Update croissant.json

croissant.json

@@ -42,20 +42,26 @@
     "separator": "cr:separator",
     "source": "cr:source",
     "subField": "cr:subField",
-    "transform": "cr:transform"
+    "transform": "cr:transform",
+    "prov": "http://www.w3.org/ns/prov#"
   },
   "@type": "sc:Dataset",
   "name": "feedback-forensics-annotations",
   "description": "Annotation data from Feedback Forensics experiments. See GitHub repo for generating code: https://github.com/rdnfn/feedback-forensics",
   "conformsTo": "http://mlcommons.org/croissant/1.0",
-  "url": "https://huggingface.co/datasets/rdnfn/feedback-forensics-annotations/",
+  "url": "https://huggingface.co/datasets/rdnfn/ff-annotations",
+  "license": "https://opendatacommons.org/licenses/by/1-0/",
+  "datePublished": "2025-05-16",
+  "dateModified": "2026-11-07",
+  "version": "1.0",
+  "citeAs": "@inproceedings{findeis2026feedbackforensics, title={Feedback Forensics: A Toolkit to Measure AI Personality}, author={Findeis, Arduin and Kaufmann, Timo and H{\\\"u}llermeier, Eyke and Mullins, Robert}, booktitle={Advances in Neural Information Processing Systems (Evaluations and Datasets Track)}, volume={40}, year={2026}}",
   "distribution": [
     {
       "@type": "cr:FileObject",
       "@id": "repo",
       "name": "repo",
       "description": "The Hugging Face git repository.",
-      "contentUrl": "https://huggingface.co/datasets/rdnfn/feedback-forensics-annotations/tree/main",
+      "contentUrl": "https://huggingface.co/datasets/rdnfn/ff-annotations/tree/main",
       "encodingFormat": "git+https",
       "sha256": "main"
     },
@@ -70,5 +76,71 @@
       "encodingFormat": "application/json",
       "includes": "data/*.json"
     }
+  ],
+  "rai:dataLimitations": "All measurements are relative to the underlying distribution of prompts and responses. When measuring personality in a model, the strength of a trait is relative to the reference models it is compared to. Similarly, the personality traits encouraged by a feedback dataset depend on its distribution of prompts and responses, and the same annotators may encourage different traits in different contexts. Trait annotations are made by AI annotators (LLM-as-a-Judge). Trait agreement analysis shows consistent annotator behaviour across related traits (paper, Appendix G.1), and annotations agree well with human judgements (paper, Appendix G.2), but LLM judges may introduce their own biases. Results also depend on the prompting and sampling strategy and vary by trait. For some value-related traits, annotation may be inherently ambiguous and therefore noisy, and we encourage manual inspection alongside the data. Finally, correlation does not imply causation: a correlation between preference labels and a trait does not mean that the original annotators followed that trait as a criterion.",
+  "rai:dataBiases": "The AI annotators may introduce biases of their own. The underlying prompt and interaction datasets are not representative of the general public. The human trait annotations come from two of the authors.",
+  "rai:personalSensitiveInformation": "The dataset does not contain such information to the best of our knowledge.",
+  "rai:dataUseCases": "Helping evaluate the personality of AI models and the development of future methods for tracking related behavioural traits.",
+  "rai:dataSocialImpact": "Positive: helps mitigate potential negative effects of misaligned personalities.",
+  "rai:hasSyntheticData": true,
+  "prov:wasDerivedFrom": [
+    {
+      "@id": "https://hf.co/datasets/lmarena-ai/arena-human-preference-100k",
+      "prov:label": "General Chatbot Arena data (arena-human-preference-100k)",
+      "sc:license": "User prompts CC-BY-4.0, model outputs governed by the terms of use of model providers"
+    },
+    {
+      "@id": "https://huggingface.co/spaces/lmarena-ai/Llama-4-Maverick-03-26-Experimental_battles/blob/main/data/clean-llama4.jsonl",
+      "prov:label": "Llama-4-Maverick arena data (Llama-4-Maverick-03-26-Experimental_battles)",
+      "sc:license": "User prompts CC-BY-4.0, model outputs governed by the terms of use of model providers"
+    },
+    {
+      "@id": "https://huggingface.co/datasets/allenai/multipref",
+      "prov:label": "MultiPref (Miranda et al., 2025)",
+      "sc:license": "ODC-By"
+    },
+    {
+      "@id": "https://huggingface.co/datasets/lmsys/chatbot_arena_conversations",
+      "prov:label": "Chatbot Arena conversations, as included in MultiPref",
+      "sc:license": "Prompts CC-BY-4.0, model outputs CC-BY-NC-4.0"
+    },
+    {
+      "@id": "https://huggingface.co/datasets/allenai/WildChat-1M",
+      "prov:label": "WildChat (Zhao et al., 2024), as included in MultiPref",
+      "sc:license": "ODC-By"
+    },
+    {
+      "@id": "https://github.com/anthropics/hh-rlhf",
+      "prov:label": "Anthropic Harmless/Helpful (Bai et al., 2022), as included in MultiPref",
+      "sc:license": "MIT"
+    },
+    {
+      "@id": "https://huggingface.co/datasets/HannahRoseKirk/prism-alignment",
+      "prov:label": "PRISM (Kirk et al., 2024)",
+      "sc:license": "Human-written texts CC-BY-4.0, model responses CC-BY-NC-4.0 and subject to the terms of use of model providers"
+    },
+    {
+      "@id": "https://huggingface.co/datasets/rdnfn/ff-model-personality",
+      "prov:label": "Feedback Forensics model personality data (model generations for the model comparisons)",
+      "sc:license": "ODC-By"
+    }
+  ],
+  "prov:wasGeneratedBy": [
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q109719325"
+      },
+      "prov:label": "AI annotation",
+      "sc:description": "Personality trait annotation of response pairs with the Feedback Forensics toolkit and a fixed set of personality traits, using Gemini-2.5-Flash as AI annotator. For comparison with the human labels, data/human_personality_annotations.json additionally contains annotations by GPT-4o-mini, GPT-4.1-mini, GPT-5-mini and multi-vote Gemini-2.5-Flash."
+    },
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q109719325"
+      },
+      "prov:label": "Human annotation",
+      "sc:description": "Two of the authors independently labelled ten traits on 101 randomly sampled general Chatbot Arena comparisons, first judging whether a trait applies and then which response expresses it more, without seeing AI annotations. data/human_personality_annotations.json holds both labellers' labels, their soft unanimous aggregate, and the AI annotations they are compared with."
+    }
   ]
-}
+}