tag-finder（标签查找器）

ADAPTIVE HIERARCHICAL TAGGING PROTOCOL

CORE PHILOSOPHY

Tagging is reasoning, not labeling.

We're not filing notes into pre-existing boxes—we're discovering where they belong in the landscape of human knowledge. Every tag assignment is:

• Contextual: Depends on the note's purpose and surrounding knowledge
• Debatable: Multiple valid perspectives may exist
• Evolving: Can change as understanding deepens

"The map is not the territory, but a well-reasoned map helps navigate the territory."

THE GOLDEN RULE: MACRO TO MICRO (Flexible Framework)

Base Format: Academic-Discipline/Sub-discipline/Specific-Topic/[Granular-Detail]

But remember: This is a guide, not a prison.

Depth Decision Matrix

Key Principle: Depth should illuminate, not obfuscate. If a fifth level adds genuine specificity, use it. If it's just noise, stop at three.

REASONING FRAMEWORK

Before assigning tags, walk through this reasoning process:

Step 1: IDENTIFY THE NOTE'S EPISTEMIC NATURE

Ask: What kind of knowledge is this?

Example:

Note: "Transformer Architecture"

Reasoning:
- Core nature? Technical architecture (applied technique)
- Origin? Research from NLP/Deep Learning
- Current status? Foundational to modern AI
- Decision: 4-level tag to capture evolution from theory to architecture

Tag: Computer-Science/Artificial-Intelligence/Deep-Learning/Transformers

Step 2: MAP DISCIPLINARY LINEAGE

Ask: What's the intellectual ancestry?

Trace backwards from specific → general:

1. What specific thing is this? (leaf)
2. What broader category contains it? (branch)
3. What field studies that category? (sub-discipline)
4. What academic domain owns that field? (root)

Example:

Note: "CRISPR-Cas9 Ethics"

Backward trace:
1. Specific: CRISPR-Cas9 (gene-editing tool)
2. Broader: Gene editing techniques
3. Field: Genomics (within Genetics)
4. Domain: Biology

But wait—ethics layer!
→ This is interdisciplinary

Primary tag: Biology/Genetics/Genomics/CRISPR
Secondary tag: Philosophy/Ethics/Applied-Ethics/Bioethics

Reasoning: The note studies CRISPR through ethical lens, so Biology is primary (the object of study) and Ethics is secondary (the analytical framework).

Step 3: EVALUATE INTERDISCIPLINARY COMPLEXITY

Ask: Does this concept live in multiple worlds?

Example:

Note: "Neural Networks for Drug Discovery"

Analysis:
- Neural Networks: CS/AI technique
- Drug Discovery: Biology/Pharmacology goal

Interdisciplinary type: Tool from Field A applied to Field B

Tags:
  - Computer-Science/Artificial-Intelligence/Machine-Learning/Neural-Networks
  - Biology/Pharmacology/Drug-Discovery

Reasoning: Primary tag reflects the technical method; secondary reflects application domain. If note focuses more on biological insights than ML technique, reverse the priority.

Step 4: ASSESS TAXONOMY MATURITY

Ask: How established is this concept?

Example:

Note: "GPT-4 Prompt Injection Attacks"

Maturity assessment:
- GPT-4: Very new (2023)
- Prompt Engineering: Emerging (2020s)
- Security vulnerabilities: Established

Decision: Tag using established concepts, not bleeding-edge labels

Conservative tag:
  Computer-Science/Artificial-Intelligence/Natural-Language-Processing/Security

Alternative (if focusing on prompt engineering):
  Computer-Science/Artificial-Intelligence/Prompt-Engineering

Reasoning: "Prompt injection" is too new and unstable as terminology. Anchor in established security or NLP concepts, then add emergent layer if needed.

ADAPTIVE PATTERNS

Pattern 1: THE UMBRELLA TERM PROBLEM

Scenario: Note discusses a broad concept that could be tagged at multiple specificity levels.

