Datasets:
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,41 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
task_categories:
|
| 4 |
+
- question-answering
|
| 5 |
+
tags:
|
| 6 |
+
- finance
|
| 7 |
+
- insurance
|
| 8 |
+
size_categories:
|
| 9 |
+
- 10K<n<100K
|
| 10 |
+
---
|
| 11 |
+
# INS-MMBench Dataset
|
| 12 |
+
|
| 13 |
+
**INS-MMBench** INS-MMBench is the first comprehensive LVLMs benchmark for the insurance domain, proposed in our paper:
|
| 14 |
+
|
| 15 |
+
> **INS-MMBench: A Comprehensive Benchmark for Evaluating LVLMs' Performance in Insurance**
|
| 16 |
+
|
| 17 |
+
[📄Paper Link](https://arxiv.org/pdf/2406.09105) | [🐙GitHub Repository](https://github.com/FDU-INS/INS-MMBench)
|
| 18 |
+
|
| 19 |
+
---
|
| 20 |
+
|
| 21 |
+
## Overview
|
| 22 |
+
|
| 23 |
+
INS-MMBench is the first comprehensive LVLMs benchmark for the insurance domain, it covers four representative insurance types: auto, property, health, and agricultural insurance and key insurance stages such as risk underwriting, risk monitoring and claim processing. INS-MMBench consists of three layers task:
|
| 24 |
+
- Fundamental task, which focuses on the understanding of individual insurance-related visual elements;
|
| 25 |
+
- Meta-task, which involves the compositional understanding of multiple insurance-related visual elements;
|
| 26 |
+
- Scenario task, which pertains to real-world insurance tasks requiring multi-step reasoning and decision-making.
|
| 27 |
+
|
| 28 |
+
INS-MMBench includes a total of 12,052 images, 10,372 thoroughly designed questions (including multiple-choice visual questions and free-text visual questions), comprehensively covering 5 scenario tasks, 12 meta-tasks and 22 fundamental tasks.
|
| 29 |
+
|
| 30 |
+
This dataset includes **six TSV files**, each corresponding to a specific task described in our paper:
|
| 31 |
+
|
| 32 |
+
| File Name | Task Type |
|
| 33 |
+
|-----------|-----------|
|
| 34 |
+
| `INS_MMBench_fundamental.tsv` | Fundamental Task |
|
| 35 |
+
| `multi_step_claim.tsv` | Scenario Task - Auto Insurance Claim Processing |
|
| 36 |
+
| `multi_step_liability.tsv` | Scenario Task - Auto Insurance Accident Liability Determination |
|
| 37 |
+
| `multi_step_health.tsv` | Scenario Task - Health Insurance Risk Assessment |
|
| 38 |
+
| `multi_step_property.tsv` | Scenario Task - Property Insurance Risk Management |
|
| 39 |
+
| `multi_step_agri.tsv` | Scenario Task - Agricultural Insurance Claim Processing |
|
| 40 |
+
|
| 41 |
+
---
|