Description

This PR fixes critical data parsing and logic errors in the Government RAG benchmarking example that previously broke location-based filtering and caused the [other] mode to crash.

Changes

1. Location Extraction Overhaul (basemodel.py)

Issue

• Used os.path.basename(os.getcwd()) to determine province
• Always returned project root (e.g., "ianvs")
• Broke RAG filtering logic completely

Fix

• Added _load_locations_from_dataset(data) helper method

Strategy

• Primary: Reads .jsonl dataset and extracts province from level_4_dim
• Fallback: Uses NLP-based text matching + internal dictionary to detect and translate Chinese province names

2. Fixed Missing all_locations Attribute

Issue

• [other] mode used self.all_locations without initialization
• Caused AttributeError

Fix

• Properly initializes self.all_locations during location extraction
• Stores all valid provinces dynamically
• Enables correct inverse filtering logic

Testing Performed

• ✅ Verified correct province mapping per query (e.g., "Shanghai")
• ✅ Confirmed RAG filtering works as expected
• ✅ Tested [other] mode → no crashes, correct inverted dataset behavior

Type of Change

• Bug fix (non-breaking change)
• Performance improvement

Impact

• Restores correct RAG filtering behavior
• Fixes [other] mode crash
• Improves robustness of location detection across environments

Fixes : #379