During the Covid-19 pandemic, many local restaurants have been hard hit by lost customers and revenue. For us, it was Agas, Myung Dong, Sha la la Ramen, and numerous other hole-in-the-walls that simply disappeared into thin air. In their time of need, they could have benefited from an app that encourages delivery options and boosts smaller, high-quality businesses. And on our side, lack of other options besides indoor dining kept us from our favorite comfort foods; an interface that connected hungry college students with their favorite eateries would help both sides.
What it does
We've built a website that gives you recommendations for businesses big and small, making it easier to spot high-quality, lesser known places to eat. You can search any restaurant and find the closest recommendations, whether it be the most popular or an unknown. These restaurants can make pages within the website, featuring delivery options(with links!) and analytics to help them manage orders. We aim to ease the barrier of online food delivery and servicing on both the customer and the restaurant sides, and maybe introduce people to new places they'll love in the process. :)
How we built it
We used several academic datasets from Yelp in order to build a recommendation system in Python. By combining checkins, ratings, location, cuisine categories, and number of reviews, we were able to determine which restaurants were most similar to each other. We then searched for businesses that were relatively unknown but still had outstanding reviews, making a point to recommend those in a separate search specifically for up and coming businesses. The recommendation system itself was built with ski-kit learn, by creating only numerical features and taking cosine similarities between all the rows. This correlation matrix is the basis for our recommendations. The website itself was built and designed in weebly, with example restaurants and corresponding analytics populated within as a proof of concept. These pages contain Grubhub, Doordash, or Ubereats links to the restaurant, as well as recommendations and other information about it.
Challenges we ran into
The recommendation system came with many challenges, such as the categorical encodings and the repeated chain restaurants. Data ingestion was also a new challenge because of the size and format of the datasets, which none of us had worked with before. Additionally, embedding the recommendations into the website was a challenge with our minimal HTML/JS knowledge and weebly's resistance to outside elements. The website itself took much time to design and integrate with other services.
Accomplishments that we're proud of
This was a new machine learning challenge for us, and I believe we tackled it with efficiency and tenacity. Our recommendations are without a doubt related to the original and boost small businesses effectively, something that we have not found elsewhere. This was also our first time creating a more custom website, but it was a challenge we overcame with creativity and experience. Additionally, I believe our successful execution of our idea is both useful, relevant, and new, and I hope you share this belief.
What we learned
We gained a lot of experience in more machine learning topics, data ingestion, web embeddings, and a hefty amount of respect for website creators and designers.
What's next for RestauranTours
Expanding our restaurants beyond Austin, creating sign ins/verification for restaurant owners, further development of restaurant pages, and completing the integration between our website to delivery services. Thank you for reading and we hope you enjoy looking through our project!