Deep feature fusion‐based stacked denoising autoencoder for tag recommendation systems

Abstract With the rapid development of artificial intelligence technology, commercial robots have gradually entered our daily lives. In order to promote product dissemination, shopping guide are a new service options commerce platforms that use tag recommendation systems identify users' intentions. A large number applications combine user historical tagging information with multi‐round dialogue ability help users efficiently search for and retrieve products interest. Recently, tensor decomposition methods become common approach modelling entity interaction relationships in systems. However, due sparsity data, these only consider low‐order entities, making it difficult capture higher‐order collaborative signals among entities. Recommendation by autoencoders can effectively extract abstract feature representations while they focus on two‐dimensional relationship between items, ignoring users, items tags real complex scenarios. The authors similarity entities propose method called deep fusion (DFFT) based stacked denoising autoencoders. This high‐order different embedding dimensions fuse them unified framework. To robust representations, inject random noise (mask‐out/drop‐out noise) into corresponding generate corrupted input then utilise encode further obtain dimensions, encoding layers combined produce better expanded model which reinforce each other. Finally, decoding component is used reconstruct original data. According experimental results two datasets, proposed DFFT outperforms other baselines terms F1@N, NDCG@N Recall@N evaluation metrics.