From Search to Discovery at Housing
Mudit Gupta
@mudit-housing
The objective of this session is to introduce a framework and models for search recommendations through real-time user click stream analysis. We will be talking about various architectural challenges and challenges in modeling the expert system and how it can be used in different domains.
Outline
The problem of search discovery at Housing can be broken down into two verticals - personalizing/improving relevance of the result set, and guiding the users to select a search criteria that has a higher chances of conversion. The components of the search recommendation service are user click stream processing and expert system to generate search recommendations. Stream processing builds session profile for the users and generate relevant signals for searches/session with low chances of conversion (broken search). The expert system handles such broken searches and suggests alternate but relevant search criteria. The expert system and the broken search models are updated using user activity and feedback. Result set is personalized based on user profiles and, supply and demand biases in the search criteria.
Requirements
None
Speaker bio
Ravikiran Gunale is a software developer at Housing. His interests include supervised learning systems, NLP and new technologies. He has worked on big data projects related to recommendation system, predictive analytics, fraud detection.
Mudit is a developer at Housing.com and has been leading the search and realtime recommendations at Housing. He is a FOSS enthusiast and has contributed actively to various project inlcuding the collaborative filtering module at mlpack.
{{ errorMsg }}