Items where author is affiliated with Yahoo! Research

Motivated by the emergence of auction-based marketplaces for display ads such as the Right Media Exchange, we study the design of a bidding agent that implements a display advertising campaign by bidding in such a marketplace. The bidding agent must acquire a given number of impressions with a given target spend, when the highest external bid in the marketplace is drawn from an unknown distribution P . The quantity and spend constraints arise from the fact that display ads are usually sold on a CPM basis. We consider both the full information setting, where the winning price in each auction is announced publicly, and the partially observable setting where only the winner obtains information about the distribution; these differ in the penalty incurred by the agent while attempting to learn the distribution. We provide algorithms for both settings, and prove performance guarantees using bounds on uniform closeness from statistics, and techniques from online learning. We experimentally evaluate these algorithms: both algorithms perform very well with respect to both target quantity and spend; further, our algorithm for the partially observable case performs nearly as well as that for the fully observable setting despite the higher penalty incurred during learning.

Web search engines use indexes to efficiently retrieve pages containing specified query terms, as well as pages linking to specified pages. The problem of compressed indexes that permit such fast retrieval has a long history. We consider the problem: assuming that the terms in (or links to) a page are generated from a probability distribution, how well compactly can we build such indexes that allow fast retrieval? Of particular interest is the case when the probability distribution is Zipfian (or a similar power law), since these are the distributions that arise on the web. We obtain sharp bounds on the space requirement of Boolean indexes for text documents that follow Zipf’s law. In the process we develop a general technique that applies to any probability distribution, not necessarily a power law; this is the first analysis of compression in indexes under arbitrary distributions. Our bounds lead to quantitative versions of rules of thumb that are folklore in indexing. Our experiments on several document collections show that the distribution of terms appears to follow a double-Pareto law rather than Zipf’s law. Despite widely varying sets of documents, the index sizes observed in the experiments conform well to our theoretical predictions.

Web mining deals with understanding, and discovering information in, the World Wide Web. Web mining focuses on analyzing three different sources of information: web structure, user activity and the contents. When referring to the Web 2.0, web structure and user activity related data can be dealt with in a very similar way that in the case of the traditional Web, however, in the case of contents, conventional analysis and mining procedures are not suitable anymore. This is mainly because, in the Web 2.0, contents are generated by users, who make a very free use of language and are constantly incorporating new communication elements which are generally context dependent. This kind of language can also be found on chats, SMS, e-mails and other channels of informal textual communication. This workshop focuses on the problem of making Web 2.0 both searchable and analyzable in terms of its contents. This is an extremely important endeavor for current web mining technologies because of two reasons: first, user generated content (UGC) is growing faster than ever in the cyberspace and, two, automatic analysis of UGC will allow improving the user experience of common citizens about Internet resources and opportunities, while, simultaneously, detecting and tracking criminal and terrorist activity. In this first edition of the workshop we attempt to focus the attention of interested research groups and companies into the new challenges and opportunities related to Web 2.0 content analysis. More specifically, we will focus on specific tasks on the scope of text content mining, with the intention of extending the coverage to multimedia data in future editions of the workshop. According to this, for the first edition of the workshop, we will collect and provide a corpus which should be used as experimental collection to conduct research in three specific shared tasks: text normalization, opinion mining and misbehavior detection. In the text normalization shared task we want to address the problem related to chat-speak style of communication. Recently, some research has been carried out in this area for SMS communications and from the perspective of machine translation approaches. In this shared task we attempt to generalize the problem to Web 2.0 contents and to explore additional alternatives the participants can come out with. In the opinion mining shared task we want to address problems such as determining text subjectivity and polarity, and sentiment analysis. Although these problems have been already approached from different perspectives, most of the research has been carried out on specific domain data and applications where users are requested to rate services or products. Our intention is to focus the attention into the more general domain in which Web 2.0 users express their sentiments and opinions in their daily interaction within a virtual community. Finally, in the misbehavior detection shared task, we want to address the problems of detecting inappropriate activity in which some users in a virtual community can be molesting or offensive to some other members of the community. We consider that this shared task can provide a good starting point for a future shared task with the more ambitious goal of classifying users and detecting identity supplantation for on-line criminal activity.

Query processing is a major cost factor in operating large web search engines. In this paper, we study query result caching, one of the main techniques used to optimize query processing performance. Our first contribution is a study of result caching as a weighted caching problem. Most previous work has focused on optimizing cache hit ratios, but given that processing costs of queries can vary very significantly we argue that total cost savings also need to be considered. We describe and evaluate several algorithms for weighted result caching, and study the impact of Zipf-based query distributions on result caching. Our second and main contribution is a new set of feature-based cache eviction policies that achieve significant improvements over all previous methods, substantially narrowing the existing performance gap to the theoretically optimal (clairvoyant) method. Finally, using the same approach, we also obtain performance gains for the related problem of inverted list caching.

