The Effect of Feedback Granularity on Recommender Systems Performance

Abstract

The main source of knowledge utilized in recommender systems (RS) is users' feedback. While the usage of implicit feedback (i.e. user's behavior statistics) is gaining in prominence, the explicit feedback (i.e. user's ratings) remain an important data source. This is true especially for domains, where evaluation of an object does not require an extensive usage and users are well motivated to do so (e.g., video-on-demand services or library archives). So far, numerous rating schemes for explicit feedback have been proposed, ranging both in granularity and presentation style. There are several works studying the effect of rating's scale and presentation on user's rating behavior, e.g. willingness to provide feedback or various biases in rating behavior. Nonetheless, the effect of ratings granularity on RS performance remain largely under-researched. In this paper, we studied the combined effect of ratings granularity and supposed probability of feedback existence on various performance statistics of recommender systems. Results indicate that decreasing feedback granularity may lead to changes in RS's performance w.r.t. nDCG for some recommending algorithms. Nonetheless, in most cases the effect of feedback granularity is surpassed by even a small decrease in feedback's quantity. Therefore, our results corroborate the policy of many major real-world applications, i.e. preference of simpler rating schemes with the higher chance of feedback reception instead of finer-grained rating scenarios.