Search

Conventional recommender systems (RSs) rely on consumers' feedback like product ratings to elicit parameters for personalized recommendations. Such an approach suffers severely from the biases caused by consumers' self-selection behaviors. RSs fed with biased input may reinforce the biases and result in biased models that are incapable to effectively predict consumer preferences. By examining the holistic process of consumer purchase and rating, three types of self-selection biases, i.e., the exposure, acquisition, and under-report biases, are considered in this paper. To mitigate these biases in training RSs, we propose a generative modeling approach that jointly incorporates consumer behavioral patterns in the exposure, purchase, and rating stages. To rigorously evaluate the performance of the proposed approach, two bias-free datasets are used as testbeds. The experimental results show that the proposed approach outperforms state-of-the-art methods. Our research contributes to the literature and practice of RSs by providing an innovative debiasing approach to dealing with biased input. [ABSTRACT FROM AUTHOR]

By using propensity score matching method of random sample, we matched simple random sample units and volunteer panel Web survey sample units based on the equal or similar propensity score. The unbiased estimators of the population parameters are constructed by using the matching simple random sample, and the self-selection bias is estimated. We propose propensity score weighted and matching sample post stratification weighted methods to estimate the population parameters, and the self-selection bias in volunteer panel Web Surveys are adjusted. [ABSTRACT FROM AUTHOR]