Insight Generation on Internal non-scientific document collections Instagram Marketing
4.76Set-level Score (Gemini-2.5-Flash Judge)INSIGHTGEN
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| INSIGHTGENBase Model=GPT-4o2026.04 | 4.76 | 3.67 | |
| INSIGHTGENBase Model=Claude-3.5-Sonnet2026.04 | 4.38 | 3.43 | |
| FAISS+CoTBase Model=Claude-3.5-Sonnet2026.04 | 4.24 | 2.62 | |
| GPT+CoTBase Model=Claude-3.5-Sonnet2026.04 | 3.71 | 2.48 | |
| Direct GPTBase Model=GPT-4o2026.04 | 3.62 | 2.29 | |
| FAISSBase Model=GPT-4o2026.04 | 3.33 | 2.24 | |
| GPT+CoTBase Model=GPT-4o2026.04 | 3.24 | 2.52 | |
| FAISS+CoTBase Model=GPT-4o2026.04 | 3.1 | 2.19 | |
| Direct GPTBase Model=Claude-3.5-Sonnet2026.04 | 3 | 1.71 | |
| FAISSBase Model=Claude-3.5-Sonnet2026.04 | 1.43 | 1.52 |