From dfe6fe364c89afed4ef89dd0b52724a5ac205976 Mon Sep 17 00:00:00 2001 From: Pratyush Niraula Date: Sun, 26 Apr 2026 23:58:30 -0500 Subject: [PATCH 1/2] misinfo fixed for some more narratives --- pipelines/misinfo_checker.py | 29 +++++++++++++++++++++++++++++ 1 file changed, 29 insertions(+) diff --git a/pipelines/misinfo_checker.py b/pipelines/misinfo_checker.py index 96a904a..2999e61 100644 --- a/pipelines/misinfo_checker.py +++ b/pipelines/misinfo_checker.py @@ -1151,7 +1151,36 @@ def process_narratives_table(batch_size: int = 9999, force_reprocess: bool = Fal print(f" risk_score: {risk_result['risk_score']}") print(f" claims analyzed: {risk_result['details']['total_claims']}") else: + # No claims linked - evaluate the narrative text itself print(" No claims linked to this narrative") + print(" Evaluating narrative text directly...") + + entailment_label, entailment_conf = check_entailment(narrative_text) + + # Map entailment to risk score + if entailment_label == "refuted" and entailment_conf >= 0.75: + risk_score = 3.0 # High risk - strongly contradicts medical science + elif entailment_label == "refuted" and entailment_conf >= 0.65: + risk_score = 2.5 # Medium-high risk + elif entailment_label == "refuted": + risk_score = 2.0 # Medium risk - contradicts but low confidence + else: + risk_score = 1.0 # Low risk - supported or neutral + + updates["narrative_risk_score"] = risk_score + updates["narrative_details"] = json.dumps({ + "total_claims": 0, + "high_risk_claims": 0, + "medium_risk_claims": 0, + "low_risk_claims": 0, + "avg_sentiment": 0.0, + "verified_false_count": 0, + "entailment_label": entailment_label, + "entailment_confidence": entailment_conf, + "source": "narrative_text_evaluation" + }) + print(f" entailment: {entailment_label} (confidence: {entailment_conf:.4f})") + print(f" risk_score: {risk_score} (from narrative text)") # Update narrative if updates: From 795b9383aec51dbd5597f73d9ada0fbe99348239 Mon Sep 17 00:00:00 2001 From: Pratyush Niraula Date: Sun, 26 Apr 2026 23:59:44 -0500 Subject: [PATCH 2/2] fixed lint --- pipelines/misinfo_checker.py | 34 +++++++++++++++++++--------------- 1 file changed, 19 insertions(+), 15 deletions(-) diff --git a/pipelines/misinfo_checker.py b/pipelines/misinfo_checker.py index 2999e61..b273c80 100644 --- a/pipelines/misinfo_checker.py +++ b/pipelines/misinfo_checker.py @@ -1154,9 +1154,9 @@ def process_narratives_table(batch_size: int = 9999, force_reprocess: bool = Fal # No claims linked - evaluate the narrative text itself print(" No claims linked to this narrative") print(" Evaluating narrative text directly...") - + entailment_label, entailment_conf = check_entailment(narrative_text) - + # Map entailment to risk score if entailment_label == "refuted" and entailment_conf >= 0.75: risk_score = 3.0 # High risk - strongly contradicts medical science @@ -1166,20 +1166,24 @@ def process_narratives_table(batch_size: int = 9999, force_reprocess: bool = Fal risk_score = 2.0 # Medium risk - contradicts but low confidence else: risk_score = 1.0 # Low risk - supported or neutral - + updates["narrative_risk_score"] = risk_score - updates["narrative_details"] = json.dumps({ - "total_claims": 0, - "high_risk_claims": 0, - "medium_risk_claims": 0, - "low_risk_claims": 0, - "avg_sentiment": 0.0, - "verified_false_count": 0, - "entailment_label": entailment_label, - "entailment_confidence": entailment_conf, - "source": "narrative_text_evaluation" - }) - print(f" entailment: {entailment_label} (confidence: {entailment_conf:.4f})") + updates["narrative_details"] = json.dumps( + { + "total_claims": 0, + "high_risk_claims": 0, + "medium_risk_claims": 0, + "low_risk_claims": 0, + "avg_sentiment": 0.0, + "verified_false_count": 0, + "entailment_label": entailment_label, + "entailment_confidence": entailment_conf, + "source": "narrative_text_evaluation", + } + ) + print( + f" entailment: {entailment_label} (confidence: {entailment_conf:.4f})" + ) print(f" risk_score: {risk_score} (from narrative text)") # Update narrative