13 September 2016 in IEEE Transactions on Knowledge and Data Engineering
More and more applications have come with large data sets in the past decade. However, existing algorithms cannot guarantee to scale well on large data. Averaged n-Dependence Estimators (AnDE) allows for flexible learning from out-of-core data, by varying the value of n (number of super parents). Hence, AnDE is especially appropriate for large data learning. In this paper, we propose a sample-based attribute selection technique for AnDE. It needs one more pass through the training data, in which a multitude of approximate AnDE models are built and efficiently assessed by leave-one-out cross validation. The use of a sample reduces the training time. Experiments on 15 large data sets demonstrate that the proposed technique significantly reduces AnDE's error at the cost of a modest increase in training time. This efficient and scalable out-of-core approach delivers superior or comparable performance to typical in-core Bayesian network classifiers.
20 February 2025 in ACM Transactions on Multimedia Computing, Communications, and Applications
The automatic recognition of user sentiments through their music listening behavior is an important research task in cognitive studies. Whereas prior studies were conducted to identify the sentiment conveyed (or evoked) by a song that a user listens to at a particular time, we argue that a more effective method would be to identify the user’s induced sentiment based on the comprehensive list of songs they have listened to (e.g., the sequence of music being played). However, recognizing the sentiment information induced by a playlist using machine learning techniques is much more challenging than identifying the sentiment induced by a single song, as it is difficult to obtain accurately labeled training samples for playlists. In this study, we developed the List–Song Relationship Factorization (LSRF) model with the objective of efficiently identifying sentiments induced by playlists. This model employs two side information constraints: the sentiment similarity between songs, based on multimodal information, and the co-occurrence of songs in playlists. These constraints enable the simultaneous co-clustering of songs and playlists. The experimental results demonstrate that the proposed model efficiently and consistently identifies sentiment information evoked by either playlists or individual songs.