README.md

---
license: apache-2.0
task_categories:
- question-answering
- text-classification
language:
- en
tags:
- agent
---

# BYU-Idaho Web Content Dataset (NLP-Enhanced)

**State-of-the-art** university web content dataset with full NLP enrichment: entity extraction, acronym detection, domain terminology, and semantic features. Enterprise-ready for advanced RAG, semantic search, and AI applications.

## Dataset Description

**Records:** 2,448 ultra-high-quality pages
**Source:** byui.edu and subdomains
**Format:** Markdown + NLP metadata (JSON fields)
**Quality:** 40.2% filtered + 91.5/100 avg score + Full NLP extraction
**Last Updated:** December 2025

### NLP Enrichment Features

**Entity Extraction** (spaCy NER)
- **15,289 organizations** extracted
- **3,029 locations** extracted
- **4,841 people** extracted
- **94.2% page coverage**

**Acronym Detection**
- **4,036 acronyms** detected and expanded
- **81.5% page coverage**
- Includes common education acronyms (FAFSA, GPA, TOEFL, etc.)

**Domain Terminology**
- **566 BYU-Idaho specific terms** found
- **16.7% page coverage**
- Includes: I-Learn, Devotional, Pathway, Honor Code, campus buildings, etc.

**Key Phrases**
- **187 action phrases** extracted
- Common educational actions: "apply for admission", "register for classes", etc.

**Domain N-grams**
- Common domain-specific 3-word phrases
- Frequency-filtered for relevance

## Dataset Structure

### Core Fields

- **index** (`int64`): Sequential ID (1-2448)
- **url** (`string`): Source URL
- **title** (`string`): Page title
- **topic** (`string`): Main heading (H1)
- **meta_description** (`string`): SEO description
- **content** (`string`): Full Markdown content (avg 2,230 chars)
- **category** (`string`): 14 categories (Academics, Admissions, etc.)
- **content_type** (`string`): 7 types (informational, guide, FAQ, etc.)
- **quality_score** (`int64`): Quality 0-100 (avg: 91.5)
- **reading_level** (`float64`): Flesch-Kincaid grade level (avg: 14.4)

### NLP Fields

**entities** (`string` - JSON):
```json
{
  "organizations": ["BYU-Idaho", "Financial Aid Office", "College of Business"],
  "locations": ["Rexburg", "Idaho", "Manwaring Center"],
  "programs": ["Computer Science", "Nursing"],
  "people": ["David A. Bednar", "Gordon B. Hinckley"]
}
```

**byui_terms** (`string` - JSON):
```json
["Devotional", "I-Learn", "Pathway", "Honor Code", "Manwaring Center"]
```

**acronyms** (`string` - JSON):
```json
{
  "FAFSA": "Free Application for Federal Student Aid",
  "GPA": "Grade Point Average",
  "TOEFL": "Test of English as a Foreign Language"
}
```

**key_phrases** (`string` - JSON):
```json
[
  "apply for admission",
  "register for classes",
  "submit transcripts",
  "complete the FAFSA"
]
```

**domain_ngrams** (`string` - JSON):
```json
[
  "church of jesus christ",
  "brigham young university",
  "learn more about"
]
```

## Usage

### Basic Loading

```python
from datasets import load_dataset
import json

ds = load_dataset("BYU-Idaho/Web-Content")['train']

# Parse JSON fields
row = ds[0]
entities = json.loads(row['entities'])
acronyms = json.loads(row['acronyms'])
byui_terms = json.loads(row['byui_terms'])

print(f"Organizations: {entities['organizations']}")
print(f"Acronyms: {list(acronyms.keys())}")
```

### Entity-Based Filtering

```python
import json

# Find pages mentioning specific organizations
def has_organization(row, org_name):
    entities = json.loads(row['entities'])
    return org_name in entities['organizations']

financial_aid_pages = ds.filter(
    lambda x: has_organization(x, 'Financial Aid Office')
)
```

### Acronym Expansion for RAG

```python
# Build acronym lookup table
all_acronyms = {}
for row in ds:
    acronyms = json.loads(row['acronyms'])
    all_acronyms.update(acronyms)

# Use in RAG to expand user queries
def expand_acronyms(query):
    for acronym, expansion in all_acronyms.items():
        if acronym in query:
            query += f" {expansion}"
    return query

# "What is FAFSA?" → "What is FAFSA Free Application for Federal Student Aid?"
```