Web search engines use highly optimized compression schemes to decrease inverted index size and improve query through- put, and many index compression techniques have been stud- ied in the literature. One approach taken by several recent studies [7, 23, 25, 6, 24] first performs a renumbering of the document IDs in the collection that groups similar documents together, and then applies standard compression techniques. It is known that this can significantly improve index com- pression compared to a random document ordering. We study index compression and query processing tech- niques for such reordered indexes. Previous work has focused on determining the best possible ordering of documents. In contrast, we assume that such an ordering is already given, and focus on how to optimize compression methods and query processing for this case. We perform an extensive study of compression techniques for document IDs and present new optimizations of existing techniques which can achieve signif- icant improvement in both compression and decompression performances. We also propose and evaluate techniques for compressing frequency values for this case. Finally, we study the effect of this approach on query processing performance. Our experiments show very significant improvements in in- dex size and query processing speed on the TREC GOV2 collection of 25.2 million web pages.

We describe a system for synchronization and organization of user-contributed content from live music events. We start with a set of short video clips taken at a single event by multiple contributors, who were using a varied set of capture devices. Using audio fingerprints, we synchronize these clips such that overlapping clips can be displayed simultaneously. Furthermore, we use the timing and link structure generated by the synchronization algorithm to improve the findability and representation of the event content, including identifying key moments of interest and descriptive text for important captured segments of the show. We also identify the preferred audio track when multiple clips overlap. We thus create a much improved representation of the event that builds on the automatic content match. Our work demonstrates important principles in the use of content analysis techniques for social media content on the Web, and applies those principles in the domain of live music capture.

There are several semantic sources that can be found in the Web that are either explicit, e.g. Wikipedia, or implicit, e.g. derived from Web usage data. Most of them are related to user generated content (UGC) or what is called today the Web 2.0. In this talk we show several applications of mining the wisdom of crowds behind UGC to improve search. We will show live demos to find relations in the Wikipedia or to improve image search as well as our current research in the topic. Our final goal is to produce a virtuous data feedback circuit to leverage the Web itself.

Motivated by contextual advertising systems and other web applications involving efficiency–accuracy tradeoffs, we study similarity caching. Here, a cache hit is said to occur if the requested item is similar but not necessarily equal to some cached item. We study two objectives that dictate the efficiency–accuracy tradeoff and provide our caching policies for these objectives. By conducting extensive experiments on real data we show similarity caching can significantly improve the efficiency of contextual advertising systems, with minimal impact on accuracy. Inspired by the above, we propose a simple generative model that embodies two fundamental characteristics of page requests arriving to advertising systems, namely, long-range dependences and similarities. We provide theoretical bounds on the gains of similarity caching in this model and demonstrate these gains empirically by fitting the actual data to the model.

Sponsored search systems are tasked with matching queries to relevant advertisements. The current state-of-the-art matching algorithms expand the user’s query using a variety of external resources, such as Web search results. While these expansion-based algorithms are highly effective, they are largely inefficient and cannot be applied in real-time. In practice, such algorithms are applied offline to popular queries, with the results of the expensive operations cached for fast access at query time. In this paper, we describe an efficient and effective approach for matching ads against rare queries that were not processed offline. The approach builds an expanded query representation by leveraging offline processing done for related popular queries. Our experimental results show that our approach significantly improves the effectiveness of advertising on rare queries with only a negligible increase in computational cost.

Quicklinks for a website are navigational shortcuts displayed below the website homepage on a search results page, and that let the users directly jump to selected points inside the website. Since the real-estate on a search results page is constrained and valuable, picking the best set of quicklinks to maximize the benefits for a majority of the users becomes an important problem for search engines. Using user browsing trails obtained from browser toolbars, and a simple probabilistic model, we formulate the quicklink selection problem as a combinatorial optimizaton problem. We first demonstrate the hardness of the objective, and then propose an algorithm that is provably within a factor of (1 − 1/e) of the optimal. We also propose a different algorithm that works on trees and that can find the optimal solution; unlike the previous algorithm, this algorithm can incorporate natural constraints on the set of chosen quicklinks. The efficacy of our methods is demonstrated via empirical results on both a manually labeled set of websites and a set for which quicklink click-through rates for several webpages were obtained from a real-world search engine.

Contextual advertising (also called content match) refers to the placement of small textual ads within the content of a generic web page. It has become a significant source of revenue for publishers ranging from individual bloggers to major newspapers. At the same time it is an important way for advertisers to reach their intended audience. This reach depends on the total number of exposures of the ad (impressions) and its click-through-rate (CTR) that can be viewed as the probability of an end-user clicking on the ad when shown. These two orthogonal, critical factors are both difficult to estimate and even individually can still be very informative and useful in planning and budgeting advertising campaigns. In this paper, we address the problem of forecasting the number of impressions for new or changed ads in the system. Producing such forecasts, even within large margins of error, is quite challenging: 1) ad selection in contextual advertising is a complicated process based on tens or even hundreds of page and ad features; 2) the publishers’ content and traffic vary over time; and 3) the scale of the problem is daunting: over a course of a week it involves billions of impressions, hundreds of millions of distinct pages, hundreds of millions of ads, and varying bids of other competing advertisers. We tackle these complexities by simulating the presence of a given ad with its associated bid over weeks of historical data. We obtain an impression estimate by counting how many times the ad would have been displayed if it were in the system over that period of time. We estimate this count by an efficient two-level search algorithm over the distinct pages in the data set. Experimental results show that our approach can accurately forecast the expected number of impressions of contextual ads in real time. We also show how this method can be used in tools for bid selection and ad evaluation.

