Search

Online advertisement is an important source of revenue for internet companies, so increasing click-through rates (CTR) on ads is crucial. the traditional CTR prediction model only extracts the classification features and numerical features of the advertisement, but ignores the title, description and other text features of the advertisement. However, we know that such information is very important for an advertisement. Therefore, we propose a deep multimodal network (DMN) to solve this problem. on the basis of the traditional deep model, DMN add the text features of cyclic neural network learning, so as to improve the performance of the model. we also did a lot of visual data analysis on the dataset, Finally, we conducted a comparative experiment on the real-world dataset.

The post-click conversion rate (CVR) prediction is important for the sponsored search engine. The predicted CVR can guide people on how to choose the advertisements to display, and greatly affects the revenue of the electronic business platform and merchants who pay for the advertisement. In this paper, we describe a new approach to solve the CVR prediction problem in sponsored search. The approach is Boosted Trees Model with Feature Engineering (BTMFE), which incorporates the power of feature engineering as well as the boosted trees. We discuss different issues such as feature importance, feature interaction, historical information fusion and boosted trees algorithm. Finally, we compare BTMFE with other methods and present experimental results on real-world datasets.

© 2017 IOP Publishing Ltd. This paper reports an investigation on a semi-active/passive integrated vibration isolator utilizing a magnetorheological elastomer (MRE) and spring. To overcome the main shortcoming of passive isolation systems, i.e. lack of adaptability, the semi-active/passive integrated isolator (SAPII) based on an MRE and spring is designed and prototyped. The magnetic circuit is optimized by finite element analysis to fully unlock the unique features of the MRE. The dynamic response characteristic of the SAPII is experimentally investigated under a sweep frequency test. A dynamic model of the SAPII vibration isolation system is established on the basis of the Kelvin model. The model parameters, such as equivalent stiffness and equivalent damping, are identified from experimental data. An ON-OFF control law based on the minimal displacement transmissibility is designed for isolation control of the sinusoid excitation. Two control laws, i.e. ON-OFF control and fuzzy logic control, are designed for vibration isolation of random excitation. Finally, the effectiveness of these control laws is verified by numerical simulation and experiment.