In this project, we leverage Yelp reviews for coffee shops in the Austin area to create a recommendation system. We use a variety of NLP tools. The recommender has been deployed as an app through Streamlit.
Customers looking for a coffee shop have a variety of needs, and it can be challenging to find a coffee shop that fits what they are looking for. This is especially true in areas that are unfamiliar to the person, for example when on a vacation. A system that could recommend a coffee shop to a person based on their needs and desires would quite useful. This project does exactly this, by using review data and NLP tools.
The data is a collection of reviews from Yelp. There are 7,616 reviews. The reviews were considered on a sentence level for the most part, where each sentence was treated as its own row. There were 58,864 total sentences in the dataset. The data was originally scraped from Yelp, and download from this location here.
- CountVectorizer and TFIDF were both tested for creating the document term matrix
- LSA, NMD, and LDA were all considered for matrix decomposition and dimensionality reduction
- Cosine distance was used for comparison and in the recommender system
- Python, Pandas, and Numpy for standard data science tools
- sklearn for CountVectorizer and matrix decomposition algorithms
- spaCy for lemmatization
- VADER for sentiment analysis
- NLTK for stemming and tokenizing
- Streamlit for create a web app
A summary of my project and results can be found on my slides. In addition, you can use the recommendation system here at the Streamlit app I created. Also, my code and visualizations are posted to my personal GitHub.