Many e-commerce websites use recommender systems or personalized rankers to personalize search results based on their previous interactions. However, a large fraction of users has no prior inter-actions, making it impossible to use collaborative filtering or rely on user history for personalization. Even the most active users mayvisit only a few times a year and may have volatile needs or different personas, making their personal history a sparse and noisy signal at best. This paper investigates how, when we cannot rely on the user history, the large scale availability of other user interactions still allows us to build meaningful profiles from the contextual data and whether such contextual profiles are useful to customize the ranking, exemplified by data from a major online travel agentBooking.com.Our main findings are threefold: First, we characterize the Continuous Cold Start Problem(CoCoS) from the viewpoint of typical e-commerce applications. Second, as explicit situational con-text is not available in typical real world applications, implicit cues from transaction logs used at scale can capture essential features of situational context. Third, contextual user profiles can be created offline, resulting in a set of smaller models compared to a single huge non-contextual model, making contextual ranking available with negligible CPU and memory footprint. Finally we conclude that, in an online A/B test on live users, our contextual ranker in-creased user engagement substantially over a non-contextual base-line, with click-through-rate (CTR) increased by 20%. This clearly demonstrates the value of contextual user profiles in a real world application.

Recommendation based on user preferences is a common task for e-commerce websites. New recommendation algorithms are often evaluated by offline comparison to baseline algorithms such as recommending random or the most popular items. Here, we investigate how these algorithms themselves perform and compare to the operational production system in large scale online experiments in a real-world application. Specifically, we focus on recommending travel destinations at Booking.com, a major online travel site, to users searching for their preferred vacation activities. To build ranking models we use multi-criteria rating data provided by previous users after their stay at a destination. We implement three methods and compare them to the current baseline in Booking.com: random, most popular, and Naive Bayes. Our general conclusion is that, in an online A/B test with live users, our Naive-Bayes based ranker increased user engagement significantly over the current online system.

Recommendation based on user preferences is a common task for e-commerce websites. New recommendation algorithms are often evaluated by offline comparison to baseline algorithms such as recommending random or the most popular items. Here, we investigate how these algorithms themselves perform and compare to the operational production system in large scale online experiments in a real-world application. Specifically, we focus on recommending travel destinations at Booking. com, a major online travel site, to users searching for their preferred vacation activities. To build ranking models we use multi-criteria rating data provided by previous users after their stay at a destination. We implement three methods and compare them to the current baseline in Booking. com: random, most popular, and Naive Bayes. Our general conclusion is that, in an online A/B test with live users, our Naive-Bayes based ranker increased user engagement significantly over the current online system.

Sliding window object detection has received remarkable attention in recent years due to great versatility and extraordinary detection performance. However, straightforward applications of the concept fail to meet criteria of real applications due to insufficient precision and inaccurate localization. Localization accuracy is especially important when the detection needs to be followed by recognition. In this paper, we present a detection approach which completely obviates the need for blind spatial clustering of nearby detection responses, which is known as a major factor of localization inaccuracy. The approach has been evaluated on traffic sign detection, where we consider the superclass of triangular warning signs. The obtained results confirm the viability of the approach and provide useful directions for future work.

We present a study on applying ViolaJones detection and SVM classification for recognizing traffic signs in video. Extensive experimentation has shown that this combination suffers from high incidence of false alarms and low tolerance to localization inaccuracy of the true positive detection responses. We report on three improvements which effectively alleviate these problems. Firstly, we confirm the previous result that raw detection performance of Viola-Jones detector can be improved by exploiting color. Additionally, we propose a solution for filtering false positive detection responses, based on a properly trained artificial neural network classifier in the last stage of the detection cascade. Finally, we propose a novel approach for alleviating the degradation of the classification performance due to localization inaccuracy. Experiments have been performed on several video sequences acquired from a moving vehicle, containing several hundred triangular warning signs. The results indicate a dramatic improvement in detection precision, as well as significant improvements in classification performance. At the system level, the proposed system correctly classified more than 97% of triangular warning signs, while producing only a few false alarms in more than 130000 image frames.