Create README.md

README.md

@@ -1,3 +1,194 @@
----
-license: mit
----
+---
+license: mit
+task_categories:
+  - other
+language:
+  - en
+tags:
+  - browser-extension
+  - user-study
+  - human-computer-interaction
+  - web-agents
+  - within-subjects
+pretty_name: PageGuide User Study
+size_categories:
+  - n<1K
+---
+
+# PageGuide User Study Dataset
+
+This dataset contains raw interaction data from a controlled within-subjects user study evaluating **PageGuide** (XWebAgent), an AI-powered browser extension that helps users complete web tasks via natural language.
+
+Participants performed tasks in two conditions — **with** and **without** the extension — across three task types: `find`, `guide`, and `hide`. Collected metrics include task-completion times, accuracy scores, and post-study survey responses.
+
+---
+
+## Study Design
+
+| Property | Value |
+|---|---|
+| Design | Counterbalanced within-subjects |
+| Conditions | `extension` (PageGuide active) vs. `control` (no extension) |
+| Task types | `find` · `guide` · `hide` |
+| Primary metrics | Completion time, accuracy, survey ratings |
+
+**Task types**
+- **find** — locate or highlight specific information on a webpage
+- **guide** — follow step-by-step instructions to complete a multi-step web action
+- **hide** — filter or conceal unwanted content on a webpage
+
+Participants were randomly assigned a counterbalanced order so that each person experienced both conditions. Task questions are labelled `q0`, `q1`, `q2`, etc.
+
+---
+
+## Files
+
+### `tasks.csv` (2.29 MB)
+Raw per-interaction log — the main data file. Each row is one task attempt.
+
+| Column | Description |
+|---|---|
+| `session_id` | Unique participant session identifier |
+| `condition` | `extension` or `control` |
+| `task_type` | `find`, `guide`, or `hide` |
+| `question_id` | Task question index within its type (`q0`, `q1`, …) |
+| `start_time` | Unix timestamp (ms) when the task started |
+| `end_time` | Unix timestamp (ms) when the task ended |
+| `duration_s` | Elapsed time in seconds |
+| `completed` | Boolean — whether the participant marked the task complete |
+| `accuracy` | Graded accuracy score (0–1 or 0–100) for find/hide tasks |
+| `query` | The natural-language query the participant typed (extension condition only) |
+| `chat_turns` | Number of chat interactions in the extension condition |
+
+### `sessions.csv` (8.28 kB)
+One row per participant session — demographic and counterbalancing metadata.
+
+| Column | Description |
+|---|---|
+| `session_id` | Matches `tasks.csv` |
+| `participant_id` | Anonymised participant label |
+| `order` | Condition order assigned (`extension_first` or `control_first`) |
+| `started_at` | Session start timestamp |
+| `web_experience` | Self-reported web experience level |
+
+### `paired_times.csv` (10.8 kB)
+Pre-processed paired completion times — one row per participant × task, ready for paired statistical tests.
+
+| Column | Description |
+|---|---|
+| `participant_id` | Anonymised participant label |
+| `task_type` | `find`, `guide`, or `hide` |
+| `question_id` | Task question index |
+| `time_extension` | Completion time (s) in the extension condition |
+| `time_control` | Completion time (s) in the control condition |
+| `time_diff` | `time_control − time_extension` (positive = extension faster) |
+
+### `summary.csv` (85.2 kB)
+Aggregated per-participant × per-task-type summary statistics (mean time, accuracy, completion rate) for both conditions. Useful for quick group-level analysis.
+
+### `stats_results.csv` (227 bytes)
+Results of the paired statistical tests (Wilcoxon signed-rank / paired t-test) run on completion times and accuracy. One row per metric × task-type comparison.
+
+| Column | Description |
+|---|---|
+| `metric` | e.g., `duration_s`, `accuracy` |
+| `task_type` | `find`, `guide`, `hide`, or `all` |
+| `test` | Statistical test used |
+| `statistic` | Test statistic |
+| `p_value` | p-value |
+| `significant` | Boolean (α = 0.05) |
+
+### `survey_summary.csv` (415 bytes)
+Aggregated post-study questionnaire scores per condition. Covers perceived usability (SUS-style) and cognitive load (NASA-TLX-style) dimensions.
+
+| Column | Description |
+|---|---|
+| `condition` | `extension` or `control` |
+| `dimension` | Survey dimension name |
+| `mean` | Mean rating |
+| `std` | Standard deviation |
+| `n` | Number of responses |
+
+---
+
+## Quick Start
+
+### Load with the 🤗 `datasets` library (recommended)
+
+```python
+from datasets import load_dataset
+
+# Load individual files as named splits
+tasks    = load_dataset("ttn0011/pageguide_userstudy", data_files="tasks.csv",           split="train").to_pandas()
+paired   = load_dataset("ttn0011/pageguide_userstudy", data_files="paired_times.csv",    split="train").to_pandas()
+sessions = load_dataset("ttn0011/pageguide_userstudy", data_files="sessions.csv",        split="train").to_pandas()
+survey   = load_dataset("ttn0011/pageguide_userstudy", data_files="survey_summary.csv",  split="train").to_pandas()
+stats    = load_dataset("ttn0011/pageguide_userstudy", data_files="stats_results.csv",   split="train").to_pandas()
+
+# Mean completion time by condition and task type
+tasks.groupby(["condition", "task_type"])["duration_s"].mean()
+
+# Paired time difference (positive = extension faster)
+paired.groupby("task_type")["time_diff"].mean()
+
+# Survey ratings side by side
+survey.pivot(index="dimension", columns="condition", values="mean")
+```
+
+### Or load directly with pandas
+
+```python
+import pandas as pd
+
+BASE = "https://huggingface.co/datasets/ttn0011/pageguide_userstudy/resolve/main/"
+
+tasks    = pd.read_csv(BASE + "tasks.csv")
+paired   = pd.read_csv(BASE + "paired_times.csv")
+sessions = pd.read_csv(BASE + "sessions.csv")
+survey   = pd.read_csv(BASE + "survey_summary.csv")
+stats    = pd.read_csv(BASE + "stats_results.csv")
+```
+
+### Reproduce the paired-time plot
+
+```python
+import matplotlib.pyplot as plt
+from datasets import load_dataset
+
+paired = load_dataset("ttn0011/pageguide_userstudy", data_files="paired_times.csv", split="train").to_pandas()
+
+fig, axes = plt.subplots(1, 3, figsize=(12, 4), sharey=False)
+for ax, task in zip(axes, ["find", "guide", "hide"]):
+    subset = paired[paired["task_type"] == task]
+    for _, row in subset.iterrows():
+        ax.plot([0, 1], [row["time_control"], row["time_extension"]],
+                color="steelblue", alpha=0.4, linewidth=1)
+    ax.set_xticks([0, 1])
+    ax.set_xticklabels(["Control", "Extension"])
+    ax.set_title(task.capitalize())
+    ax.set_ylabel("Completion time (s)")
+plt.tight_layout()
+plt.savefig("paired_times.png", dpi=150)
+```
+
+---
+
+## Citation
+
+If you use this dataset, please cite the associated paper:
+
+```bibtex
+@misc{pageguide2025,
+  title  = {XWebAgent: Grounded, Verifiable Web Interaction via Highlighted Chain-of-Thought},
+  author = {Tin Nguyen and others},
+  year   = {2025},
+  note   = {User study data: \url{https://huggingface.co/datasets/ttn0011/pageguide_userstudy}}
+}
+```
+
+---
+
+## License
+
+MIT — see [LICENSE](LICENSE) for details.
+All participant data is anonymised. No personally identifiable information is included.