个性化搜索中用户兴趣模型匿名化研究

User Profile Anonymization in Personalized Web Search

为了解决个性化搜索技术所潜在的用户隐私信息泄露的问题,提出了用户兴趣模型匿名化方法。首先根据用户兴趣模型之间的相似性将其聚类为满足p-链接性的等价组,然后计算聚类后兴趣条目的权值。所谓的p-链接性是指攻击者根据背景知识链接确定某一用户的概率不超过p。该方法可实现用户兴趣模型匿名化以及兴趣倾向不发生改变,既保护了用户隐私信息,同时也保证了个性化检索性能。实验表明:随着相关结果个数的增多,匿名化后搜索结果的查全率基本能保证在50%以上,另外p-链接性的减小对于查全率的影响并不是太大。

To solve potential user privacy leaking in personalized web search,an approach of anonymizing user profiles is proposed.According to the similarity of user profiles,the profiles are clustered as equivalences which meet p-linkability and then calculated the weight of interest items.Here p-linkability means that the probability that a certain user is determined by the attacker on the basis of the background knowledge link is less than p.This approach realizes the anonymization of user profiles and the constancy of the user interest tendency,and sufficiently protects user privacy while the anonymized user profiles are still effective in personalized web search.It is verified that after anonymization the recall ratio of the search result can be guaranteed over 50% as the related items increase.Meanwhile,the reduction of p-linkability does not greatly influence the recall ratio.