25. Multiple Imputation for Nonresponse in Surveys. By D. B. Rubin. ISBN 0 471 08705 X. Wiley, Chichester, 1987. 258 pp. £30.25.

Multiple Imputation for Nonresponse in Surveys

Generalized Linear Models (2nd ed.)

This chapter introduces the subject of statistical pattern recognition (SPR). It starts by considering how features are defined and emphasizes that the nearest neighbor algorithm achieves error rates comparable with those of an ideal Bayes’ classifier. The concepts of an optimal number of features, representativeness of the training data, and the need to avoid overfitting to the training data are stressed. The chapter shows that methods such as the support vector machine and artificial neural networks are subject to these same training limitations, although each has its advantages. For neural networks, the multilayer perceptron architecture and back-propagation algorithm are described. The chapter distinguishes between supervised and unsupervised learning, demonstrating the advantages of the latter and showing how methods such as clustering and principal components analysis fit into the SPR framework. The chapter also defines the receiver operating characteristic, which allows an optimum balance between false positives and false negatives to be achieved.

Statistical Pattern Recognition

Whether we're buying a pair of jeans, ordering a cup of coffee, selecting a long-distance carrier, applying to college, choosing a doctor, or setting up a 401(k), everyday decisions-both big and small-have become increasingly complex due to the overwhelming abundance of choice with which we are presented. As Americans, we assume that more choice means better options and greater satisfaction. But beware of excessive choice: choice overload can make you question the decisions you make before you even make them, it can set you up for unrealistically high expectations, and it can make you blame yourself for any and all failures. In the long run, this can lead to decision-making paralysis, anxiety, and perpetual stress. And, in a culture that tells us that there is no excuse for falling short of perfection when your options are limitless, too much choice can lead to clinical depression. In The Paradox of Choice, Barry Schwartz explains at what point choice-the hallmark of individual freedom and self-determination that we so cherish-becomes detrimental to our psychological and emotional well-being. In accessible, engaging, and anecdotal prose, Schwartz shows how the dramatic explosion in choice-from the mundane to the profound challenges of balancing career, family, and individual needs-has paradoxically become a problem instead of a solution. Schwartz also shows how our obsession with choice encourages us to seek that which makes us feel worse. By synthesizing current research in the social sciences, Schwartz makes the counter intuitive case that eliminating choices can greatly reduce the stress, anxiety, and busyness of our lives. He offers eleven practical steps on how to limit choices to a manageable number, have the discipline to focus on those that are important and ignore the rest, and ultimately derive greater satisfaction from the choices you have to make.

The Paradox of Choice : Why More is Less (Mengapa Lebih Itu Justru Kurang)

Every house is different. It is important that house price indexes take account of these quality differences. Hedonic methods which express house prices as a function of a vector of characteristics (such as number of bedrooms and bathrooms, land area and location) are particularly useful for this purpose. I consider here some developments in the hedonic methodology, as it is applied in a housing context, that have occurred in the last three decades. A number of hedonic house price indexes are now available. However, it is often difficult to see how these indexes relate to each other. For this reason I attempt to impose some structure on the literature by developing a taxonomy of hedonic indexes, and then show how existing indexes fit into this taxonomy. Also discussed are some promising areas for future research in the hedonic field. In particular, greater use needs to be made of spatial econometric and nonparametric methods to exploit the increased availability of geospatial data. The main criticisms of the hedonic approach are evaluated and compared with those of the repeat-sales and stratified median methods. The overall conclusion is that the advantages of the hedonic approach outweigh its disadvantages.

Hedonic price indexes for residential housing: a survey, evaluation and taxonomy

Hedonic pricing models attempt to model a relationship between object attributes and the object's price. Traditional hedonic pricing models are often parametric models that suffer from misspecification. In this paper we create these models by means of boosted CART models. The method is explained in detail and applied to various datasets. Empirically, we find substantial reduction of errors on out-of-sample data for two out of three datasets compared with a stepwise linear regression model. We interpret the boosted models by partial dependence plots and relative importance plots. This reveals some interesting nonlinearities and differences in attribute importance across the model types.

Boosting the accuracy of hedonic pricing models

In content- and knowledge-based recommender systems often a measure of (dis)similarity between products is used. Frequently, this measure is based on the attributes of the products. However, which attributes are important for the users of the system remains an important question to answer. In this paper, we present two approaches to determine attribute weights in a dissimilarity measure based on product popularity. We count how many times products are sold and based on this, we create two models to determine attribute weights: a Poisson regression model and a novel boosting model minimizing Poisson deviance. We evaluate these two models in two ways, namely using a clickstream analysis on four different product catalogs and a user experiment. The clickstream analysis shows that for each product catalog the standard equal weights model is outperformed by at least one of the weighting models. The user experiment shows that users seem to have a different notion of product similarity in an experimental context.

Determination of Attribute Weights for Recommender Systems Based on Product Popularity

In contentand knowledge-based recommender systems often a measure of (dis)similarity between items is used. Frequently, this measure is based on the attributes of the items. However, which attributes are important for the users of the system remains an important question to answer. In this paper, we present an approach to determine attribute weights in a dissimilarity measure using clickstream data of an e-commerce website. Counted is how many times products are sold and based on this a Poisson regression model is estimated. Estimates of this model are then used to determine the attribute weights in the dissimilarity measure. We show an application of this approach on a product catalog of MP3 players provided by Compare Group, owner of the Dutch price comparison site http://www.vergelijk.nl, and show how the dissimilarity measure can be used to improve 2D product catalog visualizations.

Choosing Attribute Weights for Item Dissimilarity using Clikstream Data with an Application to a Product Catalog Map

Many content-based recommendation approaches are based on a dissimilarity measure based on the product attributes. In this paper, we evaluate four dissimilarity measures for product recommendation using an online survey. In this survey, we asked users to specify which products they considered to be relevant recommendations given a reference product. We used microwave ovens as product category. Based on these responses, we create a relative relevance matrix we use to evaluate the dissimilarity measures with. Also, we use this matrix to estimate weights to be used in the dissimilarity measures. In this way, we evaluate four dissimilarity measures: the Euclidean Distance, the Hamming Distance, the Heterogeneous Euclidean-Overlap Metric, and the Adapted Gower Coefficient. The evaluation shows that these weights improve recommendation performance. Furthermore, the experiments indicate that when recommending a single product, the Heterogeneous Euclidean-Overlap Metric should be used and when recommending more than one product the Adapted Gower Coefficient is the best alternative. Finally, we compare these dissimilarity measures with a collaborative method and show that this method performs worse than the dissimilarity based approaches.

An Empirical Comparison of Dissimilarity Measures for Recommender Systems

We propose a new hybrid recommender system that combines some advantages of collaborative and content-based recommender systems. While it uses ratings data of all users, as do collaborative recommender systems, it is also able to recommend new items and provide an explanation of its recommendations, as do content-based systems. Our approach is based on the idea that there are communities of users that find the same characteristics important to like or dislike a product. This model is an extension of the probabilistic latent semantic model for collaborative filtering with ideas based on clusterwise linear regression. On a movie data set, we show that the model is competitive to other recommenders and can be used to explain the recommendations to the users.

Martijn Kagie

Papers

Boosting the accuracy of hedonic pricing models

Determination of Attribute Weights for Recommender Systems Based on Product Popularity

Choosing Attribute Weights for Item Dissimilarity using Clikstream Data with an Application to a Product Catalog Map

An Empirical Comparison of Dissimilarity Measures for Recommender Systems

Including Item Characteristics in the Probabilistic Latent Semantic Analysis Model for Collaborative Filtering