Deriving Differentially Private Session Logs for Query Suggestion
Sicong Zhang, Grace Hui Yang · 2017
Query logs are valuable resources for Information Retrieval (IR) research. However, the huge concern about user privacy has become an obstacle preventing data from being released and interfering with the advance of IR. It is understandable but still quite disappointing. Recent privacy research has begun to address the problem by anonymizing queries in query logs. In that, each query and its associated user actions are treated as one block for anonymization; each block is independent from each other. It might be sufficient to support ad-hoc retrieval that handles queries independently but will be not adequate for more complex IR tasks that require the knowledge of query sequences. In this paper, we tackle this challenge by keeping session information in differentially privately anonymized logs so that an anonymized log is able to support IR tasks, such as query suggestion and session search, that need query sequence information as well. We also provide analysis on how to achieve a proper balance between privacy and search utility.