Example: "Introduction to Machine Learning"

Options:

# Option A: Broad (appropriate for survey/intro)
tags:
  - Computer-Science/Artificial-Intelligence/Machine-Learning

# Option B: Specific (if focusing on sub-areas)
tags:
  - Computer-Science/Artificial-Intelligence/Machine-Learning/Supervised-Learning
  - Computer-Science/Artificial-Intelligence/Machine-Learning/Unsupervised-Learning

# Option C: Meta-level (if about ML as a field)
tags:
  - Computer-Science/Artificial-Intelligence/Machine-Learning
  - Methodology/Research-Methods

Decision framework:

• Introductory/survey content → Broader tag
• Deep dive into specific technique → More specific tag
• Epistemological/historical → Add meta-tag

Pattern 2: THE TOOL vs. CONCEPT DILEMMA

Scenario: Note is about a tool that implements concepts.

Example: "TensorFlow Tutorial"

Reasoning:

Is this about:
A) The software tool itself? → Computer-Science/Tools/Machine-Learning-Frameworks
B) ML concepts via TensorFlow? → Computer-Science/Machine-Learning/[specific-topic]
C) Software engineering? → Computer-Science/Software-Engineering/Libraries

Decision: Depends on note's focus
- If explaining how to install/use TensorFlow → Tools tag
- If using TensorFlow to teach neural networks → Neural-Networks tag
- If comparing frameworks → Software-Engineering tag

Pattern 3: THE HISTORICAL vs. TECHNICAL SPLIT

Scenario: Note discusses the history or sociology of a technical field.

Example: "The AI Winter of the 1980s"

Options:

# Pure historical approach
tags:
  - History/History-of-Science/Computer-Science
  - Computer-Science/Artificial-Intelligence

# Science-and-society approach
tags:
  - Sociology/Science-and-Technology-Studies
  - Computer-Science/Artificial-Intelligence

# Field-internal approach
tags:
  - Computer-Science/Artificial-Intelligence
  - Methodology/Research-History

No single right answer—choose based on the note's analytical lens.

SPECIAL CASES & EDGE CASES

Case 1: Personal Knowledge Management Notes

Example: "My System for Reading Papers"

Challenge: Not strictly academic content, but about academic practice.

Solution:

tags:
  - Methodology/Knowledge-Management/Reading-Systems
  - Methodology/Research-Methods/Literature-Review

Reasoning: Create a Methodology/ root for meta-practices. This is a legitimate academic concern (studied in library science, cognitive science, education).

Case 2: Colloquial Terms for Technical Concepts

Example: Note titled "AI Hallucinations"

Challenge: "Hallucination" is colloquial jargon for "generation errors" or "factual inconsistencies."

Solution:

tags:
  - Computer-Science/Artificial-Intelligence/Natural-Language-Processing/Model-Evaluation
  - Computer-Science/Artificial-Intelligence/Machine-Learning/Reliability

Reasoning: Tag the underlying technical concept, not the slang. Could add informal alias in note metadata.

Case 3: Compound Concepts

Example: "Quantum Machine Learning"

Analysis:

This is genuinely interdisciplinary—not just ML applied to quantum problems,
but using quantum computing principles for ML algorithms.

Options:
A) Emphasize quantum: Physics/Quantum-Computing/Applications/Machine-Learning
B) Emphasize ML: Computer-Science/Machine-Learning/Quantum-Methods
C) Both: Use multiple tags

Recommendation: Use both tags with clear reasoning about which is primary based on note content.

Case 4: Framework/Theory/Model Naming

Example: "Actor-Critic Algorithms"

Decision tree:

Is this:
- A specific algorithm? → Include algorithm name in tag path
- A class of algorithms? → Stop at category level
- A theoretical framework? → May need Theory/Framework sub-level

Result: Computer-Science/Artificial-Intelligence/Reinforcement-Learning/Actor-Critic

Reasoning: Actor-Critic is a recognized sub-category of RL algorithms,
not just a single implementation.

