1. Field of the Invention
This invention relates to the filtering of electronic mail (hereinafter “e-mail”). More particularly, the present invention relates to a method and apparatus by which dynamic filtering techniques are applied to filter out unwanted e-mail messages at various stages of transmission across one or more networks.
2. Description of the Related Art
The rapid increase in the number of users of e-mail and the low cost of distributing electronic messages via the Internet and other electronic communications networks has made marketing via the Internet an attractive medium for productivity, workflow, advertising, and business and personal communication. Consequently, e-mail and Internet newsgroups are now frequently used as the medium for widespread marketing broadcasts. These unwanted messages are commonly referred to as “spam.”
Spam is more than just an annoyance to Internet users—it represents a significant threat to the stability of vast numbers of computers and networks which comprise the Internet community. Internet service providers (hereinafter “ISPs”), online services, and corporations spend millions of dollars each year attempting to control spam. In fact, some spam distributions are so large in scope that they have the ability to crash large ISP and corporate servers. One of the reasons why spam is so pervasive is that spammers require only a computer, an address list and Internet access to distribute spam to potentially millions of Internet users. In sum, if not properly controlled, spam is capable of disabling significant portions of the Internet.
There are a number of known methods for filtering spam including Realtime Blackhole List (“RBL”) filtering, Open Relay Blocking System (“ORBS”); and Procmail rules and recipes. Frequently, these methods are designed to block spam from particular e-mail addresses from which spam is known to originate. For example, filtering methods used by America On Line® and Prodigy® use exclusion filters which block e-mail messages received from addresses that are suspected sources of spam. However, this approach is vulnerable to rapid changes in the source of unsolicited e-mail. Furthermore, because online services will generally not automatically block e-mail addresses from their members, these services are provided only if the user requests them.
One additional known e-mail filtering techniques are based upon an inclusion list, such that e-mail received from any source other than one listed in the inclusion list is discarded as junk. However, these methods require the user or the service provider to continually update the inclusion list manually because, like viruses, spam is constantly being modified to bypass static filters. If the inclusion list is not updated regularly, the list will quickly become outdated, resulting in the exclusion of desired e-mail messages from new sources and the continued inclusion of spam from old sources.
The Assignee of the present invention has developed improved techniques for filtering e-mail. Some of these techniques are described in related U.S. patent applications entitled U
The present application sets forth additional techniques for filtering e-mail. What is needed is an improved system and method for dynamically updating e-mail filter technology to meet the threat posed by ever-changing varieties of spam. In addition, what is needed is a system and method for filtering e-mail which can be easily implemented by the typical e-mail user. What is also needed is an anti-spam system and method which can be applied without the need to upgrade computer systems and networks currently available.
Disclosed is a server having a processor and a memory coupled to the processor, the memory having stored therein sequences of instructions which, when executed by the processor, cause the processor to perform the steps of: (1) receiving a request to retrieve e-mail messages on behalf of a client; (2) retrieving one or more e-mail messages from a mail server on behalf of the client; and (3) filtering the e-mail messages based on one or more rules to produce one or more filtered e-mail messages.
Also disclosed is a first server having a processor and a memory (e.g., machine-readable memory medium and computer-readable memory media) coupled to the processor, the memory having stored therein sequences of instructions which, when executed by the processor, cause the processor to perform the steps of: retrieving messages from a second server on behalf of a client; sorting messages into two or more groups based on one or more rules; and forwarding messages sorted into one of the groups to the client.
Also disclosed is an e-mail filter comprising an application programming interface (“API”) and a plurality of rule handling filter modules adapted to interface with the API.
A better understanding of the present invention can be obtained from the following detailed description in conjunction with the following drawings, in which:
Depending on the system configuration, however, one of ordinary skill in the art will readily recognize from the following discussion that different types of mail servers, clients and software could be employed without departing from the underlying principles and scope of the present invention. Accordingly, while the embodiment discussed below uses an Internet communication channel for communication between clients 110, 120 and mail server 130, numerous other communication schemes such as a direct connection to mail server 130, etc., could be implemented as well.
