Deep Learning Recommendation System for Property Management Software

About the Project

We created a recommender system that predicts users interests and recommends properties to users based on previous user interactions and property features. Recommender systems are among the most powerful machine learning systems that are used to help users identify items they may be interested in and also helps companies increase interaction and sales. Our sponsor is AppFolio, a tech company based in Santa Barbara that provides innovative software, services, and data analytics to the real estate industry. Building this model will allow AppFolio to analyze the results of our recommender system to assess how interaction data can be used to better their products. We explored multiple models including the Bilateral Variational Autoencoder (BiVae), TriVae, Visual Bayesian Personalized Ranking (VBPR), and Deep Cross Network (DCN) in an effort to achieve the highest prediction accuracy using 16 property features from our given dataset and images. Ultimately, we want to integrate our model into the Santa Barbara property website, which was built by the AppFolio Sponsored CS Capstone team.