Conversion rate (CVR) prediction is one of the most critical tasks for digital display advertising. Commercial systems often require to update models in an online learning manner to catch up with the evolving data distribution. However, conversions usually do not happen immediately after user clicks. This may result in inaccurate labeling, which is called delayed feedback problem. In previous studies, delayed feedback problem is handled either by waiting positive label for a long period of time, or by consuming the negative sample on its arrival and then insert a positive duplicate when conversion happens later. Indeed, there is a trade-off between waiting for more accurate labels and utilizing fresh data, which is not considered in existing works. To strike a balance in this trade-off, we propose Elapsed-Time Sampling Delayed Feedback Model (ES-DFM), which models the relationship between the observed conversion distribution and the true conversion distribution. Then we optimize the expectation of true conversion distribution via importance sampling under the elapsed-time sampling distribution. We further estimate the importance weight for each instance, which is used as the weight of loss function in CVR prediction. To demonstrate the effectiveness of ES-DFM, we conduct extensive experiments on a public data and a private industrial dataset. Experimental results confirm that our method consistently outperforms the previous state-of-the-art results.
Pooling layers are crucial components for efficient deep representation learning. As to graph data, however, it's not trivial to decide which nodes to retain in order to represent the high-level structure of a graph. Recently many different graph pooling methods have been proposed. However, they all rely on local features to conduct global pooling over all nodes, which contradicts poolings in CNNs that only use local features to conduct local pooling. We analyze why this may hinder the performance of graph pooling, then propose a novel graph pooling method called Bottom-Up and Top-Down graph POOLing (BUTDPool). BUTDPool aims to learn a more fine-grained pooling criterion based on coarse global structure information produced by a bottom-up pooling layer, and can enhance local features with global features. Specifically, we propose to use one or multiple pooling layers with a relatively high retain ratio to produce a coarse high-level graph. Injecting the high-level information back into low-level representation, BUTDPool enhances learning a better pooling criterion. Experiments demonstrate the superior performance of the proposed method over compared methods.
Transfer learning, which aims to reuse knowledge in different domains, has achieved great success in many scenarios via minimizing domain discrepancy and enhancing feature discriminability. However, there are seldom practical determination methods for measuring the transferability among domains. In this paper, we bring forward a novel meta-transfer feature method (MetaTrans) for this problem. MetaTrans is used to train a model to predict performance improvement ratio from historical transfer learning experiences, and can consider both the Transferability between tasks and the Discriminability emphasized on targets. We apply this method to both shallow and deep transfer learning algorithms, providing a detail explanation for the success of specific transfer learning algorithms. From experimental studies, we find that different transfer learning algorithms have varying dominant factor deciding their success, so we propose a multi-task learning framework which can learn both common and specific experience from historical transfer learning results. The empirical investigations reveal that the knowledge obtained from historical experience can facilitate future transfer learning tasks.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.