The following is a general description of how client 120 sends e-mail to client 110. Each client 110, 120 is capable of executing e-mail application programs, generally illustrated in
If client 110 is connected to the same LAN as mail server 130, client 110's user agent 115 may continually check mailbox 140 (e.g., every 10 minutes) for new e-mail messages. Alternatively, client 110 may need to connect to mail server 130 by dialing out over a telephone line using a modem (i.e., either by dialing directly to mail server 130 or dialing out and connecting to mail server 130 over the Internet). This would typically be the case for a home computer user who has an account with an Internet Service Provider or online service, or a user who must dial out to connect to his corporate LAN.
For the purposes of the following discussion, when client 110 dials out to connect to mail server 130 to check mailbox 140, it communicates with mail server 130 using “Post Office Protocol 3” (hereinafter “POP3”). It should be noted, however, that different, standard and proprietary mail protocols such as the Interactive Mail Access Protocol (hereinafter “IMAP”), the Distributed Mail System Protocol (hereinafter “DMSP”), X.400, and Lotus Notes may be implemented without departing from the scope of the present invention.
In the transaction state user agent 115 sends mail server 130 a LIST command at 420. If there is new mail in mailbox 140, server 130 responds to the LIST command with the number of new messages —100 in the example (at 425). User agent 115 then retrieves the e-mail messages one at a time using a RETR command (starting at 430). Finally, after user agent has received all 100 messages, user agent 115 issues the QUIT command at 460 and the POP3 connection is terminated at 465. This final state is the referred to as the update state. During the update state mail server 130 updates mailbox 140 by removing all messages marked as deleted and releases its lock on mailbox 140.
If client 120 previously sent an e-mail message to client 110 as described above, client 120's e-mail will be one of the 100 messages retrieved from mail server 130. However, if client 120 is a spammer, sending client 110 an unwanted solicitation, client 110 will have wasted hard drive space, line access time/cost and personal time downloading the e-mail message. This inefficiency and waste becomes more significant if a substantial percentage of all 100 e-mail messages stored in mailbox 140 are spam. Moreover, if client 110 is communicating with mail server 130 through a modem dial-up connection (rather than over a LAN), the time and cost associated with downloading unwanted content is even more significant given the current speed limitations of modems (currently less than 56,000 bits/sec) and the access cost of ISPs. Accordingly, it would be beneficial to provide an e-mail filter to remove spam from mailbox 140 before time is wasted downloading and storing it.
As set forth in co-pending application entitled “A
Spam collected from various spam probes is then analyzed and rules 210 are established based on this analysis. For example, in one embodiment, source header data from incoming e-mail is analyzed. If the network address contained in the source header is identified as the network address of a known spammer, a rule will be established to filter all incoming e-mail from this network address into the spam storage area 230. Rules 210 may also be established which identify spam based on a mathematical signature (e.g., a checksum) of the spam e-mail body (or portions of the e-mail body). Any incoming message which contains the identified signature will subsequently be forwarded into the spam storage area 230. Rules 210 based on keywords in the subject or body of spam e-mail may also be established. For example, all e-mails containing the two words “sex” and “free” may be identified as spam and filtered. In one embodiment, mail marked as potential spam is transferred to a control center where it is inspected (e.g., by a computer technician) before being sent to the spam storage area 230.
In addition, a dynamically updated inclusion list may be generated at server 130. The inclusion list initially screens e-mail header fields of all incoming e-mail messages. All messages that appear on the list are passed through to mailbox 140. E-mail received from sources other than those on the inclusion list are processed by additional anti-spam rules 210 as described herein. In one embodiment, the inclusion list is simply one of the plurality of rules 210 acting on filter module. In another embodiment, the inclusion list is a separate software module executing on server 130. The end result is the same regardless of which embodiment is used.
Because spam is continuously being created and modified by spammers, a system is needed in which rules for filtering spam can be easily updated. In addition, an e-mail filtering system is needed which is easily modified to suit the unique preferences of individual e-mail users and which can be easily adapted to run on proprietary e-mail systems. To address these and other issues an improved filter module 700 is illustrated in
Referring again to
Referring again to
It should be noted that the rule-based filtering method and apparatus described herein can be implemented at virtually any point along the e-mail transmission path from client 120 to server 130 to client 110. For example, on mail server 130 an embodiment of filter module 700 can be located at a point in server 130 where e-mail messages are transmitted to client 110 or at a point where the messages are received (as shown in
In addition, Client 110 can itself contain an embodiment of filter module 700. Therefore, Client 110 can apply filter module 700 periodically as described above (with reference to server 130) or, alternatively, can apply filter module 700 to all incoming e-mail messages. Filter module 700 can also be applied within a mail relay residing on network 100. In sum, filter module 700 may be applied at any node through which e-mail messages are transmitted.
