Automatic combination of multiple ranked retrieval systems
SIGIR '94 Proceedings of the 17th annual international ACM SIGIR conference on Research and development in information retrieval
Robust trainability of single neurons
Journal of Computer and System Sciences
Making large-scale support vector machine learning practical
Advances in kernel methods
Machine Learning
Modern Information Retrieval
Partially Supervised Classification of Text Documents
ICML '02 Proceedings of the Nineteenth International Conference on Machine Learning
Optimizing search engines using clickthrough data
Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining
PEBL: positive example based learning for Web page classification using SVM
Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining
Building Text Classifiers Using Positive and Unlabeled Examples
ICDM '03 Proceedings of the Third IEEE International Conference on Data Mining
Journal of Artificial Intelligence Research
Learning to classify texts using positive and unlabeled data
IJCAI'03 Proceedings of the 18th international joint conference on Artificial intelligence
| Hi-index | 0.00 |
Most existing search engines employ static ranking algorithms that do not adapt to the specific needs of users. Recently, some researchers have studied the use of clickthrough data to adapt a search engine’s ranking function. Clickthrough data indicate for each query the results that are clicked by users. As a kind of implicit relevance feedback information, clickthrough data can easily be collected by a search engine. However, clickthrough data is sparse and incomplete, thus, it is a challenge to discover accurate user preferences from it. In this paper, we propose a novel algorithm called “Spy Naïve Bayes” (SpyNB) to identify user preferences generated from clickthrough data. First, we treat the result items clicked by the users as sure positive examples and those not clicked by the users as unlabelled data. Then, we plant the sure positive examples (the spies) into the unlabelled set of result items and apply a naïve Bayes classification to generate the reliable negative examples. These positive and negative examples allow us to discover more accurate user’s preferences. Finally, we employ the SpyNB algorithm with a ranking SVM optimizer to build an adaptive metasearch engine. Our experimental results show that, compared with the original ranking, SpyNB can significantly improve the average ranks of users’ click by 20%.