8-6 テレビ番組推薦のためのメタデータ付きグラフマイニング(第8部門 メディア工学3)

抄録

We investigate how the sparseness conditions affect prediction performance of a modified Graph mining method, Random walk with Restart, enhanced by a set of metadeta nodes extracted from EPGs of each program. Experimental results show that the Precision at rank k (P@k) of conventional collaborative filtering gradually falls as we thin-out a part of given dataset of user-item ratings, while that of the original RWR and the modified RWR keep relatively high performance.