Pino: A Recommender System using Web Usage Mining

Recommender systems for websites do web personalization online and intelligently using recommendations to the users. These systems can be developed by web usage mining techniques which model user navigation patterns. In other words, data mining methods can be used to discover user access patterns from web logs. Then, recommendations to the users are provided based on these patterns. A variety of recommender systems based on web usage mining have been proposed, although improving the efficiency and complexity of them is still a challenging issue. In this paper, a recommender system called Pino has been proposed. A new approach for mining access patterns is proposed in Pino. In this approach, users' navigation patterns are modeled with a directed and weighted graph. Its vertices are webpages and the edges indicate their correlation based on co-occurrence frequencies between webpages. The weight of edges is calculated on the basis of the inverse conditional probability of viewing the webpages by considering their order. Then, webpages are clustered by partitioning this graph based on shortest paths. Recommendations will be generated based on discovered patterns with constant time complexity and with consistency to the statelessness property of HTTP protocol. Pino has been evaluated on web server logs. The effectiveness of recommendations has been evaluated by criteria applicability recommendation, correct recommendations, accuracy and coverage. Evaluation results indicate the ability of the system to improve quality of recommendations so that the harmony mean of these criteria in Pino system has reached 57% improved by 12% compared to previous systems.