Datasets:
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# InteractComp (Encrypted Release)
|
| 2 |
|
| 3 |
InteractComp accompanies the paper **"INTERACTCOMP: Evaluating Search Agents with Ambiguous Queries"** (Deng et al., 2025). The benchmark targets a capability gap in modern search agents: recognizing ambiguity, asking clarifying questions, and only then executing retrieval or answering. Despite rapid progress on fully specified web queries, the paper shows that interaction-centric performance remains stagnant—highlighting InteractComp as both an evaluation suite and a training ground for interactive behaviors.
|
|
@@ -55,4 +74,4 @@ Decrypted records should be valid UTF-8 JSON lines matching the original dataset
|
|
| 55 |
- Consider producing per-recipient encrypted variants (unique tokens) when auditing access or tracing potential leaks.
|
| 56 |
|
| 57 |
## License & Intended Use
|
| 58 |
-
This release is intended for non-commercial research on interactive search agents. Any downstream use should comply with the policies outlined in the accompanying paper and institutional review requirements. Contact the maintainers for questions about redistribution or additional licensing terms.
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
task_categories:
|
| 4 |
+
- question-answering
|
| 5 |
+
language:
|
| 6 |
+
- en
|
| 7 |
+
- zh
|
| 8 |
+
tags:
|
| 9 |
+
- music
|
| 10 |
+
- chemistry
|
| 11 |
+
- biology
|
| 12 |
+
- finance
|
| 13 |
+
- legal
|
| 14 |
+
- art
|
| 15 |
+
- medical
|
| 16 |
+
- agent
|
| 17 |
+
size_categories:
|
| 18 |
+
- n<1K
|
| 19 |
+
---
|
| 20 |
# InteractComp (Encrypted Release)
|
| 21 |
|
| 22 |
InteractComp accompanies the paper **"INTERACTCOMP: Evaluating Search Agents with Ambiguous Queries"** (Deng et al., 2025). The benchmark targets a capability gap in modern search agents: recognizing ambiguity, asking clarifying questions, and only then executing retrieval or answering. Despite rapid progress on fully specified web queries, the paper shows that interaction-centric performance remains stagnant—highlighting InteractComp as both an evaluation suite and a training ground for interactive behaviors.
|
|
|
|
| 74 |
- Consider producing per-recipient encrypted variants (unique tokens) when auditing access or tracing potential leaks.
|
| 75 |
|
| 76 |
## License & Intended Use
|
| 77 |
+
This release is intended for non-commercial research on interactive search agents. Any downstream use should comply with the policies outlined in the accompanying paper and institutional review requirements. Contact the maintainers for questions about redistribution or additional licensing terms.
|