Inspiration
Talking to Playster and Conveo at the coding challenge.
What it does
Takes Yelp user ratings and written reviews for restaurants and predict personalized restaurants ratings for users, of which we can pick the higher rated restaurants for recommendation.
How I built it
Use a Singular Value Decomposition (SVD) to factorize the sparse user ratings matrix, which then allows us to extract a K-dimensional latent vector for each individual users.
Use a Latent Dirichlet Allocation (LDA) to model the restaurant reviews and use the latent probabilities as the K-dimensional latent factors for each restaurant.
For predicting the user rating for a restaurant take the inner product of the two latent factors, taking into account potential biases each user or restaurant might have.
Challenges I ran into
I wanted to use images associated with the restaurant for latent factor modelling as well, but had no time.
Accomplishments that I'm proud of
I stayed up the whole night.
What I learned
There must be good money to be made in Data processing .
What's next for Ratings + Reviews Recommender
Nothing
Built With
- gensim
- scikit-learn
Log in or sign up for Devpost to join the conversation.