Feature average rating stabilization changes are implemented

src/reviewsense_ecom/service/FeatureExtractor.py

@@ -19,10 +19,7 @@ class FeatureReviews(BaseModel):
 
 class FeatureExtractor:
     def __init__(self):
-        api_key = os.getenv('GROQ_API_KEY')
-        if not api_key:
-            raise ValueError("GROQ_API_KEY is not set in the environment variables.")
-        self.llm = get_llm(api_key)
+        self.llm = get_llm("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")
         self.parser = self._create_reviews_parser()
         self.prompt = self._create_extraction_prompt()
 
@@ -49,12 +46,14 @@ class FeatureExtractor:
                 {{ 
                     "feature" : {feature}
                     "sentence": "relevant sentence 1",
-                    "sentiment": "positive/negative/neutral"
+                    "sentiment": "positive/negative/neutral",
+                    "confidence": "confidence score between 0 and 1"
                 }},
                 {{
                     "feature" : {feature}
                     "sentence": "relevant sentence 2",
-                    "sentiment": "positive/negative/neutral"
+                    "sentiment": "positive/negative/neutral",
+                    "confidence": "confidence score between 0 and 1"
                 }}
             ]
         }}

src/reviewsense_ecom/service/FeatureUpdater.py

@@ -27,17 +27,34 @@ class FeatureUpdater:
                     "positive_count": 0,
                     "negative_count": 0
                 }
-
             current_feature = features[feature]
             existing_review_count = current_feature["review_count"]
 
-            new_rating = new_data.get("rating", 0) or 0
+            M = 4.5
+
+            C = max(2, 10 - (existing_review_count // 20))  # Reduce prior influence faster
+
+            # if existing_review_count < 5:
+            #     C = 5
+            # elif existing_review_count < 50:
+            #     C = 10
+            # elif existing_review_count < 500:
+            #     C = 15
+            # else:
+            #     C = 20
+
+            new_ratings = new_data.get("ratings", [])
+            new_rating = sum(new_ratings) / len(new_ratings) if new_ratings else 0
             new_positive = new_data.get("positive_count", 0) or 0
             new_negative = new_data.get("negative_count", 0) or 0
 
+            # Apply the Bayesian Average Formula
             new_average_rating = (
-                                         (current_feature["average_rating"] * existing_review_count) + new_rating
-                                 ) / (existing_review_count + 1) if existing_review_count > 0 else new_rating
+                                         (current_feature["average_rating"] * existing_review_count) + (
+                                         C * M) + new_rating
+                                 ) / (existing_review_count + C + 1) if existing_review_count > 0 else new_rating
+
+            print(f"{new_average_rating}+--------------------------$$$----------------")
 
             features[feature]["average_rating"] = round(new_average_rating, 1)
             features[feature]["review_count"] += 1

src/reviewsense_ecom/service/ReviewRater.py

@@ -14,43 +14,33 @@ class ReviewRater:
 
     def generate_feature_ratings(self, reviews_by_feature: List[Dict[str, str]]) -> Dict[
         str, Dict[str, Optional[float]]]:
-        """
-        Generate ratings and count sentiment occurrences for each feature.
-
-        Args:
-            reviews_by_feature (List[Dict[str, str]]): List of feature-specific reviews with sentiment.
-
-        Returns:
-            Dict[str, Dict[str, Optional[float]]]: Feature-wise ratings and sentiment counts.
-        """
-        feature_data = defaultdict(lambda: {"ratings": [], "positive_count": 0, "negative_count": 0})
+        feature_data = defaultdict(
+            lambda: {"ratings": [], "positive_count": 0, "negative_count": 0, "neutral_count": 0}
+        )
 
         for review_data in reviews_by_feature:
-            print("Processing Review:", review_data)  # Debugging line
-
             feature = review_data.get("feature")
             sentiment = review_data.get("sentiment")
+            confidence = float(review_data.get("confidence", 1))
 
             if not feature or not sentiment:
-                print("Skipping due to missing feature or sentiment")
-                continue  # Skip invalid entries
+                continue
 
             if "positive" in sentiment:
-                score = 5
+                score = 3 + 2 * confidence
+                print(f"confidence value for positive: {confidence}")
                 feature_data[feature]["positive_count"] += 1
             elif "negative" in sentiment:
-                score = 1
+                score = 3 - 2 * confidence
+                print(f"confidence value for negative: {confidence}")
+
                 feature_data[feature]["negative_count"] += 1
             else:
-                score = 3  # Neutral sentiment
-                feature_data[feature]["neutral_count"] = feature_data[feature].get("neutral_count", 0) + 1
+                score = 3
+                feature_data[feature]["neutral_count"] += 1
 
             feature_data[feature]["ratings"].append(score)
+            print(f"Calculated score: {score}--------------------*****------------")
 
-        # Compute average ratings
-        for feature, data in feature_data.items():
-            data["rating"] = round(np.mean(data["ratings"]), 1) if data["ratings"] else None
-            del data["ratings"]  # Remove temporary list after computing rating
-
-        print("Final Feature Data:", dict(feature_data))  # Debugging line
         return feature_data
+