meetara-lab/vectorstore-mental_health

Vectorstore Dataset: Mental Health

Overview

This dataset contains pre-computed vector embeddings for the mental health domain, ready for use in Retrieval-Augmented Generation (RAG) applications, semantic search, and knowledge base systems. The embeddings are generated from high-quality source documents using state-of-the-art sentence transformers, making it easy to build production-ready RAG applications without the computational overhead of embedding generation.

Key Features

• ✅ Pre-computed embeddings: Ready-to-use vector embeddings, saving computation time
• ✅ Production-ready: Optimized for real-world RAG applications
• ✅ Comprehensive metadata: Includes source file information, page numbers, and document hashes
• ✅ LangChain compatible: Works seamlessly with LangChain and ChromaDB
• ✅ Search-optimized: Designed for fast semantic similarity search

What's Included

This dataset contains 407 text chunks from 29 source documents, each pre-embedded using the sentence-transformers/all-MiniLM-L6-v2 model. Each chunk includes:

• Text content: The original document text
• Embedding vector: 384-dimensional float32 vector
• Rich metadata: Source file, page numbers, document hash, and more

Dataset Details

Dataset Summary

• Domain: mental_health
• Total Chunks: 407
• Total Documents: 29
• Database Size: 27.12 MB (7 files)
• Embedding Model: sentence-transformers/all-MiniLM-L6-v2
• Chunk Size: 1000
• Chunk Overlap: 200

Dataset Structure

The dataset contains the following columns:

• id: Unique identifier for each chunk
• embedding: Vector embedding (numpy array, dtype=float32)
• document: Original text content of the chunk
• metadata: JSON string containing metadata (file_name, file_hash, page_number, etc.)

Embedding Model

This dataset uses embeddings from: sentence-transformers/all-MiniLM-L6-v2

Usage

Loading the Dataset

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("meetara-lab/vectorstore-mental_health")

# Access the data
print(dataset["train"][0])
# Output:
# {
#     'id': '...',
#     'embedding': array([...], dtype=float32),
#     'document': '...',
#     'metadata': '{"file_name": "...", "page": 1, ...}'
# }

Loading Back into ChromaDB

from langchain_chroma import Chroma
from langchain_huggingface import HuggingFaceEmbeddings
from datasets import load_dataset
import json

# Load dataset
dataset = load_dataset("meetara-lab/vectorstore-mental_health")["train"]

# Initialize ChromaDB
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
vectorstore = Chroma(
    persist_directory="./chroma_mental_health",
    embedding_function=embeddings
)

# Add documents to ChromaDB
documents = []
metadatas = []
ids = []
embeddings_list = []

for item in dataset:
    ids.append(item["id"])
    embeddings_list.append(item["embedding"].tolist())
    documents.append(item["document"])
    metadatas.append(json.loads(item["metadata"]))

# Note: You'll need to use ChromaDB's Python client directly for custom embeddings
import chromadb
client = chromadb.PersistentClient(path="./chroma_mental_health")
collection = client.create_collection(name="mental_health")

collection.add(
    ids=ids,
    embeddings=embeddings_list,
    documents=documents,
    metadatas=metadatas
)

Using with LangChain

from langchain_chroma import Chroma
from langchain_huggingface import HuggingFaceEmbeddings

# Initialize retriever
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
vectorstore = Chroma(
    persist_directory="./chroma_mental_health",
    embedding_function=embeddings
)

# Load from HF Hub first (see above), then use with LangChain
retriever = vectorstore.as_retriever()
results = retriever.invoke("your query here")

Domain-Specific Usage Examples

This vectorstore is optimized for Mental Health domain queries. Here are practical examples:

Example Queries

from langchain_chroma import Chroma
from langchain_huggingface import HuggingFaceEmbeddings

# Load vectorstore (see "Loading Back into ChromaDB" above)
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
vectorstore = Chroma(
    persist_directory="./chroma_mental_health",
    embedding_function=embeddings
)
retriever = vectorstore.as_retriever(search_kwargs={"k": 3})

# Example queries for mental health domain:
example_queries = [
    "What are the early warning signs of depression and when should I seek help?",
    "What are effective cognitive behavioral techniques for managing panic attacks?",
    "How can I support a loved one who is experiencing PTSD symptoms?",
    "What are evidence-based strategies for managing chronic anxiety in daily life?",
    "How does trauma therapy work and what can I expect from treatment?"
]

# Run a query
query = "What are the early warning signs of depression and when should I seek help?"
results = retriever.invoke(query)

# Display results
for i, doc in enumerate(results, 1):
    print(f"\nResult {i}:")
    print(f"  Source: {doc.metadata.get('file_name', 'Unknown')}")
    print(f"  Page: {doc.metadata.get('page', 'N/A')}")
    print(f"  Content: {doc.page_content[:200]}...")

