Release v0.3 safety and grounding quality dataset

README.md

@@ -3,7 +3,7 @@ pretty_name: LifeMultiTurnStreamingCoT
 language:
 - en
 license: apache-2.0
-version: "v0.1"
+version: "v0.3"
 task_categories:
 - text-generation
 tags:
@@ -22,24 +22,30 @@ configs:
 ---
 # LifeMultiTurnStreamingCoT
 
+Version: v0.3 — Safety and Grounding Quality Release
+
 LifeMultiTurnStreamingCoT is a text-to-text multi-turn life-domain dataset. Each row uses previous user-assistant dialogue turns as input, deterministic turn-level streaming state tracking as intermediate supervision, a compact final-state-based deep reasoning summary, and the next assistant turn as the answer.
 
-This version adds real task-oriented multi-turn dialogue data from MultiWOZ 2.2 and Taskmaster, alongside the existing DailyDialog source. The quality pipeline was upgraded with source-aware metadata, real-data checks, multi-turn validation, placeholder filtering, length checks, role-alternation checks, and task-oriented reasoning templates for slot/constraint tracking.
+This v0.3 release keeps the real DailyDialog, MultiWOZ 2.2, and Taskmaster sources while improving training suitability through safety filtering, grounded slot sanity checks, target-answer usefulness checks, taxonomy confidence/evidence, and quality tiers.
 
 ## Current Demo/Build Statistics
 
-- Version: v0.1
+- Version: v0.3 — Safety and Grounding Quality Release
 - Total rows: 30000
-- Train rows: 24215
-- Eval rows: 5785
-- High-quality train rows: 20653
-- High-quality eval rows: 4899
-- Average input turns: 9.706
-- Average streaming chunks: 9.706
+- Train rows: 24211
+- Eval rows: 5789
+- High-quality train rows: 8167
+- High-quality eval rows: 1845
+- Average input turns: 9.705
+- Average streaming chunks: 9.705
 - Source distribution: {"DailyDialog": 10000, "MultiWOZ": 10000, "Taskmaster": 10000}
-- Domain category distribution: {"customer_service": 204, "education_career": 2287, "finance_business": 180, "food_dining": 1263, "general_daily_life": 2879, "health_wellness": 220, "home_services": 227, "hospitality_lodging": 404, "personal_schedule": 683, "shopping_retail": 530, "social_relationship": 1986, "technology_support": 195, "travel_transportation": 18942}
-- Intent category distribution: {"booking_or_reservation": 15759, "confirmation_clarification": 2710, "customer_support": 1241, "emotional_support": 232, "information_request": 6236, "instruction_following": 26, "negotiation_decision": 139, "planning_coordination": 1461, "problem_solving": 445, "recommendation": 739, "small_talk": 1012}
-- Scenario category distribution: {"attraction_search": 455, "banking_support": 212, "customer_complaint": 218, "family_conversation": 603, "flight_booking": 9926, "food_ordering": 18, "friend_conversation": 459, "general_conversation": 3288, "home_repair": 97, "hotel_booking": 1557, "hotel_search": 184, "insurance_support": 95, "job_interview": 104, "medical_assistance": 290, "movie_ticketing": 218, "music_search": 141, "restaurant_booking": 1890, "restaurant_search": 819, "schedule_planning": 642, "school_life": 854, "shopping_assistance": 395, "taxi_booking": 1106, "technical_support": 460, "train_booking": 4317, "travel_planning": 252, "workplace_conversation": 1400}
+- Domain category distribution: {"customer_service": 204, "education_career": 2288, "finance_business": 180, "food_dining": 1262, "general_daily_life": 2881, "health_wellness": 221, "home_services": 226, "hospitality_lodging": 404, "personal_schedule": 688, "shopping_retail": 527, "social_relationship": 1984, "technology_support": 193, "travel_transportation": 18942}
+- Intent category distribution: {"booking_or_reservation": 15758, "confirmation_clarification": 2710, "customer_support": 1241, "emotional_support": 232, "information_request": 6242, "instruction_following": 26, "negotiation_decision": 139, "planning_coordination": 1460, "problem_solving": 440, "recommendation": 739, "small_talk": 1013}
+- Scenario category distribution: {"attraction_search": 455, "banking_support": 210, "customer_complaint": 218, "family_conversation": 600, "flight_booking": 9926, "food_ordering": 18, "friend_conversation": 456, "general_conversation": 3288, "home_repair": 101, "hotel_booking": 1557, "hotel_search": 184, "insurance_support": 95, "job_interview": 104, "medical_assistance": 291, "movie_ticketing": 218, "music_search": 141, "restaurant_booking": 1890, "restaurant_search": 819, "schedule_planning": 646, "school_life": 853, "shopping_assistance": 393, "taxi_booking": 1109, "technical_support": 453, "train_booking": 4317, "travel_planning": 253, "workplace_conversation": 1405}
+- Taxonomy confidence distribution: {"high": 20850, "medium": 9150}
+- Quality tier distribution: {"bronze": 2829, "drop": 17159, "gold": 6921, "silver": 3091}
+- Safety category distribution: {"safe": 28743, "sensitive": 1257}
+- Target answer quality distribution: {"drop": 3848, "strong": 15196, "usable": 6828, "weak": 4128}
 - Unknown/other taxonomy ratio: 0.0
 - Category distribution: {"daily_dialogue": 10000, "task_oriented_dialogue": 20000}
 
