A new collaborative filtering-based recommender system for manufacturing appstore: Which applications would be useful to your business?

Abstract

In this work, a recommender system is proposed for a manufacturing appstore which is designed and built to revitalize online application trades among application developers and small size manufacturing companies. The aim of the recommender system is to create and provide each website user an effective application recommendation list. The list for a user might include items which are not bought by the user but useful. To build the recommendation list the proposed system makes a list of users having similar purchasing pattern to the given user. To construct the user list every user is represented by a k-dimensional vector of categories which are predetermined according to industry and business area. Based on the vectors user similarities are calculated for every pair of users. With the user list the system figures out recommendation candidate items which are purchased by users in the list but by the target user. To rank items in the candidate list an item similarity metric is utilized. The metric for a given item implies how close the item is to the applications which the target user purchased. Finally, candidate items are ranked by this metric and first r items are recommended to the target user. To demonstrate the effectiveness of the proposed algorithm the proposed system is applied the manufacturing appstore (www.mfg-app.co.kr) and a numerical analysis has conducted with real data from the appstore. © Springer International Publishing Switzerland 2013.