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 actually cause over and above activity that would have occurred via other means (e.g., search) if recommendations were absent. Although the ideal way to estimate the causal impact of recommendations is via randomized experiments, such experiments are costly and may inconvenience users. I… Show more
“…The example on the left shows a focal product that receives a large and sudden shock in page visits, while direct visits to its recommended product remains relatively flat. This is reminiscent of the examples analyzed in Carmi, Oestreicher-Singer and Sundararajan [2012] and Sharma, Hofman and Watts [2015]. The example on the right, however, shows more general patterns that are accepted under the split-door criterion but not considered by these previous approaches: although direct visits to both the focal and recommended products vary substantially, they do so independently, and so are still useful in our estimate of the recommender's effect.…”
Section: 4mentioning
confidence: 77%
“…Connections to other methods. The split-door criterion is an example of methods that use empirical independence tests to identify causal effects under certain assumptions [Jensen et al, 2008;Cattaneo, Frandsen and Titiunik, 2015;Sharma, Hofman and Watts, 2015;Grosse-Wentrup et al, 2016]. By searching for subsets of the data where desired independence holds, it also shares some properties with natural experiment methods such as instrumental variables and conditioning methods such as regression.…”
Section: Requiresmentioning
confidence: 99%
“…Because the vast majority of this data is generated in non-experimental settings, researchers typically must deal with the possibility that any causal effects of interest are complicated by a number of potential confounds. For example, even effects as conceptually simple as the causal impact of recommendations on customer purchases are likely confounded by selection effects [Lewis, Rao and Reiley, 2011], correlated demand [Sharma, Hofman and Watts, 2015], or other shared causes of both exposure and purchase. Figure 1a shows this canonical class of causal inference problems in the form of a causal graphical model [Pearl, 2009], where X is the cause and Y is its effect.…”
We present a method for estimating causal effects in time series data when fine-grained information about the outcome of interest is available. Specifically, we examine what we call the split-door setting, where the outcome variable can be split into two parts: one that is potentially affected by the cause being studied and another that is independent of it, with both parts sharing the same (unobserved) confounders. We show that under these conditions, the problem of identification reduces to that of testing for independence among observed variables, and present a method that uses this approach to automatically find subsets of the data that are causally identified. We demonstrate the method by estimating the causal impact of Amazon's recommender system on traffic to product pages, finding thousands of examples within the dataset that satisfy the split-door criterion. Unlike past studies based on natural experiments that were limited to a single product category, our method applies to a large and representative sample of products viewed on the site. In line with previous work, we find that the widely-used click-through rate (CTR) metric overestimates the causal impact of recommender systems; depending on the product category, we estimate that 50-80% of the traffic attributed to recommender systems would have happened even without any recommendations. We conclude with guidelines for using the split-door criterion as well as a discussion of other contexts where the method can be applied. * We would like to thank Dean Eckles, Praneeth Netrapalli, Joshua Angrist, T. Tony Ke, and anonymous reviewers for their valuable feedback on this work.
