Nora Petrova commited on
Commit
6688953
·
1 Parent(s): db3726a

Initial dataset upload: HUMAINE evaluation data with CSV and Parquet formats

Browse files
.gitattributes CHANGED
@@ -57,3 +57,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
57
  # Video files - compressed
58
  *.mp4 filter=lfs diff=lfs merge=lfs -text
59
  *.webm filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
57
  # Video files - compressed
58
  *.mp4 filter=lfs diff=lfs merge=lfs -text
59
  *.webm filter=lfs diff=lfs merge=lfs -text
60
+ feedback_dataset.csv filter=lfs diff=lfs merge=lfs -text
61
+ conversations_metadata_dataset.parquet filter=lfs diff=lfs merge=lfs -text
62
+ feedback_dataset.parquet filter=lfs diff=lfs merge=lfs -text
63
+ conversations_metadata_dataset.csv filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -1,3 +1,193 @@
1
- ---
2
- license: mit
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ task_categories:
4
+ - text-classification
5
+ - question-answering
6
+ language:
7
+ - en
8
+ tags:
9
+ - human-ai-interaction
10
+ - model-evaluation
11
+ - preference-learning
12
+ - conversational-ai
13
+ pretty_name: HUMAINE Human-AI Interaction Evaluation Dataset
14
+ size_categories:
15
+ - 100K<n<1M
16
+ dataset_info:
17
+ features:
18
+ - name: conversations_metadata
19
+ struct:
20
+ - name: conversation_id
21
+ dtype: int64
22
+ - name: model_name
23
+ dtype: string
24
+ - name: task_type
25
+ dtype: string
26
+ - name: domain
27
+ dtype: string
28
+ - name: task_complexity_score
29
+ dtype: int64
30
+ - name: goal_achievement_score
31
+ dtype: int64
32
+ - name: user_engagement_score
33
+ dtype: int64
34
+ - name: total_messages
35
+ dtype: int64
36
+ - name: feedback_comparisons
37
+ struct:
38
+ - name: conversation_id
39
+ dtype: int64
40
+ - name: model_a
41
+ dtype: string
42
+ - name: model_b
43
+ dtype: string
44
+ - name: metric
45
+ dtype: string
46
+ - name: choice
47
+ dtype: string
48
+ - name: age
49
+ dtype: int64
50
+ - name: ethnic_group
51
+ dtype: string
52
+ - name: political_affilation
53
+ dtype: string
54
+ - name: country_of_residence
55
+ dtype: string
56
+ ---
57
+
58
+ # HUMAINE: Human-AI Interaction Evaluation Dataset
59
+
60
+ ## Dataset Description
61
+
62
+ ### Dataset Summary
63
+
64
+ The HUMAINE dataset contains human evaluations of AI model interactions across diverse demographic groups and conversation contexts. This dataset powers the [HUMAINE Leaderboard](https://huggingface.co/spaces/ProlificAI/humaine-leaderboard), providing insights into how different AI models perform across various user populations and use cases.
65
+
66
+ The dataset consists of two main components:
67
+ - **Conversations Metadata**: 40,332 conversations with task complexity, achievement, and engagement scores
68
+ - **Feedback Comparisons**: 105,220 pairwise model comparisons across multiple evaluation metrics
69
+
70
+ **Note**: There may be a slight discrepancy between the numbers in this dataset and the leaderboard app due to changes in consent related to data release and the post-processing steps involved in preparing this dataset.
71
+
72
+ ### Supported Tasks
73
+
74
+ - Model performance evaluation
75
+ - Demographic bias analysis
76
+ - Preference learning
77
+ - Human-AI interaction research
78
+ - Conversational AI benchmarking
79
+
80
+ ## Dataset Structure
81
+
82
+ ### Data Files
83
+
84
+ The dataset contains two CSV files:
85
+
86
+ 1. **`conversations_metadata_dataset.csv`** (40,332 rows)
87
+ - Metadata about individual conversations between users and AI models
88
+ - Includes task types, domains, and performance scores
89
+
90
+ 2. **`feedback_dataset.csv`** (105,220 rows)
91
+ - Pairwise comparisons between different AI models
92
+ - Includes demographic information and preference choices
93
+
94
+ ### Data Fields
95
+
96
+ #### Conversations Metadata
97
+ - `conversation_id`: Unique identifier for the conversation
98
+ - `model_name`: Name of the AI model used
99
+ - `task_type`: Type of task (information_seeking, technical_assistance, etc.)
100
+ - `domain`: Domain of the conversation (health_medical, technology, travel, etc.)
