How do social processes affect our preferences and interests? How do some items stay "niche" while others end up becoming widely popular?
Answers to these questions can help us understand how information diffuses through society, support sharing of information in online social systems and create better recommendation systems.
I use online traces of human activity to make progress on these broad, largely open, questions. I employ online experiments and data mining from a variety of domains such as entertainment, e-commerce and collaborative online activity to create general models for preference evolution and diffusion.
Microsoft Research
New York, USA
May - August 2014
Worked with Jake Hofman and Duncan Watts.
Estimation of causal impact of recommender systems.
Google Inc.
Mountain View, USA
May - August 2013
Worked with Gueorgi Kossinets.
Inference of attributes for local businesses, studying the evolution of their rating.
LinkedIn Corp.
Mountain View, USA
May - August 2012
Worked with Baoshi Yan.
Novel pairwise models for learning implicit user feedback on recommendations.
CRAFT Lab, EPFL
Lausanne, Switzerland
May - July 2009
Worked with Frederic Kaplan and Pierre Dillenbourg.
Prosodic analysis of speech and visualizations for supporting collaborative dialogue.
IBM Research
New Delhi, India
May - July 2008
Worked with Akshat Verma and Gargi Dasgupta.
Local-optimal search algorithms for dynamic composition of web services.
-
Accounting for preference similarity when estimating
social influence in online sharing networks
Amit Sharma and Dan Cosley
In submission to WWW 2015 [Abstract] [PDF] [Web]
Abstract:
Many online social networks thrive on automatic sharing of people's activities to their friends and followers. A fundamental question in the study of such online sharing networks is how exposure to their friends' activities influences people's future actions. We consider a specific form of social influence, mimicry or copying others' actions, and propose a test to estimate the extent of influence in a sharing network. Our test assumes that non-friends are unlikely to exert any influence on a user's actions, and thus comparing the extent of correlation in activity of a user with her friends versus non-friends can be used to separate the effect of social influence from personal preferences. Unlike past tests for influence from observational data, our test relies only on available activity history of users and directly gives estimates of influence at both the individual and network level.
Experiments on a dataset from Last.fm show that the proposed test can rule out influence for a majority of friends' activities that correlate with each other. To verify whether these results extend to other sites, we also applied our test to data from three other websites with preference data for books, movies and photos. We find while the fraction of correlated actions between friends that could have happened without influence varies, influence estimates for all four websites fall into a narrow range (0.3-0.7%), suggesting that only about one percent of the total user actions on such sharing networks can be attributed to social influence.
-
Estimating the causal impact of
recommender systems
Amit Sharma, Jake Hofman and Duncan Watts
In submission to WWW 2015 [Abstract] [PDF] [Web]
Abstract:
Recommendation systems are an increasingly prominent part of the web, accounting for up to a third of all traffic on several of the world's most popular sites.
Nevertheless, little is known about how much activity such systems cause over and above what would occur if recommendations were absent from these sites.
Although such questions about the causal impact of recommendations usually require costly experiments, here we present a method for estimating these effects from purely observational data.
Specifically, we show that causal identification is possible when a product experiences an instantaneous shock in direct traffic and the products recommended next to it do not.
We then apply our method to browsing logs containing anonymized activity for 2.1 million users on Amazon.com over an 9 month period and analyze over 1,000 unique products that experience such shocks.
We find that while ~30% of all views on these products do indeed come from recommendation click-throughs, at least 80% of this activity would likely occur in the absence of recommendations.
We conclude with a discussion about the assumptions under which the method is appropriate and caveats around extrapolating results to other products, sites, or settings.
-
Studying and modeling the connection between people's preferences and content sharing
Amit Sharma and Dan Cosley
CSCW 2015 [Abstract] [PDF] [Web]
Abstract:
People regularly share items using online social media. However, people's decisions around sharing---who shares what to whom and why---are not well understood. We present a user study involving 87 pairs of Facebook users to understand how people make their sharing decisions. We find that even when sharing to a specific individual, people's own preference for an item (individuation) dominates over the recipient's preferences (altruism). People's open-ended responses about how they share, however, indicate that they do try to personalize shares based on the recipient. To explain these contrasting results, we propose a novel process model of sharing that takes into account people's preferences and the salience of an item. We also present encouraging results for a sharing prediction model that incorporates both the senders' and the recipients' preferences. These results suggest improvements to both algorithms that support sharing in social media and to information diffusion models.
