An Inference Detection Algorithm Based on Related Tuples Mining

Existing algorithms on inference detection for database systems mainly employ functional dependencies in the database schema to detect inference, but what they can detect is limited. This paper presents a new data level inference detection algorithm. It can determine whether sensitive information can be disclosed from the user's query history through finding the related tuples between the return results of different queries. If two tuples are related to each other, then they will be merged into one tuple, thus the query history can be compressed. Moreover, the merged tuple has more information than the original two or more tuples. The experiment results show that, as the query number increases, our algorithm can infer almost the whole original relation; meanwhile the query history is compressed remarkablely. The system administrator should restrict user's query number and category to ensure that the database is secure.