Customer Personalization @ nearbuy
Download as PPTX, PDF0 likes145 views
Ankit Kohli from Nearbuy discusses their personalization efforts through creating a data pipeline in Kafka for clickstream data, transforming the data with Spark streaming, and storing it in HBase. Kohli then covers building an ML pipeline in Spark to extract features and run algorithms like ALS, k-means, and linear regression for recommendations and real-time analytics. Common challenges like data volume, streaming vs batch processing, and Spark cluster issues are also addressed.
Customer Personalization @ nearbuy
- 1. Personalization @ Nearbuy Ankit Kohli
- 2. About Me - Data Scientist at Nearbuy - Creating a world of personalization for Nearbuy’s customers
- 3. AGENDA - Creating Data Pipeline - Kafka ( Real Time Click Stream) - Hbase ( where data sits) - Data Transformation - ML Pipeline - Spark ( Data Feature Extraction ) - ML Algos ( how to use Spark and use cases of each algo) - ALS
- 4. Data Pipeline @ Nearbuy Schema less DB Click Stream KAFKA SPARK STREAM HBASE SPARK ML RECOMMEN DATIONS REAL TIME ANALYTICS
- 5. KAFKA EVENTS Deal view Event { “customerId”:”XXXXXXXXX”, “timeStamp”:”140983388484” “dealId”:”YYYYY”, “source”:”APP”, “os”:”android” ……. }
- 6. Spark Streaming - Details about how to implement Spark Streaming and set up jobs that runs 24x7 to ingest all click stream data - Implement sessionization of customer activity on Nearbuy - Transform data and store in HBASE
- 7. Spark ML Pipeline Data Feature Extraction - Categorical Data Will talk about Spark ML Algos - Collaborative Filtering ( Implicit & Explicit Feedback ) - K Means - Linear Regression Will explain each algo in depth and its use cases and how to implement it using SPARK.
- 8. Common Pitfalls - Too much data - Stream vs Batch Data - Customer Sessionization - Spark Cluster Mode Issues
- 9. Takeaways - Personalization of APP can be done in many forms , our use case is one of them ( ecommerce ) - Data visualization and ML Model selection - Anyone who is interested to start with Data Analytics will get great insight - Someone who has to start ML will get to know how to use SPARK . - Someone who is already doing this can get to know how other companies are implementing SPARK for ML - Spark ML best Practices