Browsing by Subject "recommendation systems"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Exploring the Balance Between Novelty and Familiarity in Recommendation Systems
    (2018-09) Kumar, Vikas
    The balance users seek between the comfort of familiar recommendations and the excitement they solicit in novel ones is a challenge for recommendation systems. On the one hand, familiar options help improve the trust and confidence of users in the system. On the other hand, novel options play a key role in providing serendipity - the delightful surprise that makes system more engaging and useful. However, in their pursuit to achieve the delicate balance, existing recommendation techniques have overlooked user-specific needs and assumed that users have the same, constant appetite for the amount of novelty and familiarity in their recommendations. This thesis highlights and emphasizes the dynamics in user consumption of familiar versus novel items and explores the balance between the two in recommendations. Studying users' consumption patterns in online music streaming we first show that users have distinct and dynamic appetites for novelty in their consumption. We show how a recommender adaptive to the varying appetite of users' novelty consumption is more accurate than traditional one-size-fits-all approaches. Second, we show that not only do users have a distinct appetite, but that there exists a systematic relationship between the novel items they consume and the time elapsed between successive sessions. Third, we address the limitations of inferences from activity logs and the assumptions we impose on actions taken by users in developing algorithms. Instead, we use a qualitative approach in which we interview users while they engage in music listening in their everyday environments to identify how a combination of factors, such as attention needs, exposure to artists or songs etc., influence the balance users seek between novelty and familiarity in their selection. Finally, apart from analysis of what users consume, this thesis also demonstrates the implications of individual familiarity and novelty on the content users produce in online social platforms. Analyzing online location-tagged photos shared by users on Flickr and the familiarity of the users with a location, we show that the locals, who are more familiar with a location, capture more diverse photos of the location, yet it is the tourists who, in their short stay and being less familiar, capture more representative photos of the location. The thesis aims to provide a guiding tool to define, measure, and model the dynamics of the familiarity and novelty balance users consume on online media platforms. The simplicity of our method and its ability to be embedded within existing recommendation techniques supports the contribution of this thesis as well as its general adaptability to other domains of user interactions.
  • Thumbnail Image
    Item
    Towards Recommendation Systems with Real-World Constraints
    (2018-09) Christakopoulou, Konstantina
    Recommendation systems have become an integral part of our everyday lives. Although there have been many works focusing on recommendation quality, many real-world aspects of the recommendation process are typically overlooked: How can we ensure that the very top recommendations users see are engaging? How to recommend venues matching user interests, while preventing many users from being directed to the same venue? Can we design recommenders which first converse with users and then give a recommendation? What is the best way to model recommendation systems as interactive systems, while learning on-the-fly the user-item structure? To what extent can a malicious party perform machine learned adversarial attacks against a recommender? The goal of this thesis is to pave the way towards the next generation of recommendation systems tackling such real-world challenges to improve the user experience, while giving good recommendations. This thesis, bridging techniques from machine learning, optimization, and real-world insights, introduces novel tools to address the above questions. We focus on three directions: (1) encoding real-world constraints into the objective functions, (2) learning to interact with users, and (3) modeling machine learned fake users with malicious goals. For the first direction, by adjusting the optimization objective to capture real-world constraints---(1a) the screen space is small, creating the need for the top recommendations to be relevant, (1b) the item capacities are limited---we suitably guide the learning of model parameters. For the second direction, to balance the need to explore users' preferences with the desire to exploit what has been learned, at a large user and item scale, we combine interactive learning techniques with the principle that similar users tend to behave similarly. This combination results in novel recommendation systems that learn to (2a) converse with new users, and (2b) collaboratively interact with users. For the third direction, taking the perspective of an adversary of the recommender, we use machine learning to learn fake user profiles, which are indistinguishable from real ones, while having a malicious goal. Illustrating the vulnerability of modern recommenders to machine learned attacks will arguably create new directions for designing robust recommendation systems against such attacks.