Developing a Hybrid Recommender System for Some of Life’s Most Important Choices
Submitted by Paul Meinshausen (@pmeins) on Monday, 15 June 2015
Recommender Systems are both an old and an active area of research. Advances in Recommender Systems can emerge from developing applications in new contexts and for new use cases. In this session we will describe the unique challenges associated with building a recommender system for real estate and we will present the work we are doing to develop a hybrid recommender system for real estate at Housing.com.
The recommendation problem in the property and real estate context has several domain specific characteristics which strongly influence the design and algorithmic approach for making recommendations to users. Some of the challenges of the real estate domain include perishable, metamorphic and rapidly changing inventory; constrained and ambivalent users with strongly conditional preferences; and the need to draw inferences about users’ preferences from infrequent transactions with little to no explicit user feedback. Some of the implications of these challenges are that content-filtering approaches are inappropriate and that collaborative filtering approaches are also incapable of providing a robust solution.
We will present an innovative approach we have developed at Housing’s Data Science Lab for modeling user preferences and a hybrid approach for making inferences about users’ interests.
There are no requirements for this session
Vedavyas Chigurupati is a Data Scientist at Housing.com. He was previously working with SAP Labs India as a Software Developer. He has been playing an active and leading part in developing the recommender system at Housing.