BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:When Diversity Becomes Relevant—A Multi-Category Utility Model o
 f Consumer Response to Content Recommendations - Elie Ofek\, Harvard Busin
 ess School
DTSTART:20170616T113000Z
DTEND:20170616T130000Z
UID:TALK73049@talks.cam.ac.uk
CONTACT:Luke Slater
DESCRIPTION:Sometimes we desire change\, a break from the same\, or an opp
 ortunity to fulfill different aspects of our needs. Noting that consumers 
 seek variety and realizing that common recommender systems often concentra
 te only on a narrow set of products\, several approaches have been develop
 ed to diversify suggested items. However\, current diversification strateg
 ies operate under a one-shot paradigm and are not guided by the ability to
  enhance consumer utility by taking into account the evolution of preferen
 ces contingent on past consumption. This often leads to inaccurate predict
 ions of what item will be most relevant for an individual at the current s
 tage. By recognizing that choices in a session are the result of a sequenc
 e of utility maximizing selections from various categories\, we show that 
 one can increase recommendation accuracy by dynamically tailoring the dive
 rsity of items suggested to the diversity sought by the consumer. Our appr
 oach is based on a multi-category utility model that captures a consumer
 ’s preference for different types of content\, how quickly she satiates 
 with one type and wishes to substitute it with another\, and how she trade
 s off her own costly search efforts with selecting from a recommended list
  to discover new content. Taken together\, these three elements allow us t
 o characterize how a consumer constructs a “basket” of content over th
 e course of each session\, and how likely she is to click on content recom
 mended to her. We estimate the model using a clickstream dataset from a la
 rge media outlet and apply it to determine the most relevant content at di
 fferent stages of an online session. We find that our approach generates r
 ecommendations that are on average about 10% more accurate than optimized 
 alternatives. Moreover\, the proposed method recommends content that more 
 closely matches the diversity sought by readers in their actual consumptio
 n—exhibiting the lowest concentration-diversification bias when compared
  to other personalized recommender systems. Using a policy simulation\, we
  estimate that recommending content using our approach would result in vis
 itors reading 23% additional articles at the studied website\, which has d
 irect revenue implication for the publisher of this site.
LOCATION:  KH107\, Keynes House\, Judge Business School
END:VEVENT
END:VCALENDAR