mirror of
https://github.com/tiennm99/goclaw.git
synced 2026-10-15 14:16:30 +00:00
feat(kg): expand entity types and improve extraction prompt
Add 3 new entity types: technology, product, document — reducing concept catch-all bucket. Add 4 new relation types: authored, references, provides, requires. Improve prompt with disambiguation guide between similar types, stricter related_to usage, and varied confidence examples. Update graph view colors and mass for new types.
This commit is contained in:
viettranx committed
1 parent
09a7823498
commit
c49952a1e5
3 files changed
+41
-20
No files matched your search
@@ -1,6 +1,6 @@
|
||||
package knowledgegraph
|
||||
|
||||
const extractionSystemPrompt = `You are a knowledge graph extractor for an AI assistant's memory system. Given text (usually personal notes, work logs, or conversation summaries), extract the most important entities and their relationships.
|
||||
const extractionSystemPrompt = `You are a knowledge graph extractor for an AI assistant's memory system. Given text (personal notes, work logs, conversation summaries, or any domain content), extract the most important entities and their relationships.
|
||||
|
||||
Output valid JSON with this schema:
|
||||
{
|
||||
@@ -8,7 +8,7 @@ Output valid JSON with this schema:
|
||||
{
|
||||
"external_id": "unique-lowercase-id",
|
||||
"name": "Display Name",
|
||||
"entity_type": "person|project|task|event|concept|location|organization",
|
||||
"entity_type": "person|organization|project|product|technology|task|event|document|concept|location",
|
||||
"description": "Brief description of the entity",
|
||||
"confidence": 0.0-1.0
|
||||
}
|
||||
@@ -30,14 +30,24 @@ Output valid JSON with this schema:
|
||||
- Same real-world entity MUST always get the same external_id across extractions
|
||||
- When a pronoun or partial reference clearly refers to a named entity, use that entity's ID — do NOT create a new entity
|
||||
|
||||
## Entity Types
|
||||
- person: named individuals
|
||||
- organization: companies, teams, departments
|
||||
- project: software projects, initiatives, products
|
||||
- task: specific work items, tickets, TODOs
|
||||
- event: meetings, releases, incidents, deadlines
|
||||
- concept: technologies, methodologies, domains
|
||||
- location: cities, offices, regions
|
||||
## Entity Types (use ONLY these 10)
|
||||
- person: named individuals (developer, manager, doctor, teacher)
|
||||
- organization: companies, teams, departments, groups (Google, marketing team, hospital)
|
||||
- project: initiatives, campaigns, programs being built or executed (GoClaw, thesis, ad campaign)
|
||||
- product: finished goods, services, SaaS, platforms being used or sold (LadiSales, iPhone, insurance plan)
|
||||
- technology: software, tools, frameworks, languages, databases, hardware (PostgreSQL, Docker, React, MRI machine)
|
||||
- task: specific work items, tickets, TODOs (fix bug #123, deploy v2, quarterly review)
|
||||
- event: meetings, releases, incidents, deadlines, milestones (launch day, sprint review, server outage)
|
||||
- document: articles, reports, specs, contracts, guides, files (Q1 report, API spec, user manual)
|
||||
- concept: abstract ideas, methodologies, domains, standards, patterns (RBAC, Agile, machine learning, GDPR)
|
||||
- location: cities, offices, regions, venues (HCM, AWS us-east-1, room 3, building A)
|
||||
|
||||
Choosing between similar types:
|
||||
- project vs product: project = being built/executed, product = being used/sold/consumed
|
||||
