saross/skills/entity-classifier

Entity Classifier Skill

Purpose

Classifies mentions of dual-nature entities (hotels, churches, schools of arts, halls, lodges, etc.) in historical newspaper text to determine whether they should be tagged as:

• Building/facility only (Built Environment facet)
• Business/organisation only (Agents facet)
• Both (polyhierarchical) (appears in both facets)

Uses natural language understanding to analyse context and apply a linguistic heuristic framework, providing more nuanced classification than deterministic regex patterns.

When to Use

• Classifying hotel mentions as buildings vs businesses
• Determining church classification (building vs religious organisation)
• Analysing schools of arts (venue vs cultural society)
• Any entity that can be both physical structure and organisational agent
• Reviewing and validating automated classifications
• Batch processing entity mentions from Zotero library

Role

You are an expert in classifying historical entities mentioned in 19th-century Australian newspaper text for a controlled vocabulary taxonomy aligned with Getty Art & Architecture Thesaurus (AAT) principles.

Task

Analyse mentions of dual-nature entities (hotels, churches, schools of arts, halls, lodges, etc.) in historical newspaper text and determine their appropriate facet classification:

• (a) Building/Facility Only — Entity is referenced solely as physical structure (Built Environment facet)
• (b) Business/Organisation Only — Entity is referenced solely as an agent/operator (Agents facet)
• (c) Both (Polyhierarchical) — Entity exhibits both spatial and agency characteristics in the context

Classification Heuristic

Building/Facility Indicators

Look for evidence that the entity is being used as a location or physical space:

Strong indicators:

• Locational phrases: "at [entity]", "in [entity]", "within [entity]", "near [entity]", "opposite [entity]"
• Movement to/from: "going to [entity]", "arriving at [entity]", "leaving [entity]", "travelled to [entity]"
• Events occurring: "meeting held at [entity]", "concert at [entity]", "ball at [entity]", "auction at [entity]"
• Accommodation: "staying at [entity]", "lodging at [entity]", "resided at [entity]"
• Physical features: "[entity]'s rooms", "[entity]'s bar", "[entity]'s verandah", "[entity] building"
• Construction/damage: "[entity] was built", "fire destroyed [entity]", "[entity] was demolished"
• Venue usage: "in [entity]'s yard", "at [entity]'s grounds"
• Elliptical spatial references: "at the room" (when discussing [entity]), "in the hall" (entity's space)
  • CRITICAL: When entity discussion includes "at/in the [room/hall/library]" without specifying whose, assume it refers to the entity's building
  • Example: "Committee of the School of Arts was to be held at the room" → "the room" = School of Arts' room = building indicator

Business/Organisation Indicators

Look for evidence that the entity is acting as an agent, operator, or organisation:

Strong indicators:

• Agency verbs: "[entity] is expanding", "[entity] announced", "[entity] opened", "[entity] refurbished"
• Ownership/management: "[entity] proprietor", "owner of [entity]", "[entity] manager", "keeper of [entity]"
• Business operations: "[entity] licensed", "[entity] trading", "[entity] conducting business"
• Services provided: "[entity] offers", "[entity] provides", "[entity] caters"
• Financial actions: "[entity] purchased", "[entity] selling", "[entity] revenue"
• Employment: "[entity] employs", "[entity] hiring staff"
• Legal agency: "[entity] was fined", "[entity] applied for licence", "[entity] prosecuted"
• Competition: "[entity] competing with", "[entity] attracting customers"

Metonymy (Important!)

When the entity name is used as shorthand for people/organisation, classify by intended referent:

• "The hotel denies the accusation" → Business (hotel = proprietor/management)
• "The church condemns the proposal" → Organisation (church = congregation/leadership)
• "Meeting at the church" → Building (church = physical venue)

Passive vs Active Voice

• Passive: "[entity] was refurbished" → Building (recipient of action)
• Active: "[entity] refurbished its premises" → Business (agent of action)

Both (Polyhierarchical) Classification

Use "both" when:

• Context contains both spatial AND agency indicators
• Text treats entity as both place and actor in same passage
• Example: "Concert at the hotel [building], with the hotel [business] providing refreshments"

Default Guidance

When indicators are weak or absent:

• Hotels/inns: Default to building (most newspaper mentions are locational)
• Churches: Consider denomination and context (worship events = building, governance = organisation)
• Schools of Arts: Consider activity (event venue = building, committee decisions = organisation)
• Halls: Usually buildings unless committee/management explicitly mentioned

Output Format

CRITICAL: Always generate classification results as markdown files, never as chat responses only.

File location: entity-tagging-system/outputs/{entity-type}/{entity-type}_classification_results.md

Template: entity-tagging-system/templates/classification-report-template.md

Include: detailed analysis for each mention, pattern summary, taxonomy recommendations, comparison tables.

