Augmenting privacy policies with inference detection
A system is provided for augmenting a privacy policy. During operation, the system obtains a set of training documents and at least one seed keyword associated with the privacy policy. The system extracts a number of candidate keywords from the training documents and formulates at least one query based on the candidate keywords. The system then issues the query to a corpus. In response to the query, the system receives a set of result documents. The system further determines whether a respective keyword extracted from the result documents matches at least one seed keyword. The system then augments the privacy policy by associating the candidate keyword corresponding to the respective keyword with the privacy policy based on the determination. In addition, the system applies the augmented privacy policy to a subject document and produces a result to indicate whether the subject document is in violation of the privacy policy.
1. A computer-executed method for augmenting a privacy policy, the method comprising:
obtaining a set of training documents and a seed keyword associated with the privacy policy;
extracting a candidate keyword from the training documents;
issuing at least one query comprising the candidate keyword to a corpus;
receiving a set of result documents;
evaluating an inference strength between the candidate keyword and a respective seed keyword, wherein evaluating the inference strength comprises evaluating a ratio between the number of search hits from a query containing both the candidate keyword and the respective seed keyword, and the number of search hits from a query containing only the candidate keyword;
determining that the evaluated inference strength is greater than a predetermined threshold ratio;
responsive to the evaluated inference strength being greater than the predetermined threshold ratio, augmenting the privacy policy by associating the candidate keyword with the privacy policy; and
applying the augmented privacy policy to a subject document to determine whether the subject document triggers the privacy policy, wherein applying the augmented privacy policy comprises searching the subject document for occurrences of any of the candidate keywords associated with the augmented privacy policy.
2. The method of claim 1 , further comprising extracting the keywords from the result documents by determining a term-frequency inverse-document-frequency (TF.IDF) weight for a respective word or phrase contained in a respective result document.
3. The method of claim 1 , wherein obtaining the seed keywords comprises issuing a query based on a sensitive topic to the corpus and extracting the seed keywords from a number of documents returned in response to the query.
4. The method of claim 1 , wherein obtaining the training documents comprises issuing a query based on a sensitive topic to the corpus and retrieving a predetermined number of highest ranked documents returned in response to the query.
5. The method of claim 1 , wherein obtaining the set of training documents comprises receiving one or more documents that are known to pertain to sensitive subject matter.
6. The method of claim 1 , wherein obtaining the set of training documents comprises:
issuing an initial query to a corpus, wherein the initial query includes the seed keyword associated with the privacy policy; and
selecting a predetermined number of top-ranking search results of the initial query.
7. A computer system for augmenting a privacy policy, the computer system comprising:
a processor;
a memory coupled to the processor;
a training-document obtaining mechanism configured to obtain a set of training documents associated with the privacy policy;
a seed-keyword obtaining mechanism configured to obtain a seed keyword associated with the privacy policy;
a keyword-extraction mechanism configured to extract a candidate keyword from the training documents;
a query-issuing mechanism configured to issue at least one query comprising the candidate keywords to a corpus;
a result-receiving mechanism configured to receive a set of result documents;
an evaluation mechanism configured to evaluate an inference strength between a candidate keyword and a respective seed keyword, wherein evaluating the inference strength comprises evaluating a ratio between the number of search hits from a query containing both the candidate keyword and the respective seed keyword, and the number of search hits from a query containing only the candidate keyword;
a comparison mechanism configured to determine that the evaluated inference strength is greater than a predetermined threshold ratio
a privacy-policy augmentation mechanism configured to augment the privacy policy by associating the candidate keyword with the privacy policy, responsive to the evaluated inference strength being greater than the predetermined threshold ratio; and
a privacy-policy enforcement mechanism configured to apply the augmented privacy policy to a subject document, wherein applying the augmented privacy policy comprises searching the subject document for occurrences of any of the candidate keywords associated with the augmented privacy policy.
8. The computer system of claim 7 , wherein the keyword-extraction mechanism is further configured to extract the keywords from the result documents by determining a term-frequency inverse-document-frequency (TF.IDF) weight for a respective word or phrase contained in a respective result document.
9. The computer system of claim 7 , wherein while obtaining the seed keywords, the seed-keyword obtaining mechanism is further configured to issue a query based on a sensitive topic to the corpus and extract the seed keywords from a number of documents returned in response to the query.
10. The computer system of claim 7 , wherein while obtaining the training documents, the training-document obtaining mechanism is further configured to issue a query based on a sensitive topic to the corpus and retrieve a predetermined number of highest ranked documents returned in response to the query.
11. The computer system of claim 7 , wherein while obtaining the set of training documents, the training-document obtaining mechanism is further configured to receive one or more documents that are known to pertain to sensitive subject matter.
12. The computer system of claim 7 , wherein while obtaining the set of training documents, the training-document obtaining mechanism is further configured to:
issue an initial query to a corpus, wherein the initial query includes the seed keyword associated with the privacy policy; and
select a predetermined number of top-ranking search results of the initial query.
13. A non-transitory computer-readable medium storing instructions which when executed by a computer cause the computer to perform a method for augmenting a privacy policy, the method comprising:
obtaining a set of training documents and a seed keyword associated with the privacy policy;
extracting a candidate keyword from the training documents;
issuing at least one query comprising the candidate keyword to a corpus;
receiving a set of result documents;
evaluating an inference strength between the candidate keyword and a respective seed keyword, wherein evaluating the inference strength comprises evaluating a ratio between the number of search hits from a query containing both the candidate keyword and the respective seed keyword, and the number of search hits from a query containing only the candidate keyword;
determining that the evaluated inference strength is greater than a predetermined threshold ratio;
responsive to the evaluated inference strength being greater than the predetermined threshold ratio, augmenting the privacy policy by associating the candidate keyword with the privacy policy; and
applying the augmented privacy policy to a subject document to determine whether the subject document triggers the privacy policy, wherein applying the augmented privacy policy comprises searching the subject document for occurrences of any of the candidate keywords associated with the augmented privacy policy.
14. The computer-readable medium of claim 13 , further comprising extracting the keywords from the result documents by determining a term-frequency inverse-document-frequency (TF.IDF) weight for a respective word or phrase contained in a respective result document.
15. The computer-readable medium of claim 13 , wherein obtaining the seed keywords comprises issuing a query based on a sensitive topic to the corpus and extracting the seed keywords from a number of documents returned in response to the query.
16. The computer-readable medium of claim 13 , wherein obtaining the training documents comprises issuing a query based on a sensitive topic to the corpus and retrieving a predetermined number of highest ranked documents returned in response to the query.
17. The computer-readable medium of claim 13 , wherein obtaining the set of training documents comprises receiving one or more documents that are known to pertain to sensitive subject matter.
18. The computer-readable medium of claim 13 , wherein obtaining the set of training documents comprises:
issuing an initial query to a corpus, wherein the initial query includes the seed keyword associated with the privacy policy; and
selecting a predetermined number of top-ranking search results of the initial query.