Update common/validation_utils.py

common/validation_utils.py

@@ -7,6 +7,18 @@ from RespondentAgent import *
 from langchain_groq import ChatGroq
 
 
+def is_first_person(answer):
+    """
+    Checks if the answer is written in first person.
+    Returns True if first person pronouns are found and third person references are not dominant.
+    """
+    # Look for first person pronouns
+    first_person = re.search(r'\b(I|my|me|mine|we|our|us|ours)\b', answer, re.IGNORECASE)
+    # Look for third person references (e.g., "he", "she", "they", or a capitalized name at the start)
+    third_person = re.search(r'\b(he|she|they|his|her|their|him|them)\b', answer, re.IGNORECASE)
+    return bool(first_person) and not bool(third_person)
+
+
 def validate_response(question, answer, user_profile_str, fast_facts_str, interview_transcript_text, respondent_type, ai_evaluator_agent, processor_llm):
     """
     Validates a response (answer) to a question using the appropriate evaluation method (exploratory or fact-based).
@@ -129,7 +141,11 @@ def validate_response(question, answer, user_profile_str, fast_facts_str, interv
                     relevance = None
         logging.info(f"Exploratory evaluation: plausibility={plausibility}, relevance={relevance}")
         if plausibility is not None and relevance is not None:
-            return plausibility >= 8.0 and relevance >= 8.0
+            if plausibility >= 8.0 and relevance >= 8.0:
+                if not is_first_person(answer):
+                    logging.warning("Did not pass style due to 3rd person use")
+                    return False
+                return True
         return False
     else:
         # Fact-based: Ask LLM for accuracy rating
@@ -157,5 +173,9 @@ def validate_response(question, answer, user_profile_str, fast_facts_str, interv
                     accuracy = None
         logging.info(f"Fact-based evaluation: accuracy={accuracy}")
         if accuracy is not None:
-            return accuracy >= 8.0
+            if accuracy >= 8.0:
+                if not is_first_person(answer):
+                    logging.warning("Did not pass style due to 3rd person use")
+                    return False
+                return True
         return False