Click-through rate (CTR) prediction is a critical task in online display advertising. The data involved in CTR prediction are typically multi-field categorical data, i.e., every feature is categorical and belongs to one and only one field. One of the interesting characteristics of such data is that features from one field often interact differently with features from different other fields. Recently, Fieldaware Factorization Machines (FFMs) have been among the best performing models for CTR prediction by explicitly modeling such difference. However, the number of parameters in FFMs is in the order of feature number times field number, which is unacceptable in the real-world production systems. In this paper, we propose Field-weighted Factorization Machines (FwFMs) to model the different feature interactions between different fields in a much more memory-efficient way. Our experimental evaluations show that FwFMs can achieve competitive prediction performance with only as few as 4% parameters of FFMs. When using the same number of parameters, FwFMs can bring 0.92% and 0.47% AUC lift over FFMs on two real CTR prediction data sets.

Real-time bidding (RTB) is an important mechanism in online display advertising, where a proper bid for each page view plays an essential role for good marketing results. Budget constrained bidding is a typical scenario in RTB where the advertisers hope to maximize the total value of the winning impressions under a pre-set budget constraint. However, the optimal bidding strategy is hard to be derived due to the complexity and volatility of the auction environment. To address these challenges, in this paper, we formulate budget constrained bidding as a Markov Decision Process and propose a model-free reinforcement learning framework to resolve the optimization problem. Our analysis shows that the immediate reward from environment is misleading under a critical resource constraint. Therefore, we innovate a reward function design methodology for the reinforcement learning problems with constraints. Based on the new reward design, we employ a deep neural network to learn the appropriate reward so that the optimal policy can be learned effectively. Different from the prior model-based work, which suffers from the scalability problem, our framework is easy to be deployed in large-scale industrial applications. The experimental evaluations demonstrate the effectiveness of our framework on large-scale real datasets.

Abstract. We conducted long-term network observations using standardized Multi-Axis Differential optical absorption spectroscopy (MAX-DOAS) instruments in Russia and ASia (MADRAS) from 2007 onwards and made the first synthetic data analysis. At seven locations (Cape Hedo, Fukue and Yokosuka in Japan, Hefei in China, Gwangju in Korea, and Tomsk and Zvenigorod in Russia) with different levels of pollution, we obtained 80 927 retrievals of tropospheric NO2 vertical column density (TropoNO2VCD) and aerosol optical depth (AOD). In the technique, the optimal estimation of the TropoNO2VCD and its profile was performed using aerosol information derived from O4 absorbances simultaneously observed at 460–490 nm. This large data set was used to analyze NO2 climatology systematically, including temporal variations from the seasonal to the diurnal scale. The results were compared with Ozone Monitoring Instrument (OMI) satellite observations and global model simulations. Two NO2 retrievals of OMI satellite data (NASA ver. 2.1 and Dutch OMI NO2 (DOMINO) ver. 2.0) generally showed close correlations with those derived from MAX-DOAS observations, but had low biases of up to ~50%. The bias was distinct when NO2 was abundantly present near the surface and when the AOD was high, suggesting a possibility of incomplete accounting of NO2 near the surface under relatively high aerosol conditions for the satellite observations. Except for constant biases, the satellite observations showed nearly perfect seasonal agreement with MAX-DOAS observations, suggesting that the analysis of seasonal features of the satellite data were robust. Weekend reduction in the TropoNO2VCD found at Yokosuka and Gwangju was absent at Hefei, implying that the major sources had different weekly variation patterns. While the TropoNO2VCD generally decreased during the midday hours, it increased exceptionally at urban/suburban locations (Yokosuka, Gwangju, and Hefei) during winter. A global chemical transport model, MIROC-ESM-CHEM (Model for Interdisciplinary Research on Climate–Earth System Model–Chemistry), was validated for the first time with respect to background NO2 column densities during summer at Cape Hedo and Fukue in the clean marine atmosphere.

No abstract

Abstract. In October 2017, the Sentinel-5 Precursor (S5P) mission was launched, carrying the TROPOspheric Monitoring Instrument (TROPOMI), which provides a daily global coverage at a spatial resolution as high as 7 km × 3.5 km and is expected to extend the European atmospheric composition record initiated with GOME/ERS-2 in 1995, enhancing our scientific knowledge of atmospheric processes with its unprecedented spatial resolution. Due to the ongoing need to understand and monitor the recovery of the ozone layer, as well as the evolution of tropospheric pollution, total ozone remains one of the leading species of interest during this mission. In this work, the TROPOMI near real time (NRTI) and offline (OFFL) total ozone column (TOC) products are presented and compared to daily ground-based quality-assured Brewer and Dobson TOC measurements deposited in the World Ozone and Ultraviolet Radiation Data Centre (WOUDC). Additional comparisons to individual Brewer measurements from the Canadian Brewer Network and the European Brewer Network (Eubrewnet) are performed. Furthermore, twilight zenith-sky measurements obtained with ZSL-DOAS (Zenith Scattered Light Differential Optical Absorption Spectroscopy) instruments, which form part of the SAOZ network (Système d'Analyse par Observation Zénitale), are used for the validation. The quality of the TROPOMI TOC data is evaluated in terms of the influence of location, solar zenith angle, viewing angle, season, effective temperature, surface albedo and clouds. For this purpose, globally distributed ground-based measurements have been utilized as the background truth. The overall statistical analysis of the global comparison shows that the mean bias and the mean standard deviation of the percentage difference between TROPOMI and ground-based TOC is within 0 –1.5 % and 2.5 %–4.5 %, respectively. The mean bias that results from the comparisons is well within the S5P product requirements, while the mean standard deviation is very close to those limits, especially considering that the statistics shown here originate both from the satellite and the ground-based measurements. Additionally, the TROPOMI OFFL and NRTI products are evaluated against already known spaceborne sensors, namely, the Ozone Mapping Profiler Suite, on board the Suomi National Polar-orbiting Partnership (OMPS/Suomi-NPP), NASA v2 TOCs, and the Global Ozone Monitoring Experiment 2 (GOME-2), on board the Metop-A (GOME-2/Metop-A) and Metop-B (GOME-2/Metop-B) satellites. This analysis shows a very good agreement for both TROPOMI products with well-established instruments, with the absolute differences in mean bias and mean standard deviation being below +0.7 % and 1 %, respectively. These results assure the scientific community of the good quality of the TROPOMI TOC products during its first year of operation and enhance the already prevalent expectation that TROPOMI/S5P will play a very significant role in the continuity of ozone monitoring from space.