The “algorithmic small-world hypothesis” states that not only are pairs of individuals in a large social network connected by short paths, but that ordinary individuals can find these paths. Although theoretically plausible, empirical evidence for the hypothesis is limited, as most chains in “small-world” experiments fail to complete, thereby biasing estimates of “true” chain lengths. Using data from two recent small-world experiments, comprising a total of 162,328 message chains, and directed at one of 30 “targets” spread across 19 countries, we model heterogeneity in chain attrition rates as a function of individual attributes. We then introduce a rigorous way of estimating true chain lengths that is provably unbiased, and can account for empiricallyobserved variation in attrition rates. Our findings provide mixed support for the algorithmic hypothesis. On the one hand, it appears that roughly half of all chains can be completed in 6-7 steps—thus supporting the “six degrees of separation” assertion—but on the other hand, estimates of the mean are much longer, suggesting that for at least some of the population, the world is not “small” in the algorithmic sense. We conclude that search distances in social networks are fundamentally different from topological distances, for which the mean and median of the shortest path lengths between nodes tend to be similar.

Web search engines are facing formidable performance challenges due to data sizes and query loads. The major engines have to process tens of thousands of queries per second over tens of billions of documents. To deal with this heavy workload, such engines employ massively parallel systems consisting of thousands of machines. The significant cost of operating these systems has motivated a lot of recent research into more efficient query processing mechanisms. We investigate a new way to build such high performance IR systems using graphical processing units (GPUs). GPUs were originally designed to accelerate computer graphics applications through massive on-chip parallelism. Recently a number of researchers have studied how to use GPUs for other problem domains such as databases and scientific computing [9, 8, 12]. Our contribution here is to design a basic system architecture for GPU-based high-performance IR, to develop suitable algorithms for subtasks such as inverted list compression, list intersection, and top-k scoring, and to show how to achieve highly efficient query processing on GPUbased systems. Our experimental results for a prototype GPU-based system on 25.2 million web pages shows promising gains in query throughput.

Due to the reliance on the textual information associated with an image, image search engines on the Web lack the discriminative power to deliver visually diverse search results. The textual descriptions are key to retrieve relevant results for a given user query, but at the same time provide little information about the rich image content. In this paper we investigate three methods for visual diversification of image search results. The methods deploy lightweight clustering techniques in combination with a dynamic weighting function of the visual features, to best capture the discriminative aspects of the resulting set of images that is retrieved. A representative image is selected from each cluster, which together form a diverse result set. Based on a performance evaluation we find that the outcome of the methods closely resembles human perception of diversity, which was established in an extensive clustering experiment carried out by human assessors. models deployed on the Web and by these photo sharing sites rely heavily on search paradigms developed within the field Information Retrieval. This way, image retrieval can benefit from years of research experience, and the better this textual metadata captures the content of the image, the better the retrieval performance will be. It is also commonly acknowledged that a picture has to be seen to fully understand its meaning, significance, beauty, or context, simply because it conveys information that words can not capture, or at least not in any practical setting. This explains the large number of papers on content-based image retrieval (CBIR) that has been published since 1990, the breathtaking publication rates since 1997 [12], and the continuing interest in the field [4]. Moving on from simple low-level features to more discriminative descriptions, the field has come a long way in narrowing down the semantic gap by using high-level semantics [8]. Unfortunately, CBIR-methods using higher level semantics usually require extensive training, intricate object ontologies or expensive construction of a visual dictionary, and their performance remains unfit for use in large scale online applications such as the aforementioned search engines or websites. Consequently, retrieval models operating in the textual metadata domain are therefore deployed here. In these applications, image search results are usually displayed in a ranked list. This ranking reflects the similarity of the image’s metadata to the textual query, according to the textual retrieval model of choice. There may exist two problems with this ranking. First, it may be lacking visual diversity. For instance, when a specific type or brand of car is issued as query, it may very well be that the top of this ranking displays many times the same picture that was released by the marketing division of the company. Similarly, pictures of a popular holiday destination tend to show the same touristic hot spot, often taken from the same angle and distance. This absence of visual diversity is due to the nature of the image annotation, which does not allow or motivate people to adequately describe the visual content of an image. Second, the query may have several aspects to it that are not sufficiently covered by the ranking. Perhaps the user is interested in a particular aspect of the query, but doesn’t know how to express this explicitly and issues a broader, more general query. It could also be that a query yields so many different results, that it’s hard to get an overview of the collection of relevant images in the database.