TeCNTS: A Robust Collaborative Filtering Recommendation Scheme Based on Time-effective Close Neighbor Trusted Selection Strategy
Zhigeng Han, Yuanzhe Fan, Geng Chen, Ting Zhou · 2022
The traditional collaborative filtering recommendation (CFR) scheme assumes that the data is static and usually ignores the dynamic phenomenon in the sample data. To solve this issue, many CFR schemes used dynamic information such as user interest change and dynamic trust relationships in their strategies. However, most of them did not handle the malicious changes in user interests and inadvertent fluctuation in users’ trust degree. When suffering from shilling attack, they still cannot get high performance in prediction accuracy and anti-attack. In order to improve CFR robustness, we present a robust CFR scheme named TeCNTS based on time-effective close neighbor trusted selection strategy. We deal the malicious changes in user interest with a user time-effective similarity measure function and inadvertent fluctuation of users’ trust degree with a reliable trust evaluation model, respectively. Based on the MovieLens dataset, we evaluate the performance of TeCNTS. The results show that TeCNTS is superior to the baseline schemes in prediction accuracy, accuracy stability, anti-attack and attacker filtering, and it is a robust CFR scheme.