“…The example on the left shows a focal product that receives a large and sudden shock in page visits, while direct visits to its recommended product remains relatively flat. This is reminiscent of the examples analyzed in Carmi, Oestreicher-Singer and Sundararajan [2012] and Sharma, Hofman and Watts [2015]. The example on the right, however, shows more general patterns that are accepted under the split-door criterion but not considered by these previous approaches: although direct visits to both the focal and recommended products vary substantially, they do so independently, and so are still useful in our estimate of the recommender's effect.…”
Section: 4mentioning
confidence: 77%
“…Connections to other methods. The split-door criterion is an example of methods that use empirical independence tests to identify causal effects under certain assumptions [Jensen et al, 2008;Cattaneo, Frandsen and Titiunik, 2015;Sharma, Hofman and Watts, 2015;Grosse-Wentrup et al, 2016]. By searching for subsets of the data where desired independence holds, it also shares some properties with natural experiment methods such as instrumental variables and conditioning methods such as regression.…”
Section: Requiresmentioning
confidence: 99%
“…Because the vast majority of this data is generated in non-experimental settings, researchers typically must deal with the possibility that any causal effects of interest are complicated by a number of potential confounds. For example, even effects as conceptually simple as the causal impact of recommendations on customer purchases are likely confounded by selection effects [Lewis, Rao and Reiley, 2011], correlated demand [Sharma, Hofman and Watts, 2015], or other shared causes of both exposure and purchase. Figure 1a shows this canonical class of causal inference problems in the form of a causal graphical model [Pearl, 2009], where X is the cause and Y is its effect.…”
We present a method for estimating causal effects in time series data when fine-grained information about the outcome of interest is available. Specifically, we examine what we call the split-door setting, where the outcome variable can be split into two parts: one that is potentially affected by the cause being studied and another that is independent of it, with both parts sharing the same (unobserved) confounders. We show that under these conditions, the problem of identification reduces to that of testing for independence among observed variables, and present a method that uses this approach to automatically find subsets of the data that are causally identified. We demonstrate the method by estimating the causal impact of Amazon's recommender system on traffic to product pages, finding thousands of examples within the dataset that satisfy the split-door criterion. Unlike past studies based on natural experiments that were limited to a single product category, our method applies to a large and representative sample of products viewed on the site. In line with previous work, we find that the widely-used click-through rate (CTR) metric overestimates the causal impact of recommender systems; depending on the product category, we estimate that 50-80% of the traffic attributed to recommender systems would have happened even without any recommendations. We conclude with guidelines for using the split-door criterion as well as a discussion of other contexts where the method can be applied. * We would like to thank Dean Eckles, Praneeth Netrapalli, Joshua Angrist, T. Tony Ke, and anonymous reviewers for their valuable feedback on this work.
“…Liang, et al [40] draw on the language of causal analysis in describing a model of user exposure to items; this is related to distinguishing between user preference and our confidence in an observation [26]. Some work has also been done to understand the causal impact of these systems on behavior by finding natural experiments in observational data [53,55] (approximating expensive controlled experiments [33]), but it is unclear how well these results generalize. Schnabel, et al [52] use propensity weighting techniques to remove users' selection bias for explicit ratings.…”
Recommendation systems are ubiquitous and impact many domains; they have the potential to influence product consumption, individuals' perceptions of the world, and life-altering decisions. These systems are often evaluated or trained with data from users already exposed to algorithmic recommendations; this creates a pernicious feedback loop. Using simulations, we demonstrate how using data confounded in this way homogenizes user behavior without increasing utility.
“…It is harder to assess whether recommendations foster or limit access to diverse types of content. The academic debate about recommendations being the bane or boon of social media is still very lively [42,10,44], with evidence brought in support of the two views. We aim to provide further evidence to shed light on this point in the context of link recommenders.…”
Ego-networks are fundamental structures in social graphs, yet the process of
their evolution is still widely unexplored. In an online context, a key
question is how link recommender systems may skew the growth of these networks,
possibly restraining diversity. To shed light on this matter, we analyze the
complete temporal evolution of 170M ego-networks extracted from Flickr and
Tumblr, comparing links that are created spontaneously with those that have
been algorithmically recommended. We find that the evolution of ego-networks is
bursty, community-driven, and characterized by subsequent phases of explosive
diameter increase, slight shrinking, and stabilization. Recommendations favor
popular and well-connected nodes, limiting the diameter expansion. With a
matching experiment aimed at detecting causal relationships from observational
data, we find that the bias introduced by the recommendations fosters global
diversity in the process of neighbor selection. Last, with two link prediction
experiments, we show how insights from our analysis can be used to improve the
effectiveness of social recommender systems.Comment: Proceedings of the 10th ACM International Conference on Web Search
and Data Mining (WSDM 2017), Cambridge, UK. 10 pages, 16 figures, 1 tabl
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.