Binary butterfly optimization approach (bBOA) is a recent high performing feature selection algorithm presented in 2018 which is based on the food foraging behavior of butterflies. This paper tries to improve the structure of the bBOA to enhance its classification accuracy, dimension reduction and reliability in feature selection task for who are interested in the fields of data mining and pattern recognition. The new initialization strategy and differential evolution strategy are applied to reduce the randomness of bBOA's initialization and local search process. Then, a new parameter is added to make the bBOA's transfer function more adaptive to the change of exploration and exploitation. Besides, evolution population dynamics (EPD) mechanism is employed as an extension of bBOA. The new method called optimization and extension of binary butterfly optimization approaches (OEbBOA) is tested with the K nearest neighbor classier in which twenty UCI datasets and seven recent algorithms are utilized to assess the performance of the OEbBOA algorithm. The experimental results and nonparametric Wilcoxons rank sum test confirm the efficiency of the proposed OEbBOA in maximizing classification accuracy while minimizing the number of features selected.

No abstract

Conversion rate (CVR) prediction is becoming increasingly important in the multi-billion dollar online display advertising industry. It has two major challenges: firstly, the scarce user history data is very complicated and non-linear; secondly, the time delay between the clicks and the corresponding conversions can be very large, e.g., ranging from seconds to weeks. Existing models usually suffer from such scarce and delayed conversion behaviors. In this paper, we propose a novel deep learning framework to tackle the two challenges. Specifically, we extract the pre-trained embedding from impressions/clicks to assist in conversion models and propose an inner/self-attention mechanism to capture the fine-grained personalized product purchase interests from the sequential click data. Besides, to overcome the time-delay issue, we calibrate the delay model by learning dynamic hazard function with the abundant post-click data more in line with the real distribution. Empirical experiments with real-world user behavior data prove the effectiveness of the proposed method.

No abstract

No abstract