Search

Expert-opinions approaches based on CF (collaborative filtering) have demonstrated a successful means for top-N recommendation, such as Expert-CF. However, the choice of the similarity measure used for evaluation of user-expert relationships is crucial for the success of such approaches. In this paper, we present an approach to calculate user-expert similarities by formulating a regression problem which enables us to extract the similarities from the data in a problem specific way. A comprehensive set of experiments is conducted by predicting a subset of the MovieLens data set. We use ratings crawled from a web portal of expert reviews to generate high quality recommendations. The experiments show that the proposed method improves upon the standard Expert-CF model and outperforms other state-of-the-art top-N recommendation approaches in terms of achieving good balance between recommendation quality and speed.

Prefetching is an important technique for single Web server to reduce the average Web access latency and applying it on cluster server will produce better performance. Two models for parallel Web prefetching on cluster server described in the form of I/O automaton are proposed in this paper according to the different service approaches of Web cluster server: session persistence and session non-persistence. Meanwhile, an advanced scheduling algorithm based on Web prefetching (Prefetch_LARD) is put forward. By mining the transition probability between pages from Web access logs, the algorithm builds up a prefetching model based on Markov chain. Experiments show that under the same test environment, Prefetch_LARD algorithm increases cache hit ratio in up to 26.9% and the throughput in up to 18.8% compared with the classical locality-aware request distribution (LARD) algorithm

Many Web mining methods have recently been used to model user navigational behavior based on log files of the Web server. Cluster-based server architectures combine good performance and low cost, and are widely used for Web service. In this paper, we propose a parallel Web mining (PWM) algorithm for link prediction in the environment of Web cluster server consisting of several nodes that act as independent Web servers. According to the PWM algorithm, the transition probability matrixes are firstly obtained from the Web log flies of each node by adopting the Markov chain model, compressed under the constraint of the probability threshold c and parallel threshold a, and then sent to the central node which combine these independent results to get an integrated result by some rules. By different accuracy requirement, the PWM algorithm can be divided into simple PWM algorithm (S-PWM), faster but less accurate, and complex PWM algorithm (C-PWM), slower but more accurate. Furthermore, a related incremental parallel Web mining (I-PWM) algorithm is put forward too. The experimental results show that PWM algorithm can not only alleviate the communication cost by sending the mined transition probability matrix and decrease the time complexity by disposing in parallel but also hardly affect the accuracy of the Web mining result.