Why markdown files:

• Creates permanent audit trail
• Enables cross-session review
• Facilitates methodology sharing
• Supports reproducible research
• Chat responses are ephemeral; files persist

Per-mention response template

For each entity mention analysed, provide a structured response:

### Entity: [Entity Name]
**Item:** [Article Title]
**Classification:** building | business | both
**Confidence:** high | medium | low

**Reasoning:**
[2-3 sentences explaining your classification, referencing specific indicators found in the context]

**Indicators Found:**
- Building: [list matched indicators, or "none"]
- Business: [list matched indicators, or "none"]

**Context:**
> [The relevant excerpt showing the entity mention with surrounding text]

Confidence Levels

• High: Multiple strong indicators present; classification unambiguous
• Medium: Some indicators present but context could support alternative reading
• Low: Minimal indicators; relying on defaults or general patterns

Special Cases Requiring "Both"

1. Mixed signals in same passage: Both spatial and agency cues present
2. Metonymic shift: Entity name refers to both structure and organisation within text
3. Parallel constructions: Text explicitly treats entity as both place and actor

Quality Standards

• Evidence-based: Every classification must be grounded in textual evidence
• Audit trail: Reasoning must be clear enough for human reviewer to verify
• Conservative: When truly ambiguous, prefer established defaults over speculation
• UK/Australian spelling: behaviour, organisation, licence (noun), etc.

Taxonomy Naming Principles

Intrinsic vs Parenthetical Disambiguation:

The project aims for unique leaf nodes in the taxonomy. When possible, distinguish between building and organisation aspects intrinsically through naming, falling back to parenthetical qualifiers only when necessary.

Use intrinsic naming distinction (no parenthetical):

• "Halls" (buildings) vs "Lodges" (organisations) — different words naturally distinguish the concepts
• "Boarding houses" (buildings) vs "Hospitality businesses" (organisations) — hierarchy context makes it clear
• Buildings use architectural terms; organisations use corporate/group terms

Use parenthetical qualifiers when necessary:

• Churches: "Methodist Church (building)" vs "Methodist Church (organisation)" — same name requires disambiguation
• Hotels: "Imperial Hotel (building)" vs "Imperial Hotel (business)" — same name requires disambiguation
• Generic terms: "church (building)" vs "church (organisation)"

Why this matters for classification:

When you recommend taxonomy placement, consider:

1. Can the building/organisation aspects be distinguished by different entity names? (e.g., "Masonic Hall" for building, "Freemasons" for organisation)
2. Or do they share the same name and require parenthetical qualifiers? (e.g., "Imperial Hotel (building)" and "Imperial Hotel (business)")

This informs whether the entity should use:

• Separate leaf nodes with intrinsic distinction (halls vs lodges)
• Disambiguated leaf nodes with parenthetical qualifiers (hotel (building) vs hotel (business))

Example Analyses

Example 1: Building Only

Context: "A concert was held at the Carrington Hotel last evening, attended by many prominent citizens."

Classification: building Reasoning: Strong locational indicator ("at the Carrington Hotel") with event occurrence. No agency or business operation signals. Hotel functions purely as venue. Indicators: Building: locational_prep, events_at | Business: none

---

Example 2: Business Only

Context: "The Imperial Hotel has announced extensive renovations and will close for the season to complete the work."

Classification: business Reasoning: Hotel acts as agent (announcing, closing, completing work). Active voice with agency verbs. No spatial or locational cues. Indicators: Building: none | Business: agency_verbs, business_ops

---

Example 3: Both (Polyhierarchical)

Context: "The Megalong Hotel proprietor applied for a licence renewal. Meanwhile, a dance was held at the hotel on Saturday evening."

Classification: both Reasoning: First sentence treats hotel as business (proprietor agency, licensing). Second sentence treats hotel as venue (event location). Both aspects present in same context. Indicators: Building: locational_prep, events_at | Business: proprietor_subject, legal_agent

---

Example 4: Metonymy — Business

Context: "The Katoomba Hotel denies responsibility for the incident and contests the charges."

Classification: business Reasoning: Metonymic usage — "hotel" stands for hotel management/proprietor (agents who can deny and contest). Not referring to physical building. Indicators: Building: none | Business: metonymy (agency verbs with abstract subject)

---

Working with Batch Data

When processing multiple entities:

1. Process each mention independently — don't let previous classifications bias current one
2. Note patterns — if same entity consistently appears as one type, mention this
3. Flag outliers — if entity classification differs from pattern, explain why
4. Preserve context — always include enough context for human to verify

Integration with Project Workflow

This skill supports the Blue Mountains folksonomy rationalisation project:

• Input: Entity mentions from Zotero library with full newspaper text
• Process: Analyse context using this classification framework
• Output: Structured recommendations for taxonomy facet assignment
• Review: Human expert reviews and approves/modifies classifications
• Application: Approved classifications update tag_map_consolidated.csv

Notes for Reuse

This skill is designed to be reusable across any dual-nature entity type:

• Hotels, inns, public houses (accommodation vs hospitality business)
• Churches, chapels (building vs religious organisation)
• Schools of Arts, Mechanics' Institutes (hall vs cultural society)
• Fraternal halls (lodge building vs fraternal order)
• Schools (educational facility vs educational institution)

Adjust the default guidance and entity-specific patterns as needed for each type.

Related Documents

• /home/shawn/Code/blue-mountains/docs/entity-classification-heuristic.md — full decision framework
• /home/shawn/Code/blue-mountains/CLAUDE.md — project taxonomy principles
• Getty AAT guidelines for polyhierarchical classification