### BYU-Idaho Term Filtering

```python
# Find pages about specific campus features
def has_byui_term(row, term):
    terms = json.loads(row['byui_terms'])
    return term in terms

devotional_pages = ds.filter(lambda x: has_byui_term(x, 'Devotional'))
pathway_pages = ds.filter(lambda x: has_byui_term(x, 'Pathway'))
```

### Location-Based Search

```python
# Find pages about specific locations
def mentions_location(row, location):
    entities = json.loads(row['entities'])
    return location in entities['locations']

rexburg_pages = ds.filter(lambda x: mentions_location(x, 'Rexburg'))
```

### Advanced: Build Entity Index

```python
from collections import defaultdict
import json

# Build inverted index: entity → list of page indices
entity_index = defaultdict(list)

for idx, row in enumerate(ds):
    entities = json.loads(row['entities'])
    for org in entities['organizations']:
        entity_index[org].append(idx)
    for loc in entities['locations']:
        entity_index[loc].append(idx)

# Quick lookup: all pages mentioning "Tutoring Center"
tutoring_pages = [ds[i] for i in entity_index['Tutoring Center']]
```

## NLP Enrichment Statistics

| Feature | Total Extracted | Page Coverage |
|---------|----------------|---------------|
| **Organizations** | 15,289 | 94.2% |
| **Locations** | 3,029 | 94.2% |
| **People** | 4,841 | 94.2% |
| **Acronyms** | 4,036 | 81.5% |
| **BYU-Idaho Terms** | 566 | 16.7% |
| **Key Phrases** | 187 | - |

## Top Entities Extracted

**Organizations** (most common):
- BYU-Idaho
- Brigham Young University-Idaho
- Ricks College
- Financial Aid Office
- Accessibility Services Office
- Academic Leadership Office
- Church of Jesus Christ of Latter-day Saints

**Locations** (most common):
- Rexburg
- Idaho
- Manwaring Center
- BYU-Idaho Center
- Taylor Chapel
- United States

**BYU-Idaho Terms**:
- Devotional
- Forum
- Pathway
- Honor Code
- I-Learn
- PathwayConnect
- Track System
- Manwaring Center
- Tutoring Center
- Writing Center

**Common Acronyms**:
- GPA, TOEFL, SAT, ACT, AP
- FAFSA, FERPA, CLEP
- NCAA, ESL, IELTS

## Use Cases

### Entity-Aware RAG
```python
# Route queries based on entities mentioned
if "Financial Aid" in query_entities:
    context = filter_to_financial_aid_entities()
```

### Acronym-Expanded Search
```python
# Automatically expand acronyms in search
query = expand_all_acronyms(user_query)
results = semantic_search(query)
```

### Faceted Navigation
```python
# Filter by entity types
filters = {
    'organization': 'College of Business',
    'location': 'Manwaring Center',
    'term': 'Devotional'
}
```

### Smart Query Routing
```python
# Detect BYU-Idaho terms and route to specialized retrievers
if any(term in query for term in byui_terms):
    use_institutional_knowledge_retriever()
```

### Relationship Extraction
```python
# Find connections between entities
# "Which offices are in Manwaring Center?"
pages_with_both = find_pages_with_entities(['Manwaring Center'], ['organizations'])
```

## Technical Details

**NLP Pipeline:**
1. spaCy en_core_web_sm for NER
2. Pattern matching for acronyms
3. Custom BYU-Idaho term dictionary
4. Regex for action phrase extraction
5. N-gram frequency analysis

**Entity Types:**
- Organizations: spaCy ORG label
- Locations: spaCy GPE label
- People: spaCy PERSON label (filtered)
- Programs: Heuristic-based extraction

**Acronym Detection:**
- Known education acronyms (pre-defined)
- Pattern: `ACRONYM (expansion)`
- Pattern: `expansion (ACRONYM)`

## Citation

```bibtex
@misc{byui-web-content-nlp-2025,
  title={BYU-Idaho Web Content Dataset (NLP-Enhanced)},
  author={Ron Vallejo},
  year={2025},
  version={4.0.0},
  publisher={Brigham Young University-Idaho},
  howpublished={\url{https://huggingface.co/datasets/BYU-Idaho/Web-Content}}
}
```