Leveraging recommender systems to personalise search results
summary
Claims that recommender outputs can re-rank search results are demonstrated as a proof of concept using Pyspark to build user/item factors and Elasticsearch to apply them via a function_score query. Dataset sizes are concrete: about 40 million customers and ~85,000 products, with users interacting ~6 products per year. Implementation computes dot products of user and item factor vectors (matrix factorization) and boosts matches; a demo UI with sliders shows relevance gains, but latency and scaling prevent production deployment yet.
AI-written summary · about
chapters (19)
- 0:00Personalizing search results
- 1:03Why search personalization matters
- 2:27Personalization as re-ranking
- 3:50Trade-offs in personalization models
- 5:38Data sources and user profiling
- 7:01Recommendation algorithms for search
- 9:53Matrix factorization for taste profiles
- 11:18Personalization proof of concept
- 13:22Elasticsearch setup
- 16:12Boosting and affinity scoring
- 18:23Demo: personalized search
- 20:21Front-end for stakeholders
- 22:55Next steps in personalization
- 25:03Testing personalization
- 26:12Shared accounts and trends
- 27:45Scalability challenges
- 29:12Real-time personalization
- 30:55Improving relevance with features
- 31:35Measuring success with KPIs
related talks
29mThe Challenges of Building a Personalized Recommendation System in E-Commerce
28mThe Power of Personalized Customer Experiences Through AI Innovations
58mRecommender Systems – A Walkthrough
26mLarge-Scale Recommendation System with Python and Spark
37mA primer on recommendation systems.
31mCreating recommender systems with Spark, Scala and Prediction.Io
35mSpotting trends and tailoring recommendations: PySpark on Big Data in fashion
41mRecommender Systems: Crafting Effective Personalised Experiences