Recent studies have shown that boosting provides excellent predictive performance across a wide variety of tasks. In Learning-to-rank, boosted models such as RankBoost and LambdaMART have been shown to be among the best performing learning methods based on evaluations on public data sets. In this paper, we show how the combination of bagging as a variance reduction technique and boosting as a bias reduction technique can result in very high precision and low variance ranking models. We perform thousands of parameter tuning experiments for LambdaMART to achieve a high precision boosting model. Then we show that a bagged ensemble of such LambdaMART boosted models results in higher accuracy ranking models while also reducing variance as much as 50%. We report our results on three public learning-to-rank data sets using four metrics. Bagged LamdbaMART outperforms all previously reported results on ten of the twelve comparisons, and bagged LambdaMART outperforms non-bagged LambdaMART on all twelve comparisons. For example, wrapping bagging around LambdaMART increases [email protected] from 0.4137 to 0.4200 on the MQ2007 data set; the best prior results in the literature for this data set is 0.4134 by RankBoost.

Wikipedia, one of the top ten most visited websites, is commonly viewed as the largest online reference for encyclopedic knowledge. Because of its open editing model -allowing anyone to enter and edit content- Wikipedia's overall quality has often been questioned as a source of reliable information.

In ubiquitous computing environments, many de- vices and agents interoperate with each other and use services provided by others. But the problem is that there may be requests that can not be fulfilled by available services. A solution to this problem is to use service brokers that compose existing services and create new value added services that can fulfill the clients' requests. In this paper, we present an architecture for a context- aware service broker that not only is able to compose services for responding to new requests, but also considers the context of its clients for providing customized and personalized outputs and behaviors.

Modern search engines use link structure of the World Wide Web in order to gain better results for ranking the results of users' queries. One of the most popular ranking algorithms which is based on link analysis is HITS. It generates very accurate outputs but because of huge amount of online computations, this algorithm is relatively slow. In this paper we introduce PHITS, a parallelized version of the HITS algorithm that is suitable for working with huge web graphs in a reason- able time. For implementing this algorithm, we use WebGraph framework and we focus on parallelizing access to web graph as the main bottleneck in the HITS algorithm. I. INTRODUCTION Search technology is one of the most important reasons for success of the web. The huge amount of information available on the web, its high growth rate, and its unstructured nature, all increase the need for search engines with high performance and accurate results. One of the major components of each search engine is its ranking algorithm. Traditional Information Retrieval (IR) systems usually use some models like VMS (4) and compute rank of results using content similarity measures between user's query and retrieved documents. But in the context of the web, there are some problems with these approaches. For example, spamming may lead to inefficient ranking. Some methods have been proposed to encounter these problems most of which uses some implicit information which is embedded in the web graph. These methods are known as Link-Analysis based algorithms. PageRank (5) and HITS (Hyperlink Induced Topic Search) (1) are the most well known algorithms in this category. PageRank, which is used by Google for ranking its results, is an offline and query-independent ranking algorithm. This means that the ranking is independent of the specific queries of users and therefore can be done once and used for all of the upcoming queries. On the other hand, HITS is an online and query-dependent algorithm. Being query dependent makes HITS more precise but it has some disadvantages too. In fact, required online computations for this algorithm is too much and the response time of the search engine after submitting queries by users is not acceptable. To overcome this problem, in this paper we will exploit the parallel processing methods to improve the execution performance of the algorithm. The rest of this paper is organized as follows. In section II, link-analysis based algorithms in general and HITS as a special case are discussed. At the end of this section, some of the variations and improvements for the HITS algorithm that are suggested in the literature are also described. Implementing the HITS algorithm and its parallel version, PHITS, are discussed in sections III and IV respectively. Finally, last section of this paper contains conclusion and some ideas for future work in this topic.

Measuring semantic similarity of Web services has several benefits and most of the proposed service discovery algorithms are based on measuring the similarity of the requested service with each of the advertised services. In this paper, we propose a method for measuring the similarity of Web services which are annotated with OWL-S ontology. First, a semantic similarity measure for determining the similarity of OWL concepts is discussed and then based on this measure, the functional similarity of services is defined. Then it is showed that the precision of algorithms that only take into account the functional properties of services for measuring their similarity are low. Therefore the textual descriptions of Web services are also taken into account and the textual similarity of services is also calculated. Then it is showed how neural networks can be used for combining these two measures for a better compound measure. The proposed technique is applied to a sample test collection and experimental results are presented which demonstrate the effectiveness of the idea