A collaborative filtering-based recommendation system for an online high-end retailer

Official URL: https://risc01.sabanciuniv.edu/record=b2579099_(Table of contents)

In online retail platforms, consumers seek to find the products that are best suited for their needs while limiting their search efforts. With the growing trend in online shopping, retail companies utilize a range of tools to assist customers in their journey and improve their purchase experience. One of the tools that can minimize these exploration efforts is recommendation systems that suggest a tailored set of available product options to consumers based on their preferences. In this thesis, we focus on a high-end Turkish retailer that did not utilize such engines in its practice and study the value that these systems can provide to the company. To that end, we implement a collaborative filtering-based recommendation system that uses the similarity of the consumers to derive their preferences and suggest item sets for their next purchase. We evaluate the recommendation model with transactions data to acquire the hyper-parameters and test it on the transactions made in the last month and provide recommendations on three granularity levels. We also analyze the predicted preferences to suggest bundling options and derive empirical insights. We found that by generating 20 suggestions for each customer in their shopping session, the engine can reach an accuracy of 38% at brand-level and 7% at itemlevel while using cosine similarity as its similarity metric.