@@ -53,6 +59,30 @@ This version adds real task-oriented multi-turn dialogue data from MultiWOZ 2.2
 
 Rows contain `id`, `source_dataset`, `source_id`, `dialogue_id`, `domain`, `task_type`, `dialogue_history`, `streaming_chunks`, `deep_reasoning`, `answer`, `metadata`, `quality_flags`, `quality_score`, `is_high_quality`, and `split`. The final schema remains unified across sources; source-specific details such as source, category, domain/services, scenario, original split, and raw file are kept in `metadata`.
 
+## Version History
+
+### v0.3 — Safety and Grounding Quality Release
+
+- Added safety/content filtering for training suitability.
+- Added grounded slot extraction checks.
+- Added target answer usefulness checks.
+- Added taxonomy confidence and evidence fields.
+- Added quality tiers: gold, silver, bronze, and drop.
+- Updated `is_high_quality` to use quality tiers.
+- Removed or downgraded unsafe, ungrounded, malformed, or low-usefulness samples.
+
+### v0.2 — Taxonomy-Aware Quality Release
+
+- Added `domain_category`, `intent_category`, and `scenario_category`.
+- Added deterministic source-aware taxonomy rules.
+- Improved category-aware quality thresholds.
+- Reduced false penalties for long task-oriented dialogues.
+
+### v0.1 — Initial Real-Source Release
+
+- Added real DailyDialog, MultiWOZ 2.2, and Taskmaster data.
+- Built a balanced multi-source dataset with rule-based streaming/deep reasoning.
+
 ## Category Taxonomy
 
 Each sample includes a coarse `metadata.category` and three additional taxonomy fields:
@@ -63,15 +93,39 @@ Each sample includes a coarse `metadata.category` and three additional taxonomy
 
 The taxonomy is deterministic and source-aware. MultiWOZ uses domain/service annotations, Taskmaster uses scenario/file metadata when available, and DailyDialog uses lightweight keyword and dialogue-pattern rules.
 
+## v0.3 Quality Metadata
+
+Each row includes additional metadata fields:
+
+- `metadata.safety_category`
+- `metadata.safety_flags`
+- `metadata.is_safe_for_training`
+- `metadata.grounding_flags`
+- `metadata.slot_grounding_score`
+- `metadata.has_grounding_issue`
+- `metadata.target_answer_flags`
+- `metadata.target_answer_quality`
+- `metadata.taxonomy_confidence`
+- `metadata.taxonomy_evidence`
+- `metadata.quality_tier`
+
+Recommended default training filter:
+
+```python
+row["is_high_quality"] is True
+and row["metadata"]["is_safe_for_training"] is True
+and row["metadata"]["quality_tier"] in ["gold", "silver"]
+```
+
 ## Reasoning
 
 Streaming reasoning is generated by deterministic rule-based state tracking over turn-level chunks. DailyDialog rows focus on daily intent, tone, and continuity. MultiWOZ and Taskmaster rows use task-oriented templates for goal, known constraints, missing information, scenario/domain, and next-step policy. Deep reasoning is a compact global summary from the final tracked state, dialogue history, and target answer. The answer is not rewritten by default; it comes from the original next assistant turn.
 
 ## Quality Filtering
 
-The quality checks are category-aware. Long task-oriented conversations are no longer penalized in the same way as short daily dialogues. Some heuristic checks, including weak final state and premature response detection, are kept as diagnostic warnings rather than hard filters.
+The quality checks are category-aware and now include safety, grounding, target usefulness, and tiering. Long task-oriented conversations are no longer penalized in the same way as short daily dialogues. Some heuristic checks, including weak final state and premature response detection, are kept as diagnostic warnings rather than hard filters.
 
