1. Case Analysis of Trusted Attacks
For example, the recommendation system is attacked and the recommendation result is shifted.
表1.1 用户评分矩阵
Tab. 1.1 User rating matrix
Table 1.1 is a user rating matrix. The rows represent users, among which User1 and User6 are normal users, Attacker1, Attacker2 and Attacker3 are attacking users, and the column represents the project Movie1.Movie6 is six different movies. The number represents the user's rating for the project. The rating range is 1-5, 5 means like it very much, 1 means not like it very much,-means not rated,? Means to predict the user User6's rating of the project Movie6 score. Using the idea of collaborative filtering, use Euclidean distance to calculate the similarity between User1~User5 and User6. The similarity values are 3.60, 5.92, 3.46, 4.80 and 7.41. The smaller the similarity value, the more similar to User6, and then select 3. The one with the smallest similarity value is regarded as the neighbor of User6, and the minimum values are 3.60, 3.46, and 4.80, corresponding to User1, User3, and User4, respectively. Based on User1, User3, and User4’s scores on the project Movie6, we comprehensively predict User6’s relative Score, User1, User3, and User4 have scores of 3, 2 and 1 for the project Movie6. After calculation, User6’s predicted score for the project Movie6 is 1.35, indicating that User6 does not like the project Movie6 very much, and the system will not recommend Movie6 to User6. Next, add attack users Attack1, Attack2, and Attack3, and recalculate the similarity. The similarity values are 3.60, 5.92, 3.46, 4.80, 7.41, 3.16, 1.73, and 2.82. It can be seen that the three attacking users have the smallest similarity values. , So now we select attacking users Attacker1, Attacker2, and Attacker3 as the neighbors of User6 to predict User6’s rating for the project Movie6. The three attacking users’ ratings for the project Movie6 are all 5. After calculation, User6’s predicted rating for the project Movie6 is 4.80, which means User6 likes the project Movie6, and the system will recommend Movie6 to User6. Before being attacked, User6 predicted that it did not like the project Movie6 very much, and the system would not recommend Movie6 to User6. After the attack, User6 liked the project Movie6, and the system would recommend Movie6 to User6. This is the recommended system being attacked. Recommended results.