-
Friends, strangers and the value of ego networks for recommendation
Amit Sharma, Mevlana Gemici and Dan Cosley
ICWSM 2013 [Abstract] [PDF] [Web]
Abstract: Two main approaches to using social network information in recommendation have emerged: augmenting collaborative filtering with social data and algorithms that use only ego-centric data. We compare the two approaches using movie and music data from Facebook, and hashtag data from Twitter. We find that recommendation algorithms based only on friends perform no worse than those based on the full network, even though they require much less data and computational resources. Further, our evidence suggests that locality of preference, or the non-random distribution of item preferences in a social network, is a driving force behind the value of incorporating social network information into recommender algorithms. When locality is high, as in Twitter data, simple k-nn recommenders do better based only on friends than they do if they draw from the entire network. These results help us understand when, and why, social network information is likely to support recommendation systems, and show that systems that see ego-centric slices of a complete network (such as websites that use Facebook logins) or have computational limitations (such as mobile devices) may profitably use ego-centric recommendation algorithms.
-
Do social explanations work? Studying and modeling the effects of social explanations in recommender systems.
Amit Sharma and Dan Cosley
WWW 2013 [Abstract]
[Details] [PDF] [Web]
Abstract: Recommender systems associated with social networks often use social explanations (e.g. "X,Y and 2 friends like this") to support the recommendations. We present a study of the effects of these social explanations in a music recommendation context. We start with an experiment with 237 users, in which we show explanations with varying levels of social information and analyze their effect on users' decisions. We distinguish between two key decisions: the likelihood of checking out the recommended artist, and the actual rating of the artist once the user has listened to several songs. We find that while the explanations do have some influence on the likelihood, there is little correlation between the likelihood and actual (listening) rating for the same artist. Based on these insights, we present a generative probabilistic model that explains the interplay between explanations and beckground information on music preferences, and how that leads to a final likelihood rating for an artist. Acknowledging the impact of explanations, we discuss a general recommendation framework that models external informational elements in the recommendation interface, in addition to inherent preferences of users.
-
Pairwise learning in recommendation: Experiments with pairwise recommendation on LinkedIn
Amit Sharma and Baoshi Yan
ACM RecSys 2013 [Abstract]
[Details] [PDF][Slides]
Abstract: Many online systems present a list of recommendations and infer user interests implicitly from clicks or other contextual actions. For modeling user feedback in such settings, a common approach is to consider items acted upon to be relevant to the user, and irrelevant otherwise. However, clicking some but not others conveys an implicit ordering of the presented items. Pairwise learning, which leverages such implicit ordering between a pair of items, has been successful in areas such as search ranking. In this work, we study whether pairwise learning can improve community recommendation. We first present two novel pairwise models adapted from logistic regression. Both offline and online experiments in a large real-world setting show that incorporating pairwise learning improves the recommendation performance. However, the improvement is only slight. We find that users' preferences regarding the kinds of communities they like can differ greatly, which adversely affect the effectiveness of features derived from pairwise comparisons. We therefore propose a probabilistic latent semantic indexing model for pairwise learning (Pairwise PLSI), which assumes a set of users' latent preferences between pairs of items. Our experiments show favorable results for the Pairwise PLSI model and point to the potential of using pairwise learning for community recommendation.
-
Algorithms for generating ordered solutions for explicit AND/OR structures: Extended
abstract.
Priyankar Ghosh, Amit Sharma, Partha Pratim Chakrabarti, and Pallab Dasgupta.
IJCAI 2013 [Abstract] [PDF]
Abstract: We present algorithms for generating alternative
solutions for explicit acyclic AND/OR structures in
non-decreasing order of cost. Our algorithms use a
best first search technique and report the solutions
using an implicit representation ordered by cost.
Experiments on randomly constructed AND/OR
DAGs and problem domains including matrix chain
multiplication, finding the secondary structure of
RNA, etc, show that the proposed algorithms per-
form favorably to the existing approach in terms of
time and space.
-
Network-centric recommendation: Personalization with and in social networks
Amit Sharma and Dan Cosley
IEEE SocialCom 2011 [Abstract] [PDF] [IEEE]
Abstract: People often rely on the collective intelligence of their social network for making choices, which in turn influences their preferences and decisions. However, traditional recommender systems largely ignore social context, and even network-aware recommenders don't explicitly support social goals and concerns such as shared consumption and identity management. We present relevant theories and research questions for a more \emph{network-centric} approach to recommendations and introduce PopCore, a platform for studying them in Facebook. An initial 50-user study with PopCore gives insights into tradeoffs around the popularity, likeability, and rateability of recommendations made by a set of network-centric algorithms and to people's thoughts about the idea of network-centric recommendation.
-
ReComp: QoS-aware recursive service composition at minimum cost.
Vimmi Jaiswal, Amit Sharma and Akshat Verma
IEEE IM, May 2011 [PDF]
