feat: Introduce predefined topics list in TopicService and enhance topic extraction logic to validate against this list, ensuring accurate topic matching and improved error handling.

services/topic_service.py

@@ -11,19 +11,65 @@ from config import GROQ_API_KEY
 
 logger = logging.getLogger(__name__)
 
+# Predefined topics list
+PREDEFINED_TOPICS = [
+    "Assisted suicide should be a criminal offence",
+    "We should abolish intellectual property rights",
+    "Homeschooling should be banned",
+    "The vow of celibacy should be abandoned",
+    "We should legalize prostitution",
+    "We should ban private military companies",
+    "We should abolish capital punishment",
+    "Foster care brings more harm than good",
+    "Routine child vaccinations should be mandatory",
+    "We should abolish the three-strikes laws",
+    "We should subsidize student loans",
+    "We should end the use of economic sanctions",
+    "We should end mandatory retirement",
+    "We should close Guantanamo Bay detention camp",
+    "We should subsidize space exploration",
+    "We should abandon the use of school uniform",
+    "The use of public defenders should be mandatory",
+    "We should adopt an austerity regime",
+    "Social media platforms should be regulated by the government",
+    "We should ban human cloning",
+    "We should adopt atheism",
+    "We should introduce compulsory voting",
+    "We should adopt libertarianism",
+    "We should abolish the right to keep and bear arms",
+    "We should legalize sex selection",
+    "We should abandon marriage",
+    "Entrapment should be legalized",
+    "We should end affirmative action",
+    "We should prohibit women in combat",
+    "We should adopt a zero-tolerance policy in schools",
+    "We should subsidize vocational education",
+    "We should ban the use of child actors",
+    "We should legalize cannabis",
+    "We should ban cosmetic surgery",
+    "We should end racial profiling",
+    "We should prohibit flag burning",
+    "The USA is a good country to live in",
+    "We should ban algorithmic trading",
+    "We should fight for the abolition of nuclear weapons",
+    "We should fight urbanization",
+    "We should subsidize journalism",
+]
+
 
 class TopicOutput(BaseModel):
     """Pydantic schema for topic extraction output"""
-    topic: str = Field(..., description="A specific, detailed topic description")
+    topic: str = Field(..., description="The selected topic from the predefined list that most closely matches the input text")
 
 
 class TopicService:
-    """Service for extracting topics from text arguments"""
+    """Service for extracting topics from text arguments by matching to predefined topics"""
     
     def __init__(self):
         self.llm = None
         self.model_name = "openai/gpt-oss-safeguard-20b"  # Default model
         self.initialized = False
+        self.predefined_topics = PREDEFINED_TOPICS
         
     def initialize(self, model_name: Optional[str] = None):
         """Initialize the Groq LLM with structured output"""
@@ -57,16 +103,44 @@ class TopicService:
             logger.error(f"Error initializing topic service: {str(e)}")
             raise RuntimeError(f"Failed to initialize topic service: {str(e)}")
     
+    def _get_system_message(self) -> str:
+        """Generate system message with predefined topics list"""
+        topics_list = "\n".join([f"{i+1}. {topic}" for i, topic in enumerate(self.predefined_topics)])
+        
+        return f"""You are a topic classification model. Your task is to select the MOST SIMILAR topic from the predefined list below that best matches the user's input text.
+
+IMPORTANT: You MUST return EXACTLY one of the predefined topics below. Do not create new topics or modify the wording.
+
+Predefined Topics:
+{topics_list}
+
+Instructions:
+1. Analyze the user's input text carefully
+2. Identify the main theme, subject, or argument being discussed
+3. Find the topic from the predefined list that is MOST SIMILAR to the input text
+4. Return the EXACT topic text as it appears in the list above
+
+Examples:
+- Input: "I think we need to make assisted suicide illegal and punishable by law."
+  Output: "Assisted suicide should be a criminal offence"
+
+- Input: "Student debt is crushing young people. The government should help pay for college."
+  Output: "We should subsidize student loans"
+
+- Input: "Marijuana should be legal for adults to use recreationally."
+  Output: "We should legalize cannabis"
+"""
+    
     @traceable(name="extract_topic")
     def extract_topic(self, text: str) -> str:
         """
-        Extract a topic from the given text/argument
+        Extract a topic from the given text/argument by matching to predefined topics
         
         Args:
             text: The input text/argument to extract topic from
             
         Returns:
-            The extracted topic string
+            The extracted topic string (must be one of the predefined topics)
         """
         if not self.initialized:
             self.initialize()
@@ -78,16 +152,7 @@ class TopicService:
         if len(text) == 0:
             raise ValueError("Text cannot be empty")
         
-        system_message = """You are an information extraction model.
-Extract a topic from the user text. The topic should be a single sentence that captures the main idea of the text in simple english.
-
-Examples:
-- Text: "Governments should subsidize electric cars to encourage adoption."
-  Output: topic="government subsidies for electric vehicle adoption"
-
-- Text: "Raising the minimum wage will hurt small businesses and cost jobs."
-  Output: topic="raising the minimum wage and its economic impact on small businesses"
-"""
+        system_message = self._get_system_message()
         
         try:
             result = self.llm.invoke(
@@ -97,7 +162,32 @@ Examples:
                 ]
             )
             
-            return result.topic
+            selected_topic = result.topic.strip()
+            
+            # Validate that the returned topic is in the predefined list
+            if selected_topic not in self.predefined_topics:
+                logger.warning(
+                    f"LLM returned topic not in predefined list: '{selected_topic}'. "
+                    f"Attempting to find closest match..."
+                )
+                # Try to find the closest match (case-insensitive)
+                selected_topic_lower = selected_topic.lower()
+                for predefined_topic in self.predefined_topics:
+                    if predefined_topic.lower() == selected_topic_lower:
+                        selected_topic = predefined_topic
+                        logger.info(f"Found case-insensitive match: '{selected_topic}'")
+                        break
+                else:
+                    # If still no match, log error and raise
+                    logger.error(
+                        f"Could not match returned topic '{selected_topic}' to any predefined topic. "
+                        f"Available topics: {self.predefined_topics[:3]}..."
+                    )
+                    raise ValueError(
+                        f"Returned topic '{selected_topic}' is not in the predefined topics list"
+                    )
+            
+            return selected_topic
             
         except Exception as e:
             logger.error(f"Error extracting topic: {str(e)}")