- technology vs concept: technology = concrete tool/software you install/run, concept = abstract idea/methodology
|
||||
- technology vs product: technology = technical tool (PostgreSQL, Docker), product = commercial offering (Salesforce, iPhone)
|
||||
- document vs concept: document = a specific artifact (Q1 report), concept = an abstract idea (quarterly reporting)
|
||||
- task vs event: task = actionable work item with an owner, event = a point in time that happened/will happen
|
||||
|
||||
## Relation Types (use ONLY these)
|
||||
- works_on, manages, reports_to, collaborates_with (people↔work)
|
||||
@@ -45,31 +55,39 @@ Output valid JSON with this schema:
|
||||
- created, completed, assigned_to, scheduled_for (actions)
|
||||
- located_in, based_at (location)
|
||||
- uses, implements, integrates_with (technology)
|
||||
- related_to (fallback — use sparingly)
|
||||
- authored, references (documents: who wrote it, what it refers to)
|
||||
- provides, requires (capabilities: what an entity offers or needs)
|
||||
- related_to (LAST RESORT — if no specific relation fits, prefer omitting the relation entirely)
|
||||
|
||||
## Rules
|
||||
- Extract 3-15 entities depending on text density. Short text = fewer entities
|
||||
- Confidence: 1.0 = explicitly stated, 0.7 = strongly implied, 0.4 = weakly inferred
|
||||
- Confidence: 1.0 = explicitly stated fact, 0.8 = strongly implied, 0.5 = inferred from context
|
||||
- Use varied confidence — NOT everything is 1.0. Reserve 1.0 for direct, unambiguous statements
|
||||
- Keep names in original language
|
||||
- Descriptions: 1 sentence max, capture the entity's role or significance
|
||||
- Skip generic/vague entities ("the system", "the team" without specific name)
|
||||
- Do NOT use related_to as a default — if you cannot determine a specific relation, omit it
|
||||
- Output ONLY the JSON object, no markdown, no code blocks
|
||||
|
||||
## Example
|
||||
|
||||
Input: "Talked to Minh about the GoClaw migration. He'll handle the database schema changes by Friday. The team uses PostgreSQL with pgvector."
|
||||
Input: "Talked to Minh about the GoClaw migration. He'll handle the database schema changes by Friday. The team uses PostgreSQL with pgvector. I wrote the migration guide yesterday."
|
||||
|
||||
Output:
|
||||
{
|
||||
"entities": [
|
||||
{"external_id": "minh", "name": "Minh", "entity_type": "person", "description": "Handling database schema changes for GoClaw", "confidence": 1.0},
|
||||
{"external_id": "goclaw", "name": "GoClaw", "entity_type": "project", "description": "Project undergoing migration", "confidence": 1.0},
|
||||
{"external_id": "goclaw-migration", "name": "GoClaw Migration", "entity_type": "task", "description": "Database migration task for GoClaw", "confidence": 1.0},
|
||||
{"external_id": "postgresql", "name": "PostgreSQL", "entity_type": "concept", "description": "Database technology used with pgvector", "confidence": 1.0}
|
||||
{"external_id": "goclaw-migration", "name": "GoClaw Migration", "entity_type": "task", "description": "Database migration task, deadline Friday", "confidence": 1.0},
|
||||
{"external_id": "postgresql", "name": "PostgreSQL", "entity_type": "technology", "description": "Database used with pgvector extension", "confidence": 1.0},
|
||||
{"external_id": "pgvector", "name": "pgvector", "entity_type": "technology", "description": "PostgreSQL extension for vector embeddings", "confidence": 0.8},
|
||||
{"external_id": "migration-guide", "name": "Migration Guide", "entity_type": "document", "description": "Guide for the GoClaw database migration", "confidence": 1.0}
|
||||
],
|
