Abstract. When it comes to buying expensive goods people expect to be skillfully steered through the options by well-informed sales assistants that are capable of balancing the user’s many and varied requirements. In addition users often need to be educated about the product-space, especially if they are to come to understand what is available and why certain options are being recommended by the sales-assistant. The same issues arise in interactive recommender systems, our online equivalent of a sales assistant and explanation in recommender systems, as a means to educate users and justify recommendations, is now well accepted. In this paper we focus on a novel approach to explanation. Instead of attempting to justify a particular recommendation we focus on how explanations can help users to understand the recommendation opportunities that remain if the current recommendation should not meet their requirements. We describe how this approach to explanation is tightly coupled with the generation of compound critiques, which act as a form of feedback for the users. And we argue that these explanation-rich critiques have the potential to dramatically improve recommender performance and usability. 1

Computing the frequent subsets of large multi-attribute data is both computationand data-intensive. The standard parallel algorithms require multiple passes through the data. The cost of data access may easily outweigh any performance gained by parallelizing the computational part. We address three opportunities for performance improvement: using an approximate algorithm that requires only a single pass over the data; statically partitioning the attributes to provide rudimentary load- and storage-balancing; and using a probabilistic technique to avoid generating most of the lattice of subsets implied by each object's data. The computation required is only slightly larger than levelwise algorithms, but the amount of data access is much smaller. 0 Parallel Frequent Set Counting D.B. Skillicorn skill@cs.queensu.ca Abstract: Computing the frequent subsets of large multi-attribute data is both computation- and data-intensive. The standard parallel algorithms require multiple passes th...

I want to thank Ethem Alpaydın for helping me all the time with ideas for my thesis and for his contribution to my undergraduate and graduate education. I want to thank Fikret Gürgen and Taner Bilgiç for their contribution to my undergraduate and graduate education and for participating in my thesis jury. I want to thank Dengiz Pınar, Nasuhi Sönmez and Ataman Kalkan of Gima Türk A.S¸. for supplying me the data I used in my thesis. I want to thank my family who always supported me and never left me alone during the preperation of this thesis. iv MARKET BASKET ANALYSIS FOR DATA MINING Most of the established companies have accumulated masses of data from their customers for decades. With the e-commerce applications growing rapidly, the companies will have a significant amount of data in months not in years. Data Mining, also known as Knowledge Discovery in Databases (KDD), is to find trends, patterns, correlations, anomalies in these databases which can help us to make accurate future decisions. Mining Association Rules is one of the main application areas of Data Mining. Given a set of customer transactions on items, the aim is to find correlations between the sales of items. We consider Association Mining in large database of customer transactions. We give an overview of the problem and explain approaches that have been used to attack this problem. We then give the description of the Apriori Algorithm and show results that are taken from Gima Türk A.S¸. a large Turkish supermarket chain. We also use two statistical methods: Principal Component Analysis and k-means to detect correlations between sets of items. v