-`quality_flags` and `metadata.quality_checks` support filtering by real-source status, multi-turn context, non-empty reasoning, placeholder detection, category-aware length checks, malformed-row checks, repetition checks, and role alternation. Raw external data is not committed to git; processed train/eval files are intended for upload to `skyzhou06/LifeMultiTurnStreamingCoT`.
+`quality_flags` and `metadata.quality_checks` support filtering by real-source status, multi-turn context, non-empty reasoning, placeholder detection, category-aware length checks, malformed-row checks, repetition checks, grounding checks, safety checks, target-answer checks, and role alternation. Raw external data is not committed to git; processed train/eval files are intended for upload to `skyzhou06/LifeMultiTurnStreamingCoT`.
 
 ## Leakage Control
 

dataset_info.json

@@ -1,6 +1,26 @@
 {
-  "avg_num_chunks": 9.706,
-  "avg_num_turns": 9.706,
+  "avg_num_chunks": 9.705,
+  "avg_num_turns": 9.705,
+  "build_stats": {
+    "candidate_rows_before_limit": 215796,
+    "cleaned_dialogues": 40201,
+    "constructed_samples": 216519,
+    "filtered_dialogues": 77,
+    "raw_dialogues": 40278,
+    "released_rows": 30000,
+    "safety_filtered_reason_distribution": {
+      "dangerous_activity": 89,
+      "explicit_criminal_instruction": 340,
+      "financial_context": 8,
+      "illegal_drug_sale_or_purchase": 5,
+      "medical_or_health_context": 23,
+      "mild_violence_or_conflict_context": 27,
+      "self_harm_or_suicide": 3,
+      "sexual_or_adult_content": 7,
+      "weapons_or_violent_threat": 279
+    },
+    "safety_filtered_rows": 723
+  },
   "category_distribution": {
     "daily_dialogue": 10000,
     "task_oriented_dialogue": 20000
@@ -8,101 +28,154 @@
   "dataset_name": "LifeMultiTurnStreamingCoT",
   "domain_category_distribution": {
     "customer_service": 204,
-    "education_career": 2287,
+    "education_career": 2288,
     "finance_business": 180,
-    "food_dining": 1263,
-    "general_daily_life": 2879,
-    "health_wellness": 220,
-    "home_services": 227,
+    "food_dining": 1262,
+    "general_daily_life": 2881,
+    "health_wellness": 221,
+    "home_services": 226,
     "hospitality_lodging": 404,
-    "personal_schedule": 683,
-    "shopping_retail": 530,
-    "social_relationship": 1986,
-    "technology_support": 195,
+    "personal_schedule": 688,
+    "shopping_retail": 527,
+    "social_relationship": 1984,
+    "technology_support": 193,
     "travel_transportation": 18942
   },
   "domain_distribution": {
-    "cooking": 1027,
-    "customer_service": 192,
-    "daily_advice": 726,
+    "cooking": 1026,
+    "customer_service": 195,
+    "daily_advice": 722,
     "emotional_support": 131,
-    "fitness": 226,
+    "fitness": 229,
     "health_routine": 221,
-    "home": 535,
+    "home": 538,
     "other_life": 2584,
     "personal_finance": 135,
-    "schedule": 2908,
-    "shopping": 1321,
-    "social_planning": 409,
-    "study": 1053,
-    "travel": 18532
+    "schedule": 2902,
+    "shopping": 1317,
+    "social_planning": 412,
+    "study": 1058,
+    "travel": 18530
+  },
+  "eval_rows": 5789,
+  "grounding_flag_distribution": {
+    "generic_number_misread_as_budget": 14672,
+    "percent_misread_as_budget": 8,
+    "suspicious_numeric_slot": 14679,
+    "weak_slot_evidence": 1625
   },
-  "eval_rows": 5785,
-  "high_quality_eval_rows": 4899,
-  "high_quality_percentage": 0.8517,
-  "high_quality_rows": 25552,
-  "high_quality_train_rows": 20653,
+  "high_quality_eval_rows": 1845,
+  "high_quality_percentage": 0.3337,
+  "high_quality_rows": 10012,
+  "high_quality_train_rows": 8167,
   "intent_category_distribution": {
-    "booking_or_reservation": 15759,
+    "booking_or_reservation": 15758,
     "confirmation_clarification": 2710,
     "customer_support": 1241,
     "emotional_support": 232,
-    "information_request": 6236,
+    "information_request": 6242,
     "instruction_following": 26,
     "negotiation_decision": 139,
