Inbound content filtering via automated inference detection
View Patent ↗A system is provided to detect email spam. During operation, the system receives an email, extracts a set of keywords from the email body, and constructs a first search query based a keyword extracted from the email body. The system further constructs a second search query based on the keyword in the first query and one additional word which pertains to a known spam word or to the subject of the email. Next, the system receives a first number of hits and a second number of hits in response to the first and second search queries, respectively. The system then determines whether the email is spam based on the first number and the second number. The system can also perform Website filtering using inference detection which is based on search results received in response to search queries formulated with keywords extracted from Websites.
1. A computer-executed method for detecting unwanted email, the method comprising:
extracting a set of keywords from a received email's body;
constructing a first search query using a keyword extracted from the email body;
constructing a second search query using the keyword and an additional word that is a known spam topic word;
issuing, to a search engine, the first search query and the second search query;
receiving, from the search engine, a first number of hits and a second number of hits in response to the first and second search queries respectively;
determining whether the email is unwanted, based on a comparison between the first number of hits and the second number of hits and a likelihood that the additional word corresponds to a spam topic, which involves computing a ratio between the first number and the second number; and
indicating whether the email is unwanted based on the determination.
2. The method of claim 1 , wherein extracting the set of keywords from the email body comprises determining a term-frequency inverse-document-frequency (TF.IDF) weight for a respective word or phrase contained in the email body.
3. The method of claim 2 , wherein extracting the set of keywords from the email body further comprises ranking a respective word based on its TF.IDF weight.
4. A computer system for detecting unwanted email, the computer system comprising:
a processor;
a memory coupled to the processor;
a keyword-extraction mechanism configured to extract a set of keywords from a received email's body;
a query-formulation mechanism configured to construct:
a first search query using a keyword extracted from the email body, and
a second query using the keyword and an additional word that is a known spam topic word;
a hits-receiving mechanism configured to:
issue, to a search engine, the first search query and the second search query; and
receive, from the search engine, a first number of hits and a second number of hits in response to the first and second search queries respectively;
a determination mechanism configured to determine whether the email is unwanted based on a comparison between the first number of hits and the second number of hits and a likelihood that the additional word corresponds to a spam topic, wherein determining whether the email is unwanted comprises computing a ratio between the first number and the second number; and
an indication mechanism configured to indicate whether the email is unwanted based on the determination.
5. The method of claim 4 , wherein while extracting the set of keywords from the email body, the keyword-extraction mechanism is configured to determine a term-frequency inverse-document-frequency (TF.IDF) weight for a respective word or phrase contained in the email body.
6. The computer system of claim 5 , wherein while extracting the set of keywords from the email body, the keyword-extraction mechanism is further configured to rank a respective word based on its TF.IDF weight.