Input online review data and related bias in recommender systems

A majority of extant literature on recommender systems assume the input data as a given to generate recommendations. Both implicit and/or explicit data are used as input in these systems. The existence of various challenges in using such input data including those associated with strategic source manipulations, sparse matrix, state data, among others, are sometimes acknowledged. While such input data are also known to be rife with various forms of bias, to our knowledge no explicit attempt is made to correct or compensate for them in recommender systems. We consider a specific type of bias that is introduced in online product reviews due to the sequence in which these reviews are written. We model several scenarios in this context and study their properties.

► We consider how sequential bias are introduced in online review data. ► We study sequential bias in input data used for recommender systems. ► We study the dynamic of sequential bias on later online reviews.