101
+ - `task_complexity_score`: Complexity rating (1-5)
102
+ - `goal_achievement_score`: How well the goal was achieved (1-5)
103
+ - `user_engagement_score`: User engagement level (1-5)
104
+ - `total_messages`: Total number of messages in the conversation
105
+
106
+ #### Feedback Comparisons
107
+ - `conversation_id`: Unique identifier linking to conversation metadata
108
+ - `model_a`: First model in the comparison
109
+ - `model_b`: Second model in the comparison
110
+ - `metric`: Evaluation metric (overall_winner, trust_ethics_and_safety, core_task_performance_and_reasoning, interaction_fluidity_and_adaptiveness)
111
+ - `choice`: User's choice (A, B, or tie)
112
+ - `age`: Age of the evaluator
113
+ - `ethnic_group`: Ethnic group of the evaluator
114
+ - `political_affilation`: Political affiliation of the evaluator
115
+ - `country_of_residence`: Country of residence of the evaluator
116
+
117
+ ## Usage
118
+
119
+ This dataset contains two CSV files that can be joined on the `conversation_id` field:
120
+ - `conversations_metadata_dataset.csv`: Metadata about each conversation
121
+ - `feedback_dataset.csv`: Pairwise model comparisons with demographic information
122
+
123
+ Both files are included in this single dataset repository and can be accessed using HuggingFace's dataset loading utilities.
124
+
125
+ ## Dataset Creation
126
+
127
+ ### Curation Rationale
128
+
129
+ This dataset was created to address the lack of diverse, demographically-aware evaluation data for AI models. It captures real-world human preferences and interactions across different population groups, enabling more inclusive AI development.
130
+
131
+ ### Source Data
132
+
133
+ Data was collected through structured human evaluation tasks where participants:
134
+ 1. Engaged in conversations with various AI models
135
+ 2. Provided pairwise comparisons between model outputs
136
+ 3. Rated conversations on multiple quality dimensions (metrics)
137
+
138
+ ### Annotations
139
+
140
+ All annotations were provided by human evaluators through the Prolific platform, ensuring demographic diversity and high-quality feedback.
141
+
142
+ ### Personal and Sensitive Information
143
+
144
+ The dataset contains aggregated demographic information (age groups, ethnic groups, political affiliations, countries) but no personally identifiable information. All data has been anonymized and aggregated to protect participant privacy.
145
+
146
+ ## Considerations for Using the Data
147
+
148
+ ### Social Impact
149
+
150
+ This dataset aims to promote more inclusive AI development by highlighting performance differences across demographic groups. It should be used to improve AI systems' fairness and effectiveness for all users.
151
+
152
+ ### Discussion of Biases
153
+
154
+ While efforts were made to ensure demographic diversity, the dataset may still contain biases related to:
155
+ - Geographic representation (primarily US and UK participants)
156
+ - Self-selection bias in participant recruitment
157
+ - Cultural and linguistic factors affecting evaluation criteria
158
+
159
+ ### Other Known Limitations
160
+
161
+ - Limited to English-language interactions
162
+ - Demographic categories are self-reported
163
+ - Temporal bias (models evaluated at specific points in time)
164
+
165
+ ## Additional Information
166
+
167
+ ### Dataset Curators
168
+
169
+ This dataset was curated by the Prolific AI team as part of the HUMAINE (Human-AI Interaction Evaluation) project.
170
+
171
+ ### Licensing Information
172
+
173
+ This dataset is released under the MIT License.
174
+
175
+ ### Citation Information
176
+
177
+ ```bibtex
178
+ @dataset{humaine2025,
179
+ title={HUMAINE: Human-AI Interaction Evaluation Dataset},
180
+ author={Prolific AI Team},
181
+ year={2025},
182
+ publisher={Hugging Face},
183
+ url={https://huggingface.co/datasets/ProlificAI/humaine-evaluation-dataset}
184
+ }
185
+ ```
186
+
187
+ ### Contributions
188
+
189
+ Thanks to all the human evaluators who contributed their feedback to this project!
190
+
191
+ ## Contact
192
+
193
+ For questions or feedback about this dataset, please visit the [HUMAINE Leaderboard](https://huggingface.co/spaces/ProlificAI/humaine-leaderboard) or contact the Prolific AI team.
conversations_metadata_dataset.csv ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b1b9936da135f061ea381ced9b562034edd9383f2c3e0b43228b6acee2e43d69
3
+ size 2645059
conversations_metadata_dataset.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:957c0e3c11b768d158e66bb053ad5096608dc1f80fc36e69e36b035194dc13c8
3
+ size 302115
feedback_dataset.csv ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2b9dec54e5620c35aed0534cd1d3bed3620a556a74b182729a743d2ec3a5be1f
3
+ size 11973022
feedback_dataset.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bbca035715e13e879af5e0881df43d5e9407867860ebcfce6e3a165b8631acbf
3
+ size 688645