Study of Cold-Start Product Recommendations and Its Solutions

EasyChair Preprint 9601

Abstract

In today's digital era, consumers rely more and more on the systems that provide them with a personalised experience. In interacting with the system, these consumers create more and more data of different types (click-through rates, items viewed, time spent, number of purchases, and other metrics.). This extensive collection of data from various users is used only to improve the personalising experience of the users. These systems that utilise consumers' data to create a more personalised and customised user experience are called Recommender Systems. Recommender Systems play a huge role in helping companies create a more engaging user experience. E-Commerce giants like Amazon and Flipkart employ such Recommender Systems. These can learn from the user-system interaction the likes and dislikes of users and can promote the visibility of items that interest the user. They are also helpful in luring the customer to buy those things he would have to search for manually in the absence of such a system, which can recommend the item to a user based on his previous interactions. Streaming services like YouTube, Amazon Prime Video and Netflix also use Recommender Systems to suggest movies/shows that the user might like based on the watch history. This study proposes a hybrid model with item-item collaborative filtering using a graph, user-user collaborative filtering based on textual reviews and ratings, and demographic data to generate accurate product recommendations that address the cold-start issue.

Keyphrases: Recommender Systems, cold-start issue, collaborative filtering, hybrid model, item-item graph, user-user collaborative filtering

@booklet{EasyChair:9601,
  author    = {Deep Pancholi and C. Selvi},
  title     = {Study of Cold-Start Product Recommendations and Its Solutions},
  howpublished = {EasyChair Preprint 9601},
  year      = {EasyChair, 2023}}