skip to main content
Explore Scholarly Publications and Datasets in the NSF-PAR

Title: Estimating Error and Bias in Offline Evaluation Results
Offline evaluation protocols for recommender systems are intended to estimate users' satisfaction with recommendations using static data from prior user interactions. These evaluations allow researchers and production developers to carry out first-pass estimates of the likely performance of a new system and weed out bad ideas before presenting them to users. However, offline evaluations cannot accurately assess novel, relevant recommendations, because the most novel recommendations items that were previously unknown to the user; such items are missing from the historical data, so they cannot be judged as relevant. A breakthrough that reliably produces novel, relevant recommendations would score poorly with current offline evaluation techniques. While the existence of this problem is noted in the literature, its extent is not well-understood. We present a simulation study to estimate the error that such missing data causes in commonly-used evaluation metrics in order to assess its prevalence and impact. We find that missing data in the rating or observation process causes the evaluation protocol to systematically mis-estimate metric values, and in some cases erroneously determine that a popularity-based recommender outperforms even a perfect personalized recommender. Substantial breakthroughs in recommendation quality, therefore, will be difficult to assess with existing offline techniques.
Authors:
;
Award ID(s):
1751278
Publication Date:
NSF-PAR ID:
10146883
Journal Name:
Proceedings of the 2020 Conference on Computer-Human Interaction and Information Retrieval
Page Range or eLocation-ID:
392 - 396
Sponsoring Org:
National Science Foundation
More Like this
  1. ; ( , ArXivorg)
    Today’s recommender systems are criticized for recommending items that are too obvious to arouse users’ interest. That is why the recommender systems research community has advocated some ”beyond accuracy” evaluation metrics such as novelty, diversity, coverage, and serendipity with the hope of promoting information discovery and sustain users’ interest over a long period of time. While bringing in new perspectives, most of these evaluation metrics have not considered individual users’ difference: an open-minded user may favor highly novel or diversified recommendations whereas a conservative user’s appetite for novelty or diversity may not be that large. In this paper, we developedmore »a model to approximate an individual’s curiosity distribution over different levels of stimuli guided by the well-known Wundt curve in Psychology. We measured an item’s surprise level to assess the stimulation level and whether it is in the range of the user’s appetite for stimulus. We then proposed a recommendation system framework that considers both user preference and appetite for stimulus where the curiosity is maximally aroused. Our framework differs from a typical recommender system in that it leverages human’s curiosity to promote intrinsic interest with the system. A series of evaluation experiments have been conducted to show that our framework is able to rank higher the items with not only high ratings but also high response likelihood. The recommendation list generated by our algorithm has higher potential of inspiring user curiosity compared to traditional approaches. The personalization factor for assessing the stimulus (surprise) strength further helps the recommender achieve smaller (better) inter-user similarity.« less
  2. ; ( , REVEAL 2018 Workshop on Offline Evaluation in Recommender Systems)
    Traditional offline evaluations of recommender systems apply metrics from machine learning and information retrieval in settings where their underlying assumptions no longer hold. This results in significant error and bias in measures of top-N recommendation performance, such as precision, recall, and nDCG. Several of the specific causes of these errors, including popularity bias and misclassified decoy items, are well-explored in the existing literature. In this paper we survey a range of work on identifying and addressing these problems, and report on our work in progress to simulate the recommender data generation and evaluation processes to quantify the extent of evaluationmore »metric errors and assess their sensitivity to various assumptions.« less
  3. ( , Proceedings of the ACM Conference on Recommender Systems (ACM RecSys))
    Recent work in recommender systems has emphasized the importance of fairness, with a particular interest in bias and transparency, in addition to predictive accuracy. In this paper, we focus on the state of the art pairwise ranking model, Bayesian Personalized Ranking (BPR), which has previously been found to outperform pointwise models in predictive accuracy, while also being able to handle implicit feedback. Specifically, we address two limitations of BPR: (1) BPR is a black box model that does not explain its outputs, thus limiting the user's trust in the recommendations, and the analyst's ability to scrutinize a model's outputs; andmore »(2) BPR is vulnerable to exposure bias due to the data being Missing Not At Random (MNAR). This exposure bias usually translates into an unfairness against the least popular items because they risk being under-exposed by the recommender system. In this work, we first propose a novel explainable loss function and a corresponding Matrix Factorization-based model called Explainable Bayesian Personalized Ranking (EBPR) that generates recommendations along with item-based explanations. Then, we theoretically quantify additional exposure bias resulting from the explainability, and use it as a basis to propose an unbiased estimator for the ideal EBPR loss. The result is a ranking model that aptly captures both debiased and explainable user preferences. Finally, we perform an empirical study on three real-world datasets that demonstrate the advantages of our proposed models.« less
  4. ; ; ( , Proceedings of the 38th IEEE International Conference on Data Engineering (ICDE))
    Context has been recognized as an important factor to consider in personalized recommender systems. Particularly in location-based services (LBSs), a fundamental task is to recommend to a mobile user where he/she could be interested to visit next at the right time. Additionally, location-based social networks (LBSNs) allow users to share location-embedded information with friends who often co-occur in the same or nearby points-of-interest (POIs) or share similar POI visiting histories, due to the social homophily theory and Tobler’s first law of geography. So, both the time information and LBSN friendship relations should be utilized for POI recommendation. Tensor completion hasmore »recently gained some attention in time-aware recommender systems. The problem decomposes a user-item-time tensor into low-rank embedding matrices of users, items and times using its observed entries, so that the underlying low-rank subspace structure can be tracked to fill the missing entries for time-aware recommendation. However, these tensor completion methods ignore the social-spatial context information available in LBSNs, which is important for POI recommendation since people tend to share their preferences with their friends, and near things are more related than distant things. In this paper, we utilize the side information of social networks and POI locations to enhance the tensor completion model paradigm for more effective time-aware POI recommendation. Specifically, we propose a regularization loss head based on a novel social Hausdorff distance function to optimize the reconstructed tensor. We also quantify the popularity of different POIs with location entropy to prevent very popular POIs from being over-represented hence suppressing the appearance of other more diverse POIs. To address the sensitivity of negative sampling, we train the model on the whole data by treating all unlabeled entries in the observed tensor as negative, and rewriting the loss function in a smart way to reduce the computational cost. Through extensive experiments on real datasets, we demonstrate the superiority of our model over state-of-the-art tensor completion methods.« less
  5. ; ; ( , 2021 IEEE International Conference on Big Data (Big Data))
    Matrix completion is a well-known approach for recommender systems. It predicts the values of the missing entries in a sparse user-item interaction matrix, based on the low-rank structure of the rating matrix. However, existing matrix completion methods do not take node polysemy and side information of social relationships into consideration, which can otherwise further improve the performance. In this paper, we propose a novel matrix completion method that employs both users’ friendships and rating entries to predict the missing values in a user-item matrix. Our approach adopts a graph-based modeling where nodes are users and items, and two types ofmore »edges are considered: user friendships and user-item interactions. Polysemy-aware node features are extracted from this heterogeneous graph through a graph convolution network by considering the multifaceted factors for edge formation, which are then connected to a hybrid loss function with two heads: (1) a social-homophily head to address node polysemy, and (2) an error head for user-item rating regression. The latter is formulated on all matrix entries to combat the sensitivity of negative sampling of the vast majority of missing entries during training, with a smart technique to reduce the time complexity. Extensive experiments over real datasets verify that our model outperforms the state-of-the-art matrix completion methods by a significant margin.« less
  • Citation Formats
  • MLA
  • APA
  • Chicago
  • BibTeX
  • Save / Share this Record
  • Send to Email