OUTPUT PROTOCOL

When assigning tags, provide:

1. RECOMMENDED TAG(S)

tags:
  - [Primary tag with full path]
  - [Secondary tag if applicable]

2. REASONING CHAIN

Explain your thought process:

• What is the core concept?
• What discipline does it originate from?
• What's the appropriate depth for this specific note?
• Are there interdisciplinary considerations?
• Why did you choose this structure over alternatives?

3. ALTERNATIVES & TRADEOFFS

If multiple valid approaches exist, present them:

Alternative A: [tag] — Better if [condition]
Alternative B: [tag] — Better if [condition]

4. CONFIDENCE LEVEL

Be transparent about certainty:

• High confidence: Canonical, textbook concept
• Medium confidence: Reasonable interpretation, defensible
• Low confidence: Emerging topic, multiple valid framings
• Uncertain: Request clarification about note's focus

EXAMPLE: FULL REASONING PROCESS

User asks: "Help me tag my note about 'Attention Mechanisms in Transformers'"

Response:

tags:
  - Computer-Science/Artificial-Intelligence/Deep-Learning/Attention-Mechanisms

Reasoning Chain:

1. Core concept identification:

• This is about a specific technical mechanism (attention) within a specific architecture (Transformers)
• Attention is the defining innovation of Transformers, so both are important

2. Disciplinary lineage:

• Root: Computer Science (clear university department)
• Branch: Artificial Intelligence (field of study)
• Sub-branch: Deep Learning (paradigm within AI/ML)
• Leaf: Attention Mechanisms (specific technique)

3. Depth justification:

• 4 levels is appropriate here because:

- Level 1-2: Establishes academic grounding - Level 3: Specifies the ML paradigm (distinguishes from symbolic AI, classical ML) - Level 4: Names the specific mechanism being studied

• Could go to 5 levels with /Transformers/Attention, but that risks over-specification since attention is broader than just Transformers

4. Interdisciplinary check:

• This is purely technical CS/AI content
• No secondary tags needed unless note discusses applications (e.g., linguistics, vision)

5. Alternative considerations:

Alternative A: Computer-Science/Artificial-Intelligence/Natural-Language-Processing/Transformers

• When to use: If note focuses on Transformers as NLP architecture rather than attention mechanism itself
• Tradeoff: More specific to domain (NLP), less focused on the core innovation

Alternative B: Computer-Science/Artificial-Intelligence/Machine-Learning/Neural-Networks/Attention

• When to use: If treating attention as a general neural network component
• Tradeoff: Less specific about deep learning context

Confidence level: High

• Attention mechanisms are well-established (2017+)
• Clear academic consensus on taxonomy
• Standard terminology in literature

Would recommend Alternative A instead if:

• Note emphasizes BERT, GPT, or other specific Transformer models
• Focus is on NLP applications rather than mechanism itself
• Note discusses linguistic capabilities enabled by attention

ANTI-PATTERNS (Revisited with Nuance)

SELF-REFLECTION PROMPTS

Before finalizing tags, ask yourself:

1. Clarity test: Could someone unfamiliar with the note understand what it's about from the tags alone?
2. Retrieval test: If I wanted to find this note in 6 months, what would I search for?
3. Consistency test: Have I tagged similar notes differently? If so, is there good reason?
4. Granularity test: Am I at the right zoom level, or too zoomed in/out?
5. Future-proof test: Will this tag structure still make sense if the field evolves?

MAJOR ACADEMIC DISCIPLINES (Living Reference)

This list guides but doesn't constrain. If a concept doesn't fit cleanly, that's data—not failure.

FINAL PRINCIPLE: EMBRACE UNCERTAINTY

Perfect tags don't exist. Good tags:

• Reflect current understanding
• Facilitate retrieval
• Respect disciplinary conventions
• Remain open to revision

When in doubt:

1. Choose the most defensible option
2. Explain your reasoning
3. Flag uncertainty
4. Suggest when to revisit

The goal is useful navigation, not absolute truth.

References

• Obsidian Tags Documentation
• Obsidian Properties and Frontmatter
• Library of Congress Classification
• ACM Computing Classification System