David Snelling
196690863d
- Fixed imports in examples/tests/ to use correct Brainy import - Fixed imports in tests/benchmarks/ to use correct paths - Updated bin/brainy-interactive.js to use Brainy instead of BrainyData - Corrected documentation references throughout codebase - Removed duplicate imports in benchmark files - All files now consistently use 'Brainy' class from dist/index.js
8.2 KiB
8.2 KiB
Natural Language Queries with Brainy
Current Status: Basic natural language support with 220+ patterns. Advanced NLP features coming in Q1 2025.
Brainy's find() method understands natural language, allowing you to query your data using plain English instead of complex query syntax.
Overview
The natural language processing (NLP) system is powered by 220+ pre-computed patterns that understand common query intents, temporal expressions, numeric comparisons, and domain-specific terminology.
What Works Today
The current NLP implementation supports:
- ✅ Basic pattern matching (220+ patterns)
- ✅ Simple temporal expressions ("recent", "today", "last week")
- ✅ Common query intents ("find", "show", "search")
- ✅ Domain keywords recognition
- ⚠️ Limited entity extraction
- ⚠️ Basic numeric comparisons
Basic Usage
import { Brainy } from 'brainy'
const brain = new Brainy()
await brain.init()
// Simply ask in natural language
const results = await brain.find("show me recent articles about AI")
Supported Query Types
✅ Currently Working
These query patterns are supported today.
⚠️ Basic Support
These work with limitations.
🚧 Coming Soon
Planned for future releases.
Temporal Queries ⚠️ Basic Support
// Relative time expressions
await brain.find("documents from last week")
await brain.find("posts created yesterday")
await brain.find("data from the past 30 days")
// Specific dates and ranges
await brain.find("articles published in Q3 2024")
await brain.find("reports from January to March")
await brain.find("meetings scheduled for tomorrow")
// Named periods
await brain.find("quarterly reports from this year")
await brain.find("summer vacation photos")
await brain.find("holiday sales data")
Numeric Filters ⚠️ Basic Support
// Comparisons
await brain.find("products with price under $100")
await brain.find("articles with more than 1000 views")
await brain.find("employees with salary above 75000")
// Ranges
await brain.find("items priced between $50 and $200")
await brain.find("posts with 10 to 50 likes")
await brain.find("companies with 100-500 employees")
// Percentages and metrics
await brain.find("stocks with growth over 20%")
await brain.find("projects with completion above 80%")
await brain.find("products with 5 star ratings")
Entity and Relationship Queries 🚧 Coming Soon
// People and organizations
await brain.find("articles by John Smith")
await brain.find("employees at TechCorp")
await brain.find("papers from Stanford University")
// Relationships
await brain.find("documents related to Project X")
await brain.find("products similar to iPhone")
await brain.find("people who work with Sarah")
// Ownership and attribution
await brain.find("repos owned by user123")
await brain.find("designs created by the marketing team")
await brain.find("patents filed by Apple")
Combined Complex Queries 🚧 Coming Soon
// Multiple conditions
await brain.find("verified research papers about machine learning from 2024 with high citations")
// Business queries
await brain.find("quarterly financial reports from Q2 2024 with revenue over 10M")
// Content queries
await brain.find("popular blog posts by tech influencers published this month about AI")
// E-commerce queries
await brain.find("electronics under $500 with 4+ star reviews and free shipping")
// Academic queries
await brain.find("peer-reviewed papers on quantum computing published in Nature after 2020")
How It Works
1. Pattern Matching
The NLP system first matches your query against 220+ pre-built patterns to identify:
- Query intent (search, filter, aggregate)
- Temporal expressions
- Numeric comparisons
- Entity mentions
- Relationship indicators
2. Entity Extraction
Named entities are extracted and classified:
- People names
- Organization names
- Product names
- Locations
- Dates and times
3. Intent Classification
The query intent is determined:
- Search: Finding similar items
- Filter: Applying specific criteria
- Aggregate: Grouping or summarizing
- Navigate: Following relationships
4. Query Construction
The natural language is converted to a structured Triple Intelligence query:
// Input: "recent AI papers with high citations"
// Output:
{
like: "AI papers",
where: {
type: "paper",
citations: { $gte: 100 },
published: { $gte: "2024-01-01" }
},
boost: "recent"
}
5. Execution
The structured query is executed using Triple Intelligence, combining:
- Vector similarity search
- Metadata filtering
- Graph traversal
Advanced Features
Contextual Understanding 🚧 Coming Soon
// Brainy understands context and synonyms
await brain.find("latest ML research") // Understands ML = Machine Learning
await brain.find("top rated items") // Understands top = high score/rating
await brain.find("trending topics") // Understands trending = recent + popular
Fuzzy Matching 🚧 Coming Soon
// Handles typos and variations
await brain.find("articals about blockchian") // Still finds blockchain articles
await brain.find("Jon Smith papers") // Matches "John Smith"
Domain-Specific Understanding ⚠️ Basic Support
// Tech domain
await brain.find("repos with MIT license")
await brain.find("APIs with OAuth support")
await brain.find("npm packages with zero dependencies")
// Business domain
await brain.find("SaaS companies with ARR over 1M")
await brain.find("startups in Series A")
await brain.find("B2B products with enterprise pricing")
// Academic domain
await brain.find("papers with h-index above 50")
await brain.find("journals with impact factor over 10")
await brain.find("conferences with double-blind review")
Fallback to Structured Queries
When natural language isn't sufficient, you can always use structured queries:
// Structured query for precise control
const results = await brain.find({
$and: [
{ $vector: { $similar: "neural networks", threshold: 0.8 } },
{ category: "research" },
{ year: { $gte: 2023 } },
{ $or: [
{ author: "LeCun" },
{ author: "Hinton" },
{ author: "Bengio" }
]}
],
limit: 50
})
Performance Tips
- Be specific: More specific queries execute faster
- Use proper nouns: Names and specific terms improve accuracy
- Include time frames: Temporal filters reduce search space
- Specify limits: Always include reasonable result limits
Examples by Use Case
Customer Support
await brain.find("urgent tickets from VIP customers today")
await brain.find("unresolved issues older than 3 days")
await brain.find("positive feedback about product X this month")
Content Management
await brain.find("draft posts scheduled for next week")
await brain.find("published articles needing review")
await brain.find("videos with over 10k views")
E-commerce
await brain.find("best selling products in electronics")
await brain.find("items with low stock under 10 units")
await brain.find("orders from California pending shipping")
Analytics
await brain.find("user sessions longer than 5 minutes yesterday")
await brain.find("conversion events from mobile users")
await brain.find("page views for /pricing in the last hour")
Limitations
While powerful, the NLP system has some limitations:
- Complex logic: Very complex boolean logic may require structured queries
- Ambiguity: Ambiguous queries may not parse as expected
- Domain terms: Highly specialized terminology may need training
- Languages: Currently optimized for English queries
Best Practices
- Start simple: Begin with simple queries and add complexity
- Test understanding: Use explain mode to see how queries are interpreted
- Provide feedback: Help improve the system by reporting misunderstood queries
- Combine approaches: Use NLP for exploration, structured for precision