # ذخیره مدل
         self.save_model()
         accuracy = self.classifier.score(X_test, y_test)
-        return accuracy
-    
+        return round(accuracy, 4)
+
     def predict(self, text, mode=0):
-        """پیش‌بینی احساس متن با قابلیت انتخاب حالت تحلیل
+        """
+        پیش‌بینی احساس متن
         
         پارامترها:
+            text (str): متن ورودی
             mode (int): 
                 0 = تحلیل توصیه (پیش‌فرض)
-                1 = تحلیل احساسات (Fear, Joy, etc.)
+                1 = تحلیل احساسات با مدل BERT
+                
+        بازگشتی:
+            dict یا str: نتایج تحلیل
         """
-        if not self.classifier or not self.vectorizer:
-            raise Exception("Model not trained! Call train() first or load a pretrained model.")
-            
-        tokens = self._preprocess_text(text)
-        
         if mode == 0:
+            if not self.classifier or not self.vectorizer:
+                raise Exception("Recommendation model not trained! Call train() first.")
+                
+            tokens = self._preprocess_text(text)
             vector = self._sentence_vector(tokens, self.vectorizer)
             prediction = self.classifier.predict([vector])[0]
             return {
@@ -110,15 +133,17 @@ class PersianSentimentAnalyzer:
             }[prediction]
             
         elif mode == 1:
-            return self._emotion_analysis(tokens)
+            return self._emotion_analysis(text)
             
         else:
-            raise ValueError("Invalid mode. Use 0 (recommendation analysis) or 1 (emotion analysis)")
-    
+            raise ValueError("Invalid mode. Use 0 (recommendation) or 1 (emotion)")
+
     def save_model(self):
+        """ذخیره مدل تحلیل توصیه"""
         joblib.dump(self.classifier, os.path.join(self.model_dir, 'classifier.joblib'))
         self.vectorizer.save(os.path.join(self.model_dir, 'word2vec.model'))
-    
+
     def load_model(self):
+        """بارگذاری مدل تحلیل توصیه"""
         self.classifier = joblib.load(os.path.join(self.model_dir, 'classifier.joblib'))
         self.vectorizer = Word2Vec.load(os.path.join(self.model_dir, 'word2vec.model'))

{persent-0.1.2 → persent-0.1.3}/PerSent.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: PerSent
-Version: 0.1.2
+Version: 0.1.3
 Summary: Persian Sentiment Analysis Toolkit
 Home-page: https://github.com/RezaGooner/PerSent
 Author: RezaGooner

{persent-0.1.2 → persent-0.1.3}/setup.py

@@ -5,7 +5,7 @@ with open("README.md", "r", encoding="utf-8") as fh:
 
 setup(
     name="PerSent",
-    version="0.1.2",
+    version="0.1.3",
     author="RezaGooner",
     author_email="RezaAsadiProgrammer@Gmail.com",
     description="Persian Sentiment Analysis Toolkit",