A Time-Window Approach to Recommending Emerging and On-the-rise Items

The recommendation system (RS) filters large sales transaction data, to promote item Y as an alternative or pair to item X that the application user is looking for. Items that are no longer in season usually decrease in transactions, even to zero. Constantly recommending these items is irrelevant, while other items that are more relevant and profitable are not recommended, such as the emerging and on-the-rise items. The proposed solution to this problem is the time window approach, which is a block of data of a certain size and recorded at a certain time stamp. Item combinations (itemsets) are mined from each window with an association rule approach, and changes in the number of transactions containing these items are evaluated from window to window. From the number of transactions that contain the itemsets, the system distinguishes status items into four types: risky, normal, emerging, and on-the-rise. Items that are currently suffering from zero sales risk the same fate in the next window, so they are called risky items. The system will notify admins to review this item, and look for other emerging or on-the-rise items to promote along with item X. Experiments were carried out on two datasets, and it was found that 42.7% and 59.4% items from each dataset are risky.