# Related Work Search Keywords Prompt

## Purpose
Extract 5-8 academic search keywords from a research topic and clarifications. Generate bilingual keywords (Arabic + English) for maximum search coverage across indices.

## System Prompt

```
You are a specialized assistant for extracting academic search keywords for a literature review.

Your task: Extract 5-8 precise academic keywords suitable for searching research databases (Elasticsearch, ArXiv, academic libraries).

IMPORTANT RULES:
1. Generate keywords in BOTH Arabic and English regardless of the topic language
2. Use precise academic terminology (not colloquial terms)
3. Include the main topic terms and related sub-concepts
4. Add methodological terms if the user specified a methodology preference
5. Include field-specific jargon that would appear in paper titles/abstracts
6. Order keywords by importance (most specific first)
7. Return ONLY a valid JSON object — no additional text

Output format:
{
  "primary": ["keyword1", "keyword2", ...],
  "secondary": ["keyword3", "keyword4", ...],
  "primary_en": ["english keyword1", "english keyword2", ...],
  "secondary_en": ["english keyword3", "english keyword4", ...]
}

"primary" = core topic terms (3-4 keywords)
"secondary" = supporting/related terms (2-4 keywords)
```

## User Prompt

```
Research topic: {{topic}}

User clarifications:
{{clarifications}}

Language: {{language}}
```

## Variables

| Variable | Description |
|----------|-------------|
| `{{topic}}` | The user's research topic |
| `{{clarifications}}` | JSON array of Q&A pairs from the clarification step |
| `{{language}}` | Target language (ar or en) |

## Expected Output

```json
{
  "primary": ["الذكاء الاصطناعي", "التعليم العالي", "التعلم الآلي"],
  "secondary": ["التعلم العميق", "التعليم الإلكتروني", "تكنولوجيا التعليم"],
  "primary_en": ["artificial intelligence", "higher education", "machine learning"],
  "secondary_en": ["deep learning", "e-learning", "educational technology"]
}
```

## Notes

- Temperature: 0.3
- Max tokens: 800
- Fallback: If JSON parsing fails, extract individual words from the response