
Text in English ; Abstract: English and Turkish Includes bibliographical references (leaves 37-42) ix, 43 leaves The web has grown so rapidly in the last decade and it brought the need for proper ranking. Learning to rank (LTR) is the collection of machine learning technolo- gies that construct a ranking model using training data. The model can sort documents according to their degrees of relevance or preference. In this thesis, we introduce LTR technologies and divide them into three ap- proaches: the point-wise, pair-wise and list-wise. We review the theoritical aspects of each category and introduce the representative algorithms of them. We also introduce a new LTR method GRwC which uses classifîcation and graph algorithms. We reduce the ranking problem to a two class classifîcation problem and apply KNN algorithm on a modified LTR dataset. We compared it with the popular ranking algorithm RankingSVM. Experiments on the well-known ranking datasets show that our proposed method gives slightly worse results than RankingSVM. ıralama öğrenimi örnek verileri kullanarak bunlardan bir sıralama modeli oluşturan makine öğrenimi metotlarıdır. Bu model dokümanları önemine ya da uygunluğuna bağlı olarak sıralayabilir. Birçok Bilgiye Erişim teknolojisinin temelinde sıralama vardır. Bu yüzden Sıralama öğrenimi teknolojisi ile varolan bu teknolojiler daha da iyileştirilebilir. Sıralama öğrenimi son yıllarda artan bir popülariteye sahip olmuştur. Bunun temel sebebi Sıralama öğrenimi metotlarının arama motorları tarafından kullanılmaya başlanmış olmasıdır. Büyük arama motoru şirketleri son zamanlarda bir çok Sıralama öğrenimi algoritmaları geliştirmiş ve bu algoritmaları arama sistemlerinde kullanarak iyi sonuçlar almışlardır. Bu tezde, Sıralama öğrenimi teknolojilerini inceledik ve üç ayrı kategoriye ayırdık: nokta-bazlı, çift-bazlı ve liste-bazlı yaklaşımlar. Ayrıca yeni bir Sıralama öğrenimi algoritması tasarlayıp bunu popüler bir algoritma olan RankingSVM ile karşılatırdık. Introduction Ranking in Information Retrieval Ranking Models in IR Query-dependent Ranking Models Query-independent Ranking Models Query-level Evaluation in Information Retrieval Learning to Rank The point-wise approach Multi-class Classification for Ranking Subset Ranking with Regression Other Point-wise Algorithms Pair-wise Approach RankBoost Ranking SVM Other Pair-wise Algorithms List-wise approach RankCosine ListNet Other List-wise Algorithms Graph Ranking with Classification Experiments and Results Setup Results Conclusion
TK5102.9 .K55 2011, Algorithms., Machine learning, Machine learning., Algorithms
TK5102.9 .K55 2011, Algorithms., Machine learning, Machine learning., Algorithms
| selected citations These citations are derived from selected sources. This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically). | 0 | |
| popularity This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network. | Average | |
| influence This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically). | Average | |
| impulse This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network. | Average |
