Research Query Management

Create and manage automated research queries that run on schedule to keep users updated on topics they care about.

Tools to Use

For query management, use these tools:

Setting Up a New Recurring Search

Step 1: Understand User Needs

Ask these questions:

1. "What topic do you want to track?"
2. "How often do you want updates?" (daily/weekly)
3. "How many papers per update is manageable for you?"

Step 2: Build the Query

list_available_sources()  # Show options

create_research_question(
  title="Descriptive title for this search",
  keywords=["primary_term", "secondary_term", "synonym"],
  sources=["source1", "source2"],
  max_papers=15,
  relevance_threshold=0.7,
  schedule="daily"  # or "weekly"
)

Step 3: Test the Query

run_discovery_for_question(question_id="[new query ID]")

Review results with user:

• Are papers relevant?
• Too many/too few results?
• Missing important keywords?

Step 4: Confirm Setup

"Your recurring search is set up:

📋 **Query**: [title]
🔑 **Keywords**: [list]
📚 **Sources**: [list]
⏰ **Schedule**: [daily/weekly]
📊 **Expected**: ~[X] papers per run

The system will automatically search for new papers and add them
to your collection. You'll find new papers in your daily/weekly digest."

Refining Existing Queries

Diagnosis Questions

When user reports issues, ask:

Common Refinements

Too much noise:

get_research_question(question_id="[query ID]")

update_research_question(
  question_id="[query ID]",
  relevance_threshold=0.8,  # Raise from 0.7
  keywords=["more", "specific", "terms"]  # Add specificity
)

Missing papers:

update_research_question(
  question_id="[query ID]",
  relevance_threshold=0.65,  # Lower threshold
  sources=["add", "more", "sources"],  # Add sources
  keywords=["existing", "plus", "synonyms"]  # Add synonyms
)

Wrong domain:

update_research_question(
  question_id="[query ID]",
  keywords=["topic", "-exclude_term"],  # Negative keywords
  sources=["domain_specific_source"]  # Change sources
)

Schedule Recommendations

Query Health Check

Periodically review query performance:

list_research_questions()

For each query:
  get_research_question(question_id="...")

  Check:
  - Last run date
  - Papers found per run
  - User engagement (if tracked)

  Recommend adjustments if needed

Workflow Examples

Example 1: New Topic Tracking

User: "Keep me updated on new papers about state space models"

1. "I'll set up a recurring search for you. A few questions:
    - How often should I check for new papers? (daily/weekly)
    - How many papers can you review? (~10-15 daily or ~30-50 weekly)"

2. [User: "Weekly, about 30 papers"]

3. create_research_question(
     title="State Space Models Research Updates",
     keywords=["state space model", "SSM", "Mamba", "S4", "linear attention"],
     sources=["arxiv", "semantic_scholar"],
     max_papers=35,
     relevance_threshold=0.7,
     schedule="weekly"
   )

4. run_discovery_for_question(question_id="...")

5. "Found 28 papers in the initial test. Here are the top 5:
    [list]

    Does this look like what you're looking for?"

Example 2: Refining a Noisy Query

User: "My attention mechanisms search is giving me too many irrelevant papers"

1. list_research_questions()  # Find the query

2. get_research_question(question_id="attention-query-id")
   → Current settings: threshold=0.65, keywords=["attention"]

3. "I see your current search uses broad keywords. Let's refine:
    - What specific type of attention are you interested in?
    - Any domains you want to exclude?"

4. [User: "Just efficient attention for transformers, not general attention"]

5. update_research_question(
     question_id="attention-query-id",
     keywords=["efficient attention", "linear attention", "sparse attention", "flash attention"],
     relevance_threshold=0.75
   )

6. run_discovery_for_question(question_id="attention-query-id")

7. "Updated! Test run found 18 papers (down from 45).
    Here are the top 5 - are these more relevant?"

Example 3: Multiple Related Queries

User: "I'm researching LLM efficiency - help me set up comprehensive tracking"

1. "For comprehensive coverage, I recommend multiple focused queries:

   Query 1: Model Architecture Efficiency
   Query 2: Training Efficiency
   Query 3: Inference Optimization

   Each will capture different aspects. Shall I set these up?"

2. [Create each query with specific keywords]

3. "You now have 3 queries running weekly:
    - Architecture: ~15 papers/week
    - Training: ~20 papers/week
    - Inference: ~25 papers/week

    Total: ~60 papers/week. Manageable for your schedule?"

Response Template

For new query setup:

## Research Query Created ✓

**Title**: [query title]
**ID**: [query_id]

**Configuration**:
- Keywords: [list]
- Sources: [list]
- Schedule: [frequency]
- Max papers: [count]
- Relevance threshold: [value]

**Test Results**:
- Papers found: [count]
- Sample papers: [top 3]

**Next steps**:
- Query will run automatically on [schedule]
- Use `get_research_question(question_id="[ID]")` to check status
- Ask me to refine if results aren't quite right