Modified Collaborative Filtering Algorithm Based on ItemRank

Modified Collaborative Filtering Algorithm Based on ItemRank

Pengyuan Xu (Institute of System Engineering, Dalian University of Technology, Dalian, China) and Yanzhong Dang (Institute of System Engineering, Dalian University of Technology, Dalian, China)
Copyright: © 2014 |Pages: 9
DOI: 10.4018/ijkss.2014010103
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

The most crucial component of collaborative filtering recommendation algorithm (CF) is the mechanism of calculating similarities among items or users. In this paper, a new CF algorithm based on ItemRank Similarity (IRS) is proposed, which extracts items' quality characteristics from the similar matrix. The corresponding algorithmic accuracy is measured by the ranking score, precision, recall and F-measure. This algorithm provides remarkably higher accurate predictions than other modified CF algorithm.
Article Preview

1. Introduction

Nowadays, information overload is a big problem for our social (Brodera, et al., 2000). It results from the exponential growth of the Internet and World-Wide-Web. There is so much information and knowledge on the Internet and it is a hard work for us to find out the piece which is appropriate for us (Brin& Page, 1998; Kleinberg, 1999). Research on knowledge management can help us to obtain potential solutions to these problems. One of the effective tools for this problem is search engine. Search engine is a useful way for information filtering by which the relevant information could be found. But, it is to be regretted that the search engine directed towards the key words, it can not deal with personalization of information service. At the present time, to provide personal recommendations is a unique method to solve the problem. Various kinds of algorithms have been proposed, including similarity-based methods, content-based analysis (Balabanovic & Shoham, 1997; Pazzani, 1999), spectral analysis (Goldberg, et al., 2001), iteratively self-consistent refinement (Ren, et al., 2008), net-based method (Zhang, et al., 2007), and so on.

In both research and practice, collaborative filtering (CF) algorithm is one of the most famous recommendation technologies (Brin & Page, 1998; Kleinberg, 1999). A number of CF methods have been developed in the past (Liu, et al., 2006; Liu, et al., 2007; Liu, et al., 2009; Liu, et al., 2008). To make useful predictions about potential interests of a given user, the CF firstly identifies a set of similar users (items) from the past records and then makes a prediction based on the weighted combination of those similar users’ opinions (items rates). Recently, some new methods for recommendation systems are proposed. Including network-based model (Zhang, et al., 2007; Zhang, et al., 2008; Zhou, et al., 2007; Zhou, et al., 2008), these approaches has been demonstrated to be of high accuracy. When the degrees of the nodes have been considered in this model, it can deal with the problem of “dark matter” in the networks. The recommendation algorithm based on the heat conduction or diffusion process has been proved to be high accuracy and low computational complexity.

In order to increase the accuracy of the standard CF, several algorithms have been proposed. All these algorithms based on the main idea that people have their habits and favorite characteristics. A user co-collect item i and j seems they have his favorite taste. It seems that item i and j have the common characteristics which is loved by the users who co-collect it. The mainly problem is that these algorithms do not consider many users like the high-quality items even these items are not their favorite types. In other words, they have not taken into account the fact that high-quality items would be co-collected with not similar item by users. For example, many users who like funny movie very much also collected some high-quality action movies. It results large similarity between different types of items by algorithms mention above. And many action movies will be recommended to users who love funny movies. It makes wrong recommendations. How to weigh the quality of items is the key question to modify the similarity algorithm. In the field of Search Engine, the quality of web page can be measured by PageRank. Similarly, in this paper, a new definition of ItemRank is proposed. Quality of items can be measured by ItemRank.

Complete Article List

Search this Journal:
Reset
Open Access Articles: Forthcoming
Volume 8: 4 Issues (2017): 3 Released, 1 Forthcoming
Volume 7: 4 Issues (2016)
Volume 6: 4 Issues (2015)
Volume 5: 4 Issues (2014)
Volume 4: 4 Issues (2013)
Volume 3: 4 Issues (2012)
Volume 2: 4 Issues (2011)
Volume 1: 4 Issues (2010)
View Complete Journal Contents Listing