arrow_back Anti-patterns in designing machine learning systems
An Approach for recommending TopK Digital Artworks
Submitted by Vivek Anand Rao T S (@vtemker) on Monday, 2 May 2016
We have shown how recommender systems apply to the online digital artwork domain. The goal was to test the ability of recommender systems to aid artists in discovering artwork relevant to their likings. The users were from the online digital artwork sharing community, using the PENUP application. We have used information retrieval based metrics to measure the performance of a few key algorithms implemented in Apache Mahout. The approach shows how we can integrate trust based social information into the standard Nearest Neighbor (NN) and Matrix factorization (MF) algorithms for artworks. It also shows how such a system can be measured for quality of recommendations, allowing one to deploy a practical system on productiont.
Recommender systems enable online users to have a more personalized experience. To the user, the system appears to somehow understand and capture the user’s likes and dislikes. There are various explicit and implicit “signals” picked from the user’s online activities. With this information, relevant content is automatically chosen and customized for the user.
Digital art is an artistic work or practice that uses digital technology as an essential part of the creative or presentation process. Samsung S Pen with Note is a tool for creating digital arts. Samsungs PEN.UP is a social networking service for people who create digital art. It has a community of over 1.7 million users and has more than a million artworks.
Interests in subjects related to artwork vary greatly and are subjective. What are very interesting to some users may be of little to others. For e.g. some people are interested in gaming related themes while others are interested in nature related themes. As the number of artworks grew, discovery of artworks and presenting the relevant artworks to the users based on their interests became a challenge. Usual methods of classification based on categories and tags did not work very well as the tags are user contributed and users who want more coverage for their arts add many categories and tags.
This paper studies the attempts towards improving the performance of standard recommender algorithm on artworks by the incorporation of social information. When an artist follows another artist, it can be said with more confidence that the likes and dislikes are in agreement. Such top – k recommender systems using social information have been studied before, but our work applies them to the field of digital artworks.
An enthusiast of machine learning and a recent entrant to the field. Enjoying learning and applying ML techniques. Working at Samsung R&D.