Proc. 2nd International Workshop on Management of Information on the Web- Web Data and Text

Collaborative filtering uses a database about consumers ’ preferences to make personal product recommendations and is achieving widespread success in E-Commerce nowadays. In this paper, we present several feature-weighting methods to improve the accuracy

Proc. 2nd International Workshop on Management of Information on the Web - Web Data and Text Mining (MIW’01)

Feature Weighting and Instance Selection for Collaborative Filtering

Kai Yu2, Zhong Wen2, Xiaowei Xu1, Martin Ester2

1

Information and Communications, Corporate Technology, Siemens AG

2

Institute for Computer Science, University of Munich

Xiaowei.Xu@mchp.siemens.de, {yu_k, wen, ester}@dbs.informatik.uni-muenchen.de

Abstract

Collaborative filtering uses a database aboutconsumers’ preferences to make personal productrecommendations and is achieving widespread success inE-Commerce nowadays. In this paper, we present severalfeature-weighting methods to improve the accuracy ofcollaborative filtering algorithms. Furthermore, wepropose to reduce the training data set by selecting onlyhighly relevant instances. We evaluate various methods onthe well-known EachMovie data set. Our experimentalresults show that mutual information achieves the largestaccuracy gain among all feature-weighting methods. Themost interesting fact is that our data reduction methodeven achieves an improvement of the accuracy of about6% while speeding up the collaborative filteringalgorithm by a factor of 15.

consumer.

Collaborative filtering has been very successful inboth research and practice. However, there still remainimportant research issues in overcoming two fundamentalchallenges for collaborative filtering [8].

The first challenge is to improve the scalability of thecollaborative filtering algorithms. Existing collaborativefiltering algorithms can deal with thousands of consumerswithin a reasonable time, but the demand of modern E-Commerce systems is to handle tens of millions ofconsumers.

The second challenge is to improve the quality of therecommendations for the consumers. Consumers needrecommendations they can trust to help them findproducts they will like. If a consumer trusts a recomendersystem, purchases a product, but finds out he does not likethe product, the consumer will be unlikely to use therecommender systems again.

In this paper, we present different feature weightingmethods to improve the accuracy of collaborative filteringalgorithm. Furthermore, we introduce a relevancemeasure of an instance to the target and propose to reducethe training data set by selecting only highly relevantinstances.

In section 2, we briefly introduce collaborativefiltering algorithms. We present different featureweighting methods including inverse user frequency,entropy and mutual information in section 3. We proposea mutual information based data reduction method forcollaborative filtering in section 4. The empiricalevaluation of these methods and results are reported insection 5. The paper ends with a summary and someinteresting future work.

1. Introduction

The Internet is increasingly used as a channel for salesand marketing. More and more people purchase productsthrough the Internet. One main problem that thecustomers face is how to find the product they like frommillions of products. For the vendor, again, it is crucial tofind out about the customers’ preferences for products.Collaborative filtering or recommender systems haveemerged in response to these problems[1] [6][10].

Collaborative filtering accumulates a database ofconsumers’ product preferences, and then uses them tomake customer-tailored recommendations for productssuch as clothing, music, books, furniture, and movies. Theconsumer's preference can be either explicit votes orimplicit usage/purchase history. Collaborative filteringcan help E-commerce in converting web surfers intobuyers by personalization of the web interface. It can alsoimproves cross-sell by suggesting other products theconsumer might be interested in. In a world where an E-commerce site's competitors are only a click or two away,gaining customer loyalty is an essential business strategy.Collaborative filtering can improve the loyalty by creatinga value-added relationship between supplier and

2. Collaborative Filtering

The task in collaborative filtering is to predict thepreference of an active consumer to a given product basedon a database of consumer' product preferences. There aretwo general classes of collaborative filtering algorithms:memory-based methods and model-based methods.

Memory-based algorithm [6][10] is the most popularprediction technique in collaborative filtering

The work was performed in Cooperate Technology, Siemens AG. The contact author is Xiaowei Xu: Xiaowei.Xu@mchp.siemens.de

第1页下一页