# How to Extract Entities and Semantic Relationships from Page Content

Entity Analysis in QueryBurst extracts every named entity (people, products, organisations, concepts, locations, events) and semantic relationship from a page, then scores the results for density, coherence, and alignment. Results are visualised as an arc diagram, a filterable entity list, relationship triples, topic clusters, and a document view with attention mode that highlights what AI focuses on. Pages with low entity density or orphan entities (mentioned but unconnected) are flagged.

## How to Access Entity Analysis

1. Navigate to the **[Page Reports](/content/docs/page-reports/index.html)** tab in QueryBurst
2. Select a page from the crawled pages list
3. Click on the **Entities** tab in the page details view
4. Click **Start Entity Analysis** to run the analysis

> **Note**: Entity Analysis is computationally intensive and may take 30-60 seconds depending on content length.

## Understanding the Results

### Summary Bar

After analysis completes, you'll see a compact summary showing:

| Metric | Description | Good Score |
| --- | --- | --- |
| **Overall Score** | Combined score of all metrics | 70+ |
| **Density** | How many entities per unit of text | 60+ |
| **Relations** | How many relationships between entities | 50+ |
| **Coherence** | How well entities cluster into topics | 65+ |
| **Alignment** | How well content matches expected topics | 60+ |
| **Semantic** | Entity signal vs filler word ratio | 40+ |

You'll also see counts of:

- **Entities** - Total unique entities extracted
- **Triples** - Total relationships discovered
- **Clusters** - Topic groups formed

### Primary Topics

Tags showing the main topics your content covers, derived from entity clustering.

### Inferred Topic

An AI-generated summary of what your page appears to be about, based on the extracted [entities](/content/docs/site-intelligence/entity-profiles/index.html) and relationships.

### Orphan Entities Warning

Entities that are mentioned but have no relationships to other entities. These represent potential content gaps where concepts are introduced but not connected to your main narrative.

## View Modes

### Graph View

An interactive **arc diagram** visualization showing:

- **Nodes** - Entities positioned in a circle, sized by importance (salience/mentions)
- **Arcs** - Curved lines connecting related entities
- **Colors** - Each entity type has a distinct color

**Interactions:**

- **Hover** over an entity to highlight its connections
- **Click** an entity to lock the selection and see all its relationships in the detail panel
- **Click again** to deselect

The right panel shows all relationships for the selected entity in the format:

> Subject → _predicate_ → Object

### Entities View

A filterable list of all extracted entities:

**Entity Types:**

| Type | Color | Examples |
| --- | --- | --- |
| **Person** | Blue | Names, roles, titles |
| **Organization** | Purple | Companies, institutions, brands |
| **Product** | Emerald | Products, services, offerings |
| **Concept** | Amber | Abstract ideas, methodologies |
| **Location** | Rose | Places, regions, addresses |
| **Event** | Cyan | Dates, occasions, milestones |
| **Attribute** | Slate | Properties, characteristics |

**Features:**

- Filter by entity type using the pills at the top
- Click any entity row to expand and see **context** - where it appears in your content
- See mention count for each entity

### Relationships View

Shows all **semantic triples** extracted from your content in the format:

> **Subject** → _predicate_ → **Object**

Examples:

- "Organic mattress" → _is made from_ → "natural latex"
- "Company" → _founded in_ → "2015"
- "Product" → _certified by_ → "GOTS"

Relationships reveal how your content connects concepts and are crucial for AI understanding.

### Clusters View

Groups of semantically related entities that form **topic clusters**:

Each cluster shows:

- **Primary topic** - The central concept
- **Entity count** - How many entities belong to this cluster
- **Member entities** - All entities in the cluster

Well-structured content typically has:

- 2-5 clear topic clusters
- Strong central themes with supporting entities
- Minimal overlap between clusters

### Document View

The most powerful view for understanding how entities appear in context. Four sub-modes:

#### Normal Mode

Full document text with entities highlighted in their type colors. Hover over any entity to see its type and mention count.

#### Attention Mode

Dims all non-entity text, making entities "pop" visually. This shows what an AI might focus on when extracting meaning from your content.

#### Condensed Mode

Removes all non-entity text entirely, showing only the entities in sequence. This represents the "essence" of your content after removing filler words.

#### Themes Mode

Shows the extracted core themes with their descriptions and member entities. This is the final "distilled" representation of your content.

**Entity Type Filters:** Use the colored pills to toggle which entity types are highlighted. This helps focus on specific aspects of your content.

**Chunk Density Scores:** Each content chunk shows a density percentage indicating how much of that section's text consists of meaningful entities vs filler.

## Watch Distillation

Click the **Watch Distillation** button to see an animated visualization of how your content is progressively condensed:

1. **Web Page** → The original page
2. **Extracted Text** → Raw text content (word count)
3. **Attention** → Key terms highlighted (filler fading)
4. **Condensed** → Only entities remain
5. **Themes** → Final core themes

This visualization helps explain how AI systems might process and understand your content.

## Interpreting Results

### Healthy Entity Structure

✅ **High entity density** - Content rich in meaningful concepts

✅ **Many relationships** - Concepts are connected, not just listed

✅ **Clear topic clusters** - Content is well-organized thematically

✅ **Few orphan entities** - All concepts tie back to main narrative

✅ **Strong coherence** - Topics stay focused, don't drift

### Warning Signs

⚠️ **Low entity density** - Too much filler, not enough substance

⚠️ **Few relationships** - Concepts mentioned but not connected

⚠️ **Many orphan entities** - Disconnected ideas that confuse readers

⚠️ **Too many clusters** - Content lacks focus, tries to cover too much

⚠️ **Low coherence** - Topics jump around without clear structure

### Score Interpretation

| Score Range | Interpretation |
| --- | --- |
| **80-100** | Excellent - Well-structured, entity-rich content |
| **60-79** | Good - Solid foundation with room for improvement |
| **40-59** | Fair - Consider adding connections and depth |
| **0-39** | Needs work - Content may be too thin or unfocused |

## Tips & Best Practices

### Improving Entity Density

1. **Be specific** - Use proper nouns and specific terms instead of generic ones
2. **Name things** - Give products, features, and concepts clear names
3. **Add context** - Explain what entities are when introducing them

### Building Stronger Relationships

1. **Use connecting language** - "X is designed for Y", "A includes B"
2. **Explain causation** - "Because of X, we developed Y"
3. **Show hierarchy** - "X, which is part of Y, enables Z"

### Reducing Orphan Entities

1. **Connect back** - When introducing a new concept, relate it to something already mentioned
2. **Use transitions** - Bridge between topics explicitly
3. **Summarize** - Recap how concepts relate at section ends

### Improving Topic Coherence

1. **Outline first** - Plan content structure before writing
2. **One topic per section** - Don't mix unrelated concepts
3. **Use headers** - Clear section breaks help maintain focus

## Technical Details

### How Entity Extraction Works

1. **Chunking** - Content is split into semantic chunks
2. **NER Processing** - Named Entity Recognition identifies entities
3. **Type Classification** - Entities are categorized by type
4. **Relationship Extraction** - Semantic triples are identified
5. **Embedding** - Entities are converted to vectors
6. **Clustering** - Similar entities are grouped by semantic similarity
7. **Topic Inference** - An overall topic is synthesized

### Entity Salience

Entities are ranked by "salience" - how important they are to the content:

- Frequency of mentions
- Position in content (earlier = more important)
- Relationship connections
- Semantic centrality in clusters

### Analysis Caching

Previous analyses are saved and can be loaded from the history panel. This lets you:

- Compare analyses over time as content changes
- Quickly review past results without re-running
- Track improvements in entity structure