Food Bars Recommender System


As a fun little experiment, we've sampled numerous food bars and collected data on our team's preferences on a number of different attributes (taste, texture smell, etc.) Using this data, we've built a recommender system that provides predicted ratings for the bars that each person hasn't yet rated. To begin, choose an attribute and click the buttons corresponding to the action you wish to take.

For more information, check out our blog.


Recommendation Data

The recommendation data below gives the standard wide format view of the data. The individual rows are the users, and the columns are the ratings on the attribute selected on the left. Where ratings are missing, the recommendation system will attempt to infer the rating of the user (See the recommendations tab).

Raw Data

Here is the original raw data prior to transformation. In this case, the different attributes are along the columns and the bars and tasters are along the rows. We transform this data to get the above structure (with columns as bars and tasters as rows) in order to perform the recommendation.

Click Predict Interests to see the predictions.

Predicted Ratings

The predicted ratings for the top 5 bars for the selected user are given in the plot below.


The same data from above is given in table format below.