PURCHASING AND SELLING GOODS RECOMMENDER SYSTEM USING PROGRESSIVE MINING OF SEQUENTIAL PATTERNS (PISA) WITH MULTI CONSTRAINT

• Main Author: Andrian Liwinata, Kevin
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/51143
• Institution: Institut Teknologi Bandung

Bukalapak recommender system has been transformed from Content-Based which uses the similarity between name and item’s description to the similarity between other users’ preferences. However, the recommender system should have attracted users’ preferences holistically. Moreover, there are several special characteristics of ecommerce transactions: have an order by time, have purchasing patterns for every user, and have stakeholders' unique paradigm to simplify decision making. From these three characteristics, PISA algorithm with multi constraint will be used because it gives time domain in form of timestamp and adds constraint as special criteria for the stakeholder. Furthermore, this algorithm eliminates obsolete transactions to give space for new transactions in form of sliding window POI which is determined by the stakeholder. This algorithm then becomes the preprocess for recommender system using users’ purchase history. This algorithm is trained with 9982 synthetic transactions from Bukalapak with 40 items and 49 users and then processed to be recommendations for users with a website as a medium. The experiment result shows that only 0 or 1 item is recommended with algorithm PISA with multi constraint as preprocess because the synthetic transactions are too few. This means more transactions are needed to give more recommendations.