Project Overview

• Name: Book Recommender (semantic + emotion-aware)
• Description: A semantic book recommender that combines document embeddings and emotion signals to surface emotionally relevant book recommendations. The project includes a Gradio dashboard (dashboard.py), data CSVs used for recommendations, and several Jupyter notebooks for exploration and modeling.

Key Files

• dashboard.py: Gradio app that runs a semantic recommender using HuggingFace sentence embeddings and a Chroma vector DB.
• books_with_emotions.csv, books_with_categories.csv, cleaned_books.csv: Datasets used by the project.
• tagged_description.txt: Text data used to build the embedding index.
• Notebooks: data-exploration.ipynb, sentiment-analysis.ipynb, text-classification.ipynb, vector-search.ipynb — for EDA and model experiments.

Requirements

• Use the provided requirements.txt to install Python dependencies:

python -m venv .venv
; .\.venv\Scripts\Activate.ps1
; pip install -r requirements.txt

Running the Dashboard

1. Ensure the virtual environment is active (see above).
2. If you rely on any environment variables, add them to a .env file in the repo root. dashboard.py calls load_dotenv().
3. Start the app:

python dashboard.py

The dashboard uses a HuggingFace sentence-transformer (sentence-transformers/all-MiniLM-L6-v2) to compute embeddings and builds a Chroma vector store from tagged_description.txt. The app loads books_with_emotions.csv to provide metadata and emotional signals.

Notebooks

• Use the notebooks for exploration and to reproduce preprocessing, sentiment analysis, and vector search experiments. They contain the data-cleaning and modeling steps used to generate the datasets.

Models Used

• sentence-transformers/all-MiniLM-L6-v2 — used via HuggingFaceEmbeddings (LangChain) to compute sentence embeddings for semantic search and the Gradio dashboard (dashboard.py and vector-search.ipynb).
• bhadresh-savani/distilbert-base-uncased-emotion — used with transformers.pipeline("text-classification") in sentiment-analysis.ipynb to infer emotions/sentiment from book descriptions.
• typeform/distilbert-base-uncased-mnli — used with transformers.pipeline("zero-shot-classification") in text-classification.ipynb for zero-shot topic/class labeling.

Datasets & Outputs

• The CSV files in the repo are the primary inputs for the dashboard. tagged_description.txt is used to build the vector index.

Troubleshooting

• If you get import errors for any langchain_* or community imports, verify package names on PyPI. The project imports community LangChain modules (e.g., langchain_community, langchain_text_splitters, langchain_chroma); if an import fails, try installing langchain-community, langchain-text-splitters, and chromadb.
• If embedding downloads fail, ensure you have internet access or have the model cached with sentence-transformers.