In another embodiment, the rules 210 used by filter module 220 are continually monitored by server 130. If a rule has not been used to filter spam for a predetermined length of time (e.g., a month) that rule may be moved from an active to an inactive state and no longer applied to filter 220 (unless the type of spam for which it was created reappears). This type of rule-aging system is useful given the fact that older types of spam are continually replaced with new types. If not removed as described above, the number of outdated rules would build up an unmanageable level and filter module 220 may become inefficient at removing spam (applying numerous obsolete rules to the incoming e-mail stream).
Additionally, in one embodiment, rules applied to filter module 220 are weighted based on one or more variables, including but not limited to spam content, probability of positive spam identification, and frequency of use. For example, a rule which is geared towards screening e-mail messages containing sexual content (e.g., in a home where children use the computer) which filters e-mail based on the keywords “sex” and “free” may be given a weight value of 10 on a scale from 1 to 10. However, a rule which screens e-mail based merely on the keyword “free” may be weighted with, e.g., a weight value of 2.
E-mail messages can also be weighted based on the probability that the filter has correctly identified the filtered e-mail message as spam. For example, a positive spam identification under a mathematical analysis (e.g., a checksum) will generally be accurate. Thus, an e-mail message identified as spam based on a mathematical calculation will be given a higher weighted value than, for example, a keyword identification.
In addition, the rules applied to filter module 220 may be additive such that it may take several rules to fire to allow filter module 220 to decide that a message is spam. In such an embodiment, the relative weights of the rules which identify the message as spam can be added together to establish a cumulative weighted value. Moreover, a filter module 220 in this embodiment may be configured based on how aggressively filter module 220 should screen e-mail messages. For example, filter module 220 may configured to screen all e-mail messages with a cumulative weighted value of 6.
In one embodiment, rules have two weights associated with them: (1) a spam weight—this weight indicates the certainty that if a rule fires successfully against a message, the message is Spam and (2) a priority weight—this weight indicates the frequency of use and most recent use of a rule. In addition, in this embodiment two different priority weights may be associated with a rule, a global priority weight and a local priority weight.
Spam Weight is an arbitrary weight chosen by a rule designer based on internal heuristics to reflect the certainty that the rule will correctly identify Spam. General rules, for example, that don't definitively identify Spam but are designed to detect typical Spammer practices, may be provided lower weights that rules which target a specific type of spam. Thus, because of their low assigned weight, general rules may need to fire along with other rules (general or not) for spam to be filtered by filter module 700 (or 310, 850).
In this embodiment, new rules are assigned the highest Priority Weight. As statistics on rule usage are compiled, the priority weight may change to reflect how often the rule is used. For example, if a new rule is generated and isn't used within a configured period of time, the rule's priority weight will decrease. Similarly, if a rule has been used recently or frequently within a designated interval, the priority weight will be increased. Global priority weight reflects a rule's usage at all known filter modules across a network (e.g., network 100) whereas local priority weight reflects a rule's usage at a specific node (e.g., filter module 310).
The spam weight and priority weight of a set of rules may be combined mathematically to prioritize rules within the set. In one embodiment, the spam weight and priority weight of each rule within the set are multiplied together and the results of the multiplications are ordered sequentially. Those rules associated with numerically higher results (i.e., which are used more frequently and are more likely to correctly identify spam) are assigned higher priorities within the set.
Moreover, rules assigned higher priorities will tend to be executed earlier against messages applied to the filter module 310, resulting in a more efficient spam filtration system. In other words, spam will be identified more quickly because rules with a higher usage frequency and detection accuracv will be implemented first. In addition, both the Global and Local Priority weights may be multiplied with a rule's Spam Weight as described above to evaluate new orderings within the set of rules.
One problem associated with the implementation of filter modules 220 and 700 as described above is that in order for client 110 to receive the benefit of a server-side filter module 220, it must be set up and maintained by the administrator of server 130. For example, if the user of client 110 belongs to an online service—say, for example, Netcom®—that user will not be able to implement a server-side filter if Netcom does not provide one. Moreover, assuming arguendo that Netcom offers a server-side e-mail filter, the user of client 110 will be limited to the type of filter offered. If the filter is merely a static filter, the user will not be able to significantly tailor it to his specific preferences.
