Anonymizing text documents such as Medical reports is usually handled by removing sensitive words. However, this defeats the purpose of releasing such documents for public use. While these documents may contain identifying information, we propose techniques based on generalizing terms so that identification is minimized, yet the information content in documents is preserved.  We use ontology (eg., wordnet) based generalization to deal with general text documents.

     This is an extended version of SKM 2008 paper, with more efficient schemes and a better formalization of Plausibly Deniable Search. In this scheme, a user query is issued as a standardized query, along with masking queries that are on different topics. This provides the user privacy through deniability - he could deny issuing any particular query as any other query could have been the actual user query. We use Latent Semantic Indexing, Query Topic Clustering, DMOZ dataset, Lemur Toolkit for the experiments.

     This paper introduces the idea of Plausibly Deniable Technique as a privacy-protection mechanism in a search environment, mainly web search. Noting that Private Information Retrieval is not practical with web search as it affects the advertisement business model, we offer a solution based on the deniable search - the user issues k queries and any one of them is plausibly original user query. The experiments used TREC-4 dataset that shows some very preliminary results - more comprehensive experiments are presented in the SDM 2009 paper.

    Suppose two parties want to find similar documents in their collection without releasing the contents of their own documents to the other party. This paper proposes a way to construct cosine similarity securely, which is then used for finding similar documents across different collections. Experiments are performed using Lemur toolkit and Paillier encryption scheme.

     Data provenance in the presence of confidential sources make it impossible to declare and tag documents with their plaintext sources. This paper proposes a framework that preserves the confidentiality of sources allowing all data provenance operations.