-    "planning_coordination": 1461,
-    "problem_solving": 445,
+    "planning_coordination": 1460,
+    "problem_solving": 440,
     "recommendation": 739,
-    "small_talk": 1012
+    "small_talk": 1013
   },
   "quality_flag_distribution": {
-    "answer_not_grounded": 1292,
+    "answer_not_grounded": 1642,
     "deep_reasoning_too_long": 1,
     "excessive_repetition": 442,
-    "generic_answer": 333,
-    "premature_respond": 4690,
+    "financial_context": 624,
+    "fragment_answer": 2611,
+    "generic_answer": 359,
+    "generic_number_misread_as_budget": 14672,
+    "grounding_issue": 14679,
+    "low_information_answer": 5011,
+    "medical_or_health_context": 327,
+    "mild_violence_or_conflict_context": 307,
+    "off_topic_answer": 534,
+    "percent_misread_as_budget": 8,
+    "premature_respond": 4693,
     "repeated_turns": 442,
+    "sensitive_content": 1257,
+    "suspicious_numeric_slot": 14679,
     "target_leakage": 176,
     "too_many_turns": 262,
+    "too_short_answer": 2759,
     "too_short_average_turn": 8,
-    "weak_final_state": 8627,
-    "weak_target_answer": 2756
+    "weak_final_state": 22334,
+    "weak_slot_evidence": 1625,
+    "weak_target_answer": 2759
+  },
+  "quality_tier_distribution": {
+    "bronze": 2829,
+    "drop": 17159,
+    "gold": 6921,
+    "silver": 3091
+  },
+  "safety_category_distribution": {
+    "safe": 28743,
+    "sensitive": 1257
+  },
+  "safety_flag_distribution": {
+    "financial_context": 624,
+    "medical_or_health_context": 327,
+    "mild_violence_or_conflict_context": 307
   },
   "scenario_category_distribution": {
     "attraction_search": 455,
-    "banking_support": 212,
+    "banking_support": 210,
     "customer_complaint": 218,
-    "family_conversation": 603,
+    "family_conversation": 600,
     "flight_booking": 9926,
     "food_ordering": 18,
-    "friend_conversation": 459,
+    "friend_conversation": 456,
     "general_conversation": 3288,
-    "home_repair": 97,
+    "home_repair": 101,
     "hotel_booking": 1557,
     "hotel_search": 184,
     "insurance_support": 95,
     "job_interview": 104,
-    "medical_assistance": 290,
+    "medical_assistance": 291,
     "movie_ticketing": 218,
     "music_search": 141,
     "restaurant_booking": 1890,
     "restaurant_search": 819,
-    "schedule_planning": 642,
-    "school_life": 854,
-    "shopping_assistance": 395,
-    "taxi_booking": 1106,
-    "technical_support": 460,
+    "schedule_planning": 646,
+    "school_life": 853,
+    "shopping_assistance": 393,
+    "taxi_booking": 1109,
+    "technical_support": 453,
     "train_booking": 4317,
-    "travel_planning": 252,
-    "workplace_conversation": 1400
+    "travel_planning": 253,
+    "workplace_conversation": 1405
   },
   "source_distribution": {
     "DailyDialog": 10000,
     "MultiWOZ": 10000,
     "Taskmaster": 10000
   },
+  "target_answer_flag_distribution": {
+    "answer_not_grounded": 1611,
+    "fragment_answer": 2611,
+    "generic_answer": 350,
+    "low_information_answer": 5011,
+    "off_topic_answer": 534,
+    "too_short_answer": 2759
+  },
+  "target_answer_quality_distribution": {
+    "drop": 3848,
+    "strong": 15196,
+    "usable": 6828,
+    "weak": 4128
+  },
+  "taxonomy_confidence_distribution": {
+    "high": 20850,
+    "medium": 9150
+  },
   "total_rows": 30000,
-  "train_rows": 24215,
+  "train_rows": 24211,
   "unknown_other_ratio": 0.0,
-  "version": "v0.1"
+  "version": "v0.3",
+  "version_label": "v0.3 — Safety and Grounding Quality Release"
 }

eval.jsonl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dcdd49b0e1a177ea2f5d321c3eb358a6ca7cedb178c18a53119a7f8f0e1328b3
-size 82708681
+oid sha256:9d72ae3fdaf48eab7f3776f9062d601cb835db334ff56bd4bfff195c7e34c4ee
+size 86682090

train.jsonl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ec644e4690c4aa89f309dc77f80920c895b907237bb049691a6b76d5f3c4b0ab
-size 348843497
+oid sha256:5315c5db299e46157cc21429bdc53b33a9b801e2e39380c75443a565d47e5d49
+size 365469434