Modeling User Exposure in Recommendation

By Dawen Liang, Laurent Charlin, James McInerney, David M. Blei. 2015

426 citations recorded by the source corpus.

Collaborative filtering analyzes user preferences for items (e.g., books, movies, restaurants, academic papers) by exploiting the similarity patterns across users. In implicit feedback settings, all the items, including the ones that a user did not consume, are taken into consideration. But this assumption does not accord with the common sense understanding that users have a limited scope and awareness of items. For example, a user might not have heard of a certain paper, or might live too far away from a restaurant to experience it. In the language of causal analysis (Imbens & Rubin, 2015), the assignment mechanism (i.e., the items that a user is exposed to) is a latent variable that may change for various user/item combinations. In this paper, we propose a new probabilistic approach that directly incorporates user exposure to items into collaborative filtering. The exposure is modeled as a latent variable and the model infers its value from data. In doing so, we recover one of the most successful state-of-the-art approaches as a special case of our model (Hu et al. 2008), and provide a plug-in method for conditioning exposure on various forms of exposure covariates (e.g., topics in text, venue locations). We show that our scalable inference algorithm outperforms existing benchmarks in four different domains both with and without exposure covariates.

Subfields: Computer Science Applications, Information Systems, Management Science and Operations Research

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Science Applications

In Information Systems

In Management Science and Operations Research

Citations in the corpus

Neural Collaborative Filtering (2017, background): “Recently, He et al. [14] and Liang et al. [23] proposed dedicated models to weight missing data, and Rendle et al. [1] developed an implicit coordinate descent (iCD) solution for feature-based factorization models, achieving state-of-the-art performance for item recommendation.”
How algorithmic confounding in recommendation systems increases homogeneity and decreases utility (2017, background): “[40] draw on the language of causal analysis in describing a model of user exposure to items; this is related to distinguishing”
Recommendations as Treatments: Debiasing Learning and Evaluation (2016, background): “Related but different from the problem we consider is recommendation from positive feedback alone (Hu et al., 2008; Liang et al., 2016).”
Hierarchical Gating Networks for Sequential Recommendation (2019, background): “Recently, several works [11, 22] are proposed to weigh the missing data.”
Model-Agnostic Counterfactual Reasoning for Eliminating Popularity Bias in Recommender System (2020, background): “• ExpoMF [31]: A probabilistic model that separately estimates the user preferences and the exposure.”
Adversarial Personalized Ranking for Recommendation (2018, background): “This is the check-in dataset constructed by [23] for item recommendation.”
Outer Product-based Neural Collaborative Filtering (2018, methodology): “Gowalla This is the check-in dataset from Gowalla, a location-based social network, constructed by [Liang et al., 2016] for item recommendation.”
Dual Graph Attention Networks for Deep Latent Representation of Multifaceted Social Effects in Recommender Systems (2019, background): “Previous studies for social recommendation attempt to model social effects in various ways, such as by trust propagation [5, 12, 22], regularization loss [20, 33, 40], matrix factorization [8, 18, 38], network embedding [17, 34], causal inference [16, 32], and graph neural network [36, 41].”
Bias and Debias in Recommender System: A Survey and Future Directions (2020, background): “Chen et al. [31, 32] give detailed analyses of EXMF and rewrite the objective function of EXMF as follows: where 𝛾 𝑢𝑖 is defined as variational parameters of the user’s exposure.”
Revisiting Graph based Collaborative Filtering: A Linear Residual Graph Convolutional Network Approach (2020, methodology): “The runtime time on the Gowalla dataset for each model is about one third of the time compared to the time cost of the Amazon Books, and the overall trend of the time comparison is similar as analyzed above.”
Causal embeddings for recommendation (2017, background): “In (Liang et al. 2016) the authors present ExpoMF, a probabilistic approach for collaborative ﬁltering on implicit data.”
Neural Graph Collaborative Filtering (2019, background): “Gowalla: This is the check-in dataset [21] obtained from Gowalla, where users share their locations by checking-in.”
Toward Implicit Sample Noise Modeling: Deviation-driven Matrix Factorization (2016, background): “(Hu, Koren, and Volinsky 2008) and (Liang et al. 2016) are designed for an implicit feedback scenario by a weighting Robust MF is a family of MF models addressing noise modeling, but lots of works are designed for full matrix factorization rather than CF.”
Causal Intervention for Leveraging Popularity Bias in Recommendation (2021, background): “Some other works have attempted to remove the impact of other bias problems [11], position bias [5], selection bias [43], and exposure bias [12, 26] etc.”