BeyondStars

Authentic Ratings Redesigned!

Comment

Inspiration

Have you ever left a restaurant feeling that the online reviews you trusted didn’t quite match your experience? How often have you found the perfect spot, only to be deterred by the snarling traffic outside? We're here to address these very questions that affect every diner's choice.

Current restaurant rankings heavily rely on customer reviews. However, a large portion of diners don't leave feedback, potentially skewing these rankings. Moreover, traffic congestion, a factor impacting the dining experience, is rarely considered. Did you know that 53% of young adults base their dining decisions on these often incomplete online reviews? Also a hidden factor often overlooked in restaurant reviews is the impact of traffic congestion and parking spaces. Isn't it curious how such a critical element of our dining experience is left out of the narrative?

What it does

We are proposing a novel approach that transcends traditional review-based rankings. We're leveraging advanced data analytics to provide a more comprehensive and dynamic restaurant ranking system. Our platform integrates examines both current and historical data regarding consumer travel behaviors and traffic trends.

How we built it

We are utilizing INRIX's Trade Area Trips API, combined with Google Places API, to analyze real-time and historical data on customer travel patterns and traffic conditions. Our system evaluates key factors like how recent the visits are, the distance customers travel, their frequency of visits, and the traffic conditions en route. Based on these factors, we develop a composite score for each restaurant in real-time.

For consumer side, we have iOS and Android compatible app where we fetch the scoring database for the restaurants nearby user. The user has 3 views: Map View, List View, and Recommendation View.

Map View: Shows the user all the restaurant along with ratings in their vicinity List View: Fetches the list of all the restaurants, and sort them according to the ratings. Recommendation View: Recommends user a carousel of top 5 restaurants, based on the score.

For client/restaurant side, we have a web dashboard which has 3 charts:

  1. Daily visits for a restaurant.
  2. Time-based visits for a restaurant.
  3. A radar chart depicting the real-time score for the particular restaurant.

Challenges we ran into

  1. The assumption that we are taking here is that if the end location coordinates is under 100m radius, the consumer is visiting the restaurant. Although the generalization might not hold true in some cases, our assumption stands on the basis of proximity.

  2. The Trade Area Trips API takes a polygon (bbox) as location coordinates, rather than commonly used circle. Accounting for the range, this might lead to a little bit of inaccuracy.

  3. We only had access to previous 15 days of data, which makes it hard to model on the basis of frequency of the restaurant visits. If we had a extended time period, frequency score would have played a bigger role

Accomplishments that we're proud of

  1. We were able to built iOS and Android apps for consumer side, and the Web Dashboard for the restaurant/client side.

  2. Although the data we received was very small for training, we were able to determine a composite score model which fit our expectations.

  3. We achieved to an extent what we set out to do. The ratings are better than the Google Reviews, covering the entire diaspora of the consumers over the entire duration.

What we learned

  1. We got to experiment with lot of new tech stack. For a lot of us, this was either a first project of React Native or Spring Boot. We faced a lot of difficulties and found a way around, through different OS deployments.

  2. We learnt how to deal with spatial data and mapping it. For all of us, this was the first time we worked with any kind of location based data.

  3. The implementation of composite score through a back-and-forth training on limited data taught us how to implement modeling in the lack of data. This could happen a lot in the real-world where we might not get the cleanest, most perfect of data.

  4. We used to think of mapped location usually spread across a radius. But we learnt new way to map the location based coordinates.

What's next for BeyondStars

  1. Firstly, the data cut we used for current training could lead to some inaccuracies. We aim to re-train the model to enhance predication accuracy.

  2. Add more reports, and analytics for the client/restaurant side. There are a lot of insights uncover, and views to driven from the data we are extracting.

Share this project:

Updates