Common Use Cases

This dataset is useful for:

• Mental wellness assessment and guidance
• Evidence-based therapy information and techniques
• Self-care and coping strategy development
• Crisis intervention and support resources
• Psychological education and awareness

Real-World Example

# Complete example: Query and use results
from langchain_chroma import Chroma
from langchain_huggingface import HuggingFaceEmbeddings

# 1. Initialize (after loading from HF Hub)
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
vectorstore = Chroma(
    persist_directory="./chroma_mental_health",
    embedding_function=embeddings
)

# 2. Create retriever with relevance filtering
retriever = vectorstore.as_retriever(
    search_type="similarity",
    search_kwargs={
        "k": 5,  # Get top 5 most relevant results
        "score_threshold": 0.7  # Minimum similarity score
    }
)

# 3. Query the vectorstore
query = "What are effective cognitive behavioral techniques for managing panic attacks?"
docs = retriever.invoke(query)

# 4. Process results
for doc in docs:
    metadata = doc.metadata
    print(f"📄 File: {metadata.get('file_name', 'Unknown')}")
    print(f"📃 Page: {metadata.get('page', 'N/A')}")
    print(f"📝 Content: {doc.page_content[:300]}...\n")

Dataset Statistics

Content Statistics

• Total Chunks: 407
• Total Documents: 29
• Average Chunks per Document: 14.0
• Database Size: 27.12 MB (7 files)

Technical Specifications

• Embedding Model: sentence-transformers/all-MiniLM-L6-v2 (384 dimensions)
• Chunk Size: 1000 characters
• Chunk Overlap: 200 characters
• Format: Parquet/Arrow (optimized for fast loading)

Performance Considerations

Loading Time

• Full dataset loads in ~5-15 seconds on average hardware
• Memory usage: ~0.6 MB for embeddings alone
• Recommended RAM: 2GB+ for full dataset operations

Search Performance

• Typical query time: <100ms for similarity search
• Optimized for retrieval of top-k results (k=5-10)
• Works best with vector databases like ChromaDB, Pinecone, or Weaviate

Citation

If you use this dataset, please cite:

@dataset{meetara_vectorstore_mental_health,
  title={meeTARA Vectorstore: Mental Health},
  author={meeTARA Lab},
  year={2024},
  url={https://huggingface.co/datasets/meetara-lab/vectorstore-mental_health}
}

Limitations and Considerations

• Language: This dataset is monolingual (English only)
• Domain specificity: Optimized for mental health domain queries
• Embedding model: Uses sentence-transformers/all-MiniLM-L6-v2 - ensure compatibility if switching models
• Update frequency: Dataset reflects state at time of publication; source documents may have been updated

Alternatives and Related Datasets

Looking for other domains? Check out other meeTARA vectorstore datasets:

• meetara-lab/vectorstore-general_health - General health and medical information
• Additional domain datasets coming soon!

Maintenance and Updates

This dataset is maintained by the meeTARA Lab team. For updates, bug reports, or feature requests, please visit our GitHub repository.

License

This dataset is released under the Apache 2.0 License. This means you are free to:

• Use the dataset commercially and non-commercially
• Modify and create derivative works
• Distribute the dataset and modifications

Please see the full license text for complete terms.

Citation

If you use this dataset in your research or applications, please cite it as:

@dataset{meetara_vectorstore_mental_health,
  title={meeTARA Vectorstore: Mental Health},
  author={meeTARA Lab},
  year={2024},
  url={https://huggingface.co/datasets/meetara-lab/vectorstore-mental_health},
  license={apache-2.0},
  task={feature-extraction, text-retrieval, rag}
}

Contact and Support

• GitHub: meetara-lab/meetara-core
• Issues: Report bugs or request features on GitHub Issues
• Documentation: Visit our repository for detailed documentation

---

Made with ❤️ by the meeTARA Lab team