Look-alike audience extension is a practically effective way to customize high-performance audience in on-line advertising. With look-alike audience extension system, any advertiser can easily generate a set of customized audience by just providing a list of existing customers without knowing the detailed targetable attributes in a sophisticated advertising system. In this paper, we present our newly developed graph-based look-alike system in Yahoo! advertising platform which provides look-alike audiences for thousands of campaigns. Extensive experiments have been conducted to compare our look-alike model with three other existing look-alike systems using billions of users and millions of user features. The experiment results show that our developed graph-based method with nearest-neighbor filtering outperforms other methods by more than 50% regarding conversion rate in app-install ad campaigns.

Look-alike models, which are efficient tools for finding similar users from a smaller user set, are quickly revolutionizing the online programmatic advertising industry. The datasets in these contexts exhibit extremely sparse feature spaces on a massive scale, so traditionally, the state-of-the-art look-alike models have used pairwise similarities to construct these similar user sets. One of the key challenges of the similarity-based models is that they do not provide a way to measure the potential value of the users to an advertiser, which is crucial in an advertising context. We propose methods to score users within the expanded audience in a way which relates directly to the business metric that the advertiser wants to optimize. We present three scoring models and show that, through empirical evaluation using real-world, large-scale data, by incorporating the potential value of a user to an advertiser into our scoring model, we can significantly improve the performance of the look-alike models over methods which only use pairwise similarities of users.

Online display advertising is a multi-billion dollar industry where advertisers promote their products to users by having publishers display their advertisements on popular Web pages. An important problem in online advertising is how to forecast the number of user visits for a Web page during a particular period of time. Prior research addressed the problem by using traditional time-series forecasting techniques on historical data of user visits; (e.g., via a single regression model built for forecasting based on historical data for all Web pages) and did not fully explore the fact that different types of Web pages and different time stamps have different patterns of user visits. In this paper, we propose a series of probabilistic latent class models to automatically learn the underlying user visit patterns among multiple Web pages and multiple time stamps. The last (and the most effective) proposed model identifies latent groups/classes of (i) Web pages and (ii) time stamps with similar user visit patterns, and learns a specialized forecast model for each latent Web page and time stamp class. Compared with a single regression model as well as several other baselines, the proposed latent class model approach has the capability of differentiating the importance of different types of information across different classes of Web pages and time stamps, and therefore has much better modeling flexibility. An extensive set of experiments along with detailed analysis carried out on real-world data from Yahoo! demonstrates the advantage of the proposed latent class models in forecasting online user visits in online display advertising.

Video surveillance is an alternative approach to staff or self- reporting that has the potential to detect and monitor aggressive behaviors more accurately. In this paper, we propose an automatic algorithm capable of recognizing aggressive behaviors from video records using local binary motion descriptors. The proposed algorithm will increase the accuracy for retrieving aggressive behaviors from video records, and thereby facilitate scientific inquiry into this low frequency but high impact phenomenon that eludes other measurement approaches.

Display advertising is a multi-billion dollar industry where advertisers promote their products to users by having publishers display their advertisements on popular Web pages. An important problem in online advertising is how to forecast the number of user visits for a Web page during a particular period of time. Prior research addressed the problem by using traditional time-series forecasting techniques on historical data of user visits; (e.g., via a single regression model built for forecasting based on historical data for all Web pages) and did not fully explore the fact that different types of Web pages have different patterns of user visits. In this paper we propose a probabilistic latent class model to automatically learn the underlying user visit patterns among multiple Web pages. Experiments carried out on real-world data demonstrate the advantage of using latent classes in forecasting online user visits.

We propose a method for forecasting high-dimensional data (hundreds of attributes, trillions of attribute combinations) for a duration of several months. Our motivating application is guaranteed display advertising, a multi-billion dollar industry, whereby advertisers can buy targeted (high-dimensional) user visits from publishers many months or even years in advance. Forecasting high-dimensional data is challenging because of the many possible attribute combinations that need to be forecast. To address this issue, we propose a method whereby only a sub-set of attribute combinations are explicitly forecast and stored, while the other combinations are dynamically forecast on-the-fly using high-dimensional attribute correlation models. We evaluate various attribute correlation models, from simple models that assume the independence of attributes to more sophisticated sample-based models that fully capture the correlations in a high-dimensional space. Our evaluation using real-world display advertising data sets shows that fully capturing high-dimensional correlations leads to significant forecast accuracy gains. A variant of the proposed method has been implemented in the context of Yahoo!'s guaranteed display advertising system.

Abstract. Knowledge about temporal and spatial variations of the O3 and NOx relationship in the urban environment are necessary to assess the exceedance of air quality standards for NO2. Both reliable measurements and validated high-resolution air quality models are important to assess the effect of traffic emission on air quality. In this study, measurements of NO, NO2 and O3 concentrations were performed in Gothenburg, Sweden, during the Gote-2005 campaign in February 2005. The aim was to evaluate the variation of pollutant concentrations in the urban landscape in relation to urban air quality monitoring stations and wind speed. A brief description of the meteorological conditions and the air pollution situation during the Gote-2005 campaign was also given. Furthermore, the Air Pollution Model (TAPM) was used to simulate the NOx-regime close to an urban traffic route and the simulations were compared to the measurements. Important conclusions were that the pollutant concentrations varied substantially in the urban landscape and the permanent monitoring stations were not fully representative for the most polluted environments. As expected, wind speed strongly influenced measured pollutant concentrations and gradients. Higher wind speeds dilute NO2 due to stronger dispersion; while at the same time vertical transport of O3 is enhanced, which produces NO2 through oxidation of NO. The oxidation effect was predominant at the more polluted sites, while the dilution effect was more important at the less polluted sites. TAPM reproduced the temporal variability in pollutant concentrations satisfactorily, but was not able to resolve the situation at the most polluted site, due to the local scale site-specific conditions.

In this paper, we present a system that automatically translates Arabic text embedded in images into English. The system consists of three components: text detection from images, character recognition, and machine translation. We formulate the text detection as a binary classification problem and apply gradient boosting tree (GBT), support vector machine (SVM), and location-based prior knowledge to improve the F1 score of text detection from 78.95% to 87.05%. The detected text images are processed by off-the-shelf optical character recognition (OCR) software. We employ an error correction model to post-process the noisy OCR output, and apply a bigram language model to reduce word segmentation errors. The translation module is tailored with compact data structure for hand-held devices. The experimental results show substantial improvements in both word recognition accuracy and translation quality. For instance, in the experiment of Arabic transparent font, the BLEU score increases from 18.70 to 33.47 with use of the error correction module.

We present a set of experiments with a video OCR system (VOCR) tailored for video information retrieval and establish its importance in multimedia search in general and for some specific queries in particular. The system, inspired by an existing work on text detection and recognition in images, has been developed using techniques involving detailed analysis of video frames producing candidate text regions. The text regions are then binarized and sent to a commercial OCR resulting in ASCII text, that is finally used to create search indexes. The system is evaluated using the TRECVID data. We compare the system's performance from an information retrieval perspective with another VOCR developed using multi-frame integration and empirically demonstrate that deep analysis on individual video frames result in better video retrieval. We also evaluate the effect of various textual sources on multimedia retrieval by combining the VOCR outputs with automatic speech recognition (ASR) transcripts. For general search queries, the VOCR system coupled with ASR sources outperforms the other system by a very large extent. For search queries that involve named entities, especially people names, the VOCR system even outperforms speech transcripts, demonstrating that source selection for particular query types is extremely essential.