Concept Knowledge Base

The Concept Knowledge Base (Concept KB) is an optional knowledge layer in the Synthesis system that provides structured management of core concepts referenced in the literature. Concepts can be overlaid onto the topic graph and reader, enriching the context for topic synthesis.

What Is a Concept?

In the Synthesis system, a concept is a term or entity with independent meaning within a research domain. Unlike the flat classification of tags, concepts can have a multi-layered structure, including senses, aliases, and relationships.

Four-Layer Structure of Concepts

Concept                 — e.g., "Transformer"
  └── Sense             — e.g., "Transformer (machine learning architecture)"
       ├── Alias        — e.g., "Transformer model", "Transformer network"
       └── Relation     — broader_than "Attention Mechanism"

Concept Types

Type	Description	Examples
method	Research methods	Deep learning, reinforcement learning
model	Models or architectures	Transformer, ResNet
dataset	Datasets	ImageNet, COCO
metric	Evaluation metrics	BLEU, F1-score
field	Research fields	Computer vision, natural language processing
task	Tasks	Image classification, machine translation
tool	Tools	PyTorch, TensorFlow

Concepts Surface Features

Concept List

On the Synthesis Workbench → Concepts page, you can browse all indexed concepts:

• Filter: By type (method / model / dataset, etc.), status, or associated topics
• Search: Search concepts by name
• View Toggle: Compact / comfortable density

Concept Details

After selecting a concept, you can view and edit:

Information	Description
Identity	Concept ID, name, type
Status	active / deprecated / pending
Definition	Descriptive definition of the concept
Senses	Specific meanings of the concept in different contexts
Aliases	Alternative names for the same concept
Relations	Associations with other concepts (broader / narrower / related)
Related Topics	Topics that reference this concept

Sense Management

The same concept may have different meanings across disciplines. The sense mechanism allows:

• Adding multiple senses to a concept, each with its own definition
• Annotating the usage context or domain for each sense
• Associating specific senses with papers or topics

Alias Management

• Recording different naming conventions for the same concept (e.g., full name, abbreviation, alternative terms)
• Aliases are used for citation matching and concept identification

Overlay Features

Concept information can be overlaid onto other surfaces:

• Overlay onto Topic Graph: Display concepts related to a topic in the Topic Graph
• Overlay onto Reader: Display concept cards on the Topic Detail page

Review

Change proposals to the concept knowledge base (new concepts, new senses, new relationships) appear in the Concepts review tab of the Review Hub. You can review and decide whether to accept these proposals.

Relationship with Tags

Concepts and tags are two complementary approaches to knowledge organization:

Dimension	Tags	Concepts
Structure	Flat, facet:value	Multi-layered (senses + aliases + relationships)
Purpose	Literature classification and filtering	Knowledge management and association analysis
Source	Controlled vocabulary + AI inference	Auto-extracted from literature + user-managed
Scope	Covers all literature	Deep coverage of selected core terms

Next Steps

• Review Hub — Review concept suggestions
• Tags Management — Manage the controlled tag vocabulary
• Topic Synthesis — Leverage concept knowledge when creating topic syntheses