A Hybrid Item-Based Recommendation Ranking Algorithm Based on User Access Patterns

Nowadays, most websites provide tremendous information organized in complex structures of web pages. Therefore, how to help users quickly find pages they are looking for is an important issue. Although a sitemap can provide navigation information

across sections

of the website, it is static and can hardly provide dynamic information based on access patterns and browsing trends. In this paper, we proposed a hybrid approach for improving recommendation ranking of the web pages for the next visit. Our raking strategy considers not only the relevance (correlation to the next page calculated by the collaborative filtering algorithm) but also the level of interest (time spent on a page) and accessibility (the distance to the next page). In order to evaluate the proposed recommendation ranking algorithm, we used the web access log (IIS log) of a website, Health 99, operated by the Bureau of Health Promotion, Taiwan. The log data was divided into training and testing sets. The measurements of the relevance, the level of interest and the distance factor were computed from the training set. The experimental results showed that the possibility of the pages in the recommendation ranking lists by our approach that were accepted by users was much higher than that proposed by the original collaborative filtering algorithm, particular in short recommendation list (< 5).