SESSION BASED RECOMMENDATION MODEL USING MULTI FILTER SCANN

• Main Author: Brian Osmond, Andrew
• Format: Dissertations
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/87580
• Institution: Institut Teknologi Bandung

The marketplace has become a part of our daily lives. With the convenience and ease it offers; an increasing number of people are turning to online shopping. However, this rapid growth has introduced new challenges in efficiently and effectively matching buyers, sellers, and couriers. The growing complexity of marketplace data further exacerbates these challenges, particularly in delivering recommendations that are relevant, fast, and accurate for users. Various approaches have been developed to address scalability challenges, including algorithms such as ScaNN, which is known for its effectiveness in performing searches on large-scale datasets. Nevertheless, ScaNN still faces limitations in handling duplicate search results and maintaining search speed. This presents an opportunity to optimize the ScaNN algorithm through novel strategies that can reduce computational load without compromising accuracy. This research aims to develop a session-based recommendation model that enhances the ScaNN algorithm through the Multi Filter approach. By incorporating a data preprocessing stage, opportunities arise to expedite computation time and minimize potential information loss in the recommendation process. The Yoochoose dataset was utilized in simulations to evaluate the performance of the proposed method. Results indicate that the Multi Filter ScaNN approach improves efficiency by up to 32 times compared to standard ScaNN while maintaining recommendation accuracy. Furthermore, implementing this model as an API demonstrated its ability to handle large-scale requests in real-time simulations. This solution opens avenues for further advancements in session-based recommendation systems across various domains.