PLENTRY: A Recipe Recommender

In this project, I worked with three students who studied Industrial Design. We aimed to develop a personalized and AI driven dinner recipe recommendation application, that allows users to maximally use leftover ingredients in their pantry and fridge. We discussed the features for the application and I took the charge of data mining and documentation because I had more background knowledge about statistics. I used WEKA (a software for operating machine learning) as a starting point to understand machine learning and acquire skills in dealing with data. With the collected data, first I merged two datasets: the user profile dataset and the interaction log dataset. Then, I cleaned up these data. For example, removing incomplete instances, attributes without significant variations, re-grouping some numeric data to reduce dimensions and making the data to be more evenly distributed. I tested the data with different kinds of algorithms (e.g., J48, RandomForest, ZeroR) and selected suitable predictors to achieve decent and reasonable predicting results. The final prototype presents users with personalized recipe recommendations after submitting their personal profiles and offers a functionality to filter the datasets on their available ingredients.

In this project I...

• Discussed and constructed the conceptual interaction interface with teammates

• Created fake persona to add more variation to user profile attributes within the collected datasets

• Employed machine learning algorithms on the data through Weka to establish a prediction model suitable for integration in the application

• Gained an initial understanding of machine learning and got acquainted with some algorithms

• Tools: Weka (Machine Learning), Xtensio (User Persona)