Long-Tail Session-based Recommendation from Calibration

Accurate prediction in session-based recommendation has achieved progress, but skewed recommendation list caused by popularity bias is rarely investigated. Existing models on mitigating the popularity bias try to reduce the over-concentration on popular items but ignore the users' different preferences towards tail items. To this end, we incorporate calibration to mitigate the popularity bias in session-based recommendation. We propose a calibration module that can predict the distribution of the recommendation list and calibrate the recommendation list to the ongoing session. Meanwhile, a separate training and prediction strategy is applied to deal with the imbalance problem. Experiments on benchmark datasets show that our model can both achieve the competitive accuracy of recommendation and provide more tail items.