As described above, user agent 115 executed on client 110 allows client 110 to check for e-mail in mailbox 140 on mail server 130. E-mail user agents known in the art include Microsoft Outlook®, Lotus cc:mail®, Lotus Notes®, Novell Groupwise®, Eudora®, and Netscape Communicator®. In order for any one of these user agents to retrieve mail from mailbox 140 on mail server 130, the user agent must be configured with the correct network address (e.g., “mail.isp.com” as shown in
Thus, referring now to
Referring now to
In one embodiment, client 115 may send a request to proxy server 300 to view messages which have been filtered into spam storage module 340 (or into a virus storage module). Client 110 can then insure that legitimate messages have not been inadvertently filtered by filter module 310. As such, in this embodiment, proxy server 300 will store spam in spam storage module 340 for a predetermined length of time. Alternatively, client 115 may be allocated a predetermined amount of memory in spam storage module 340. When this memory has been filled up with spam, the oldest spam messages will be forced out to make room for new spam messages. In addition, client 110 may be periodically notified that spam has been filtered into spam storage module 340. Once client 110 has checked spam storage area 340 to view messages which have been filtered, the user may choose to redirect one or more filtered messages back into mail storage module 330 (illustrated as signal 335 of
Proxy server 300 continues to retrieve messages from mail server 130 one at a time (signals 612 et seq.) until the last message—e.g., message 100 (not illustrated in FIG. 6)—has been retrieved. Filter module 310 applies its set of rules 310 to identify spam, computer viruses or other types of e-mail messages and sorts the retrieved messages accordingly. In the embodiment illustrated in
At some predetermined point in time (represented by the dashed line 624 in
In the signal diagram shown in
Following timeout period 624, however, proxy server 300 continues retrieving and processing the 96 e-mail messages remaining on mail server 130 as described above. At some point after proxy server has completed processing the remaining 96 messages, user agent 115 on client 110 sends another LIST command to proxy server 300, thereby requesting a list of current e-mail messages (signal 660). At this point, proxy server 300 has processed the 96 additional e-mail messages following timeout period 624. For the purposes of the following discussion it will be assumed that ½ of the 96 messages processed after timeout period 624 were identified as spam by filter module 310. Thus, by the time client 130 sends LIST signal 660, filter module 310 has transferred 48 messages into spam storage module 340 and 48 messages into mail storage module 330.
Upon receiving the LIST signal 660 from client 110, proxy server 300 sends a second list signal 665 to mail server 130 to determine whether additional e-mail messages addressed to client have been received at mailbox 140 since proxy server 300 and mail server 130 last communicated. In the example illustrated in
Thus, in response to client 110's LIST command 660, proxy server 300 sends a response signal 690 indicating that 49 e-mail messages are currently available, all of which are legitimate e-mail messages. Client 110 will subsequently proceed to retrieve each of the 49 remaining messages from mail storage module 330 on proxy server 300 (although these signals are not illustrated in the signal chart of
In one embodiment, proxy server 300 will store cleaned e-mail messages in mail storage module 330 even after the messages have been downloaded by client 110. This is done so that the messages do not have to be filtered a second time if client 110 need to access the messages again (e.g., if the user of client 110 attempts to check e-mail from a different computer).
Referring now to
In another embodiment, once a particular e-mail message has been identified as spam, only a single copy of that message is stored in spam storage module 340, regardless of how many different e-mail users the message is addressed to. This technique of caching only one copy of a spam message in spam storage module 340 allows proxy server 300 to conserve significant space in spam storage area 340 given the fact that spam is commonly distributed to thousands of e-mail users at a time.
One of ordinary skill in the art will readily recognize from the following discussion that alternative embodiments of the structures and methods illustrated herein may be employed without departing from the principles of the invention. Throughout this detailed description, numerous specific details are set forth such as specific mail protocols (i.e., POP3) and filter applications (e.g., spam removal) in order to provide a thorough understanding of the present invention. It will be appreciated by one having ordinary skill in the art, however, that the present invention may be practiced without such specific details. In other instances, well known software-implemented communication techniques have not been described in detail in order to avoid obscuring the subject matter of