||||
"relations": [
|
||||
{"source_entity_id": "minh", "relation_type": "works_on", "target_entity_id": "goclaw-migration", "confidence": 1.0},
|
||||
{"source_entity_id": "minh", "relation_type": "assigned_to", "target_entity_id": "goclaw-migration", "confidence": 1.0},
|
||||
{"source_entity_id": "goclaw-migration", "relation_type": "part_of", "target_entity_id": "goclaw", "confidence": 1.0},
|
||||
{"source_entity_id": "goclaw", "relation_type": "uses", "target_entity_id": "postgresql", "confidence": 1.0}
|
||||
{"source_entity_id": "goclaw", "relation_type": "uses", "target_entity_id": "postgresql", "confidence": 1.0},
|
||||
{"source_entity_id": "postgresql", "relation_type": "integrates_with", "target_entity_id": "pgvector", "confidence": 0.8},
|
||||
{"source_entity_id": "migration-guide", "relation_type": "references", "target_entity_id": "goclaw-migration", "confidence": 1.0}
|
||||
]
|
||||
}`
|
||||
@@ -42,7 +42,7 @@ func (t *KnowledgeGraphSearchTool) Parameters() map[string]any {
|
||||
},
|
||||
"entity_type": map[string]any{
|
||||
"type": "string",
|
||||
"description": "Filter by entity type (person, project, task, event, concept, location, organization)",
|
||||
"description": "Filter by entity type (person, organization, project, product, technology, task, event, document, concept, location)",
|
||||
},
|
||||
"entity_id": map[string]any{
|
||||
"type": "string",
|
||||
|
||||
@@ -29,17 +29,20 @@ interface TypeColor {
|
||||
}
|
||||
const TYPE_COLORS: Record<string, TypeColor> = {
|
||||
person: { border: "#E85D24", dark: { bg: "rgba(232,93,36,0.15)", text: "#f4a574" }, light: { bg: "#fde8d8", text: "#7a2610" } },
|
||||
organization: { border: "#ef4444", dark: { bg: "rgba(239,68,68,0.15)", text: "#fca5a5" }, light: { bg: "#fee2e2", text: "#991b1b" } },
|
||||
project: { border: "#22c55e", dark: { bg: "rgba(34,197,94,0.15)", text: "#86efac" }, light: { bg: "#dcfce7", text: "#166534" } },
|
||||
product: { border: "#f97316", dark: { bg: "rgba(249,115,22,0.15)", text: "#fdba74" }, light: { bg: "#ffedd5", text: "#9a3412" } },
|
||||
technology: { border: "#3b82f6", dark: { bg: "rgba(59,130,246,0.15)", text: "#93c5fd" }, light: { bg: "#dbeafe", text: "#1e40af" } },
|
||||
task: { border: "#f59e0b", dark: { bg: "rgba(245,158,11,0.15)", text: "#fcd34d" }, light: { bg: "#fef3c7", text: "#92400e" } },
|
||||
event: { border: "#ec4899", dark: { bg: "rgba(236,72,153,0.15)", text: "#f9a8d4" }, light: { bg: "#fce7f3", text: "#9d174d" } },
|
||||
document: { border: "#8b5cf6", dark: { bg: "rgba(139,92,246,0.15)", text: "#c4b5fd" }, light: { bg: "#ede9fe", text: "#6d28d9" } },
|
||||
concept: { border: "#a78bfa", dark: { bg: "rgba(167,139,250,0.12)", text: "#c4b5fd" }, light: { bg: "#ede9fe", text: "#5b21b6" } },
|
||||
location: { border: "#14b8a6", dark: { bg: "rgba(20,184,166,0.15)", text: "#5eead4" }, light: { bg: "#ccfbf1", text: "#115e59" } },
|
||||
organization: { border: "#ef4444", dark: { bg: "rgba(239,68,68,0.15)", text: "#fca5a5" }, light: { bg: "#fee2e2", text: "#991b1b" } },
|
||||
};
|
||||
const DEFAULT_TC: TypeColor = { border: "#9ca3af", dark: { bg: "rgba(156,163,175,0.12)", text: "#d1d5db" }, light: { bg: "#f3f4f6", text: "#374151" } };
|
||||
|
||||
const TYPE_MASS: Record<string, number> = {
|
||||
organization: 8, project: 6, person: 4, concept: 3, location: 3, event: 2, task: 1.5,
|
||||
organization: 8, project: 6, product: 5, person: 4, technology: 3.5, concept: 3, location: 3, document: 2.5, event: 2, task: 1.5,
|
||||
};
|
||||
const DEFAULT_MASS = 2;
|
||||
|
||||
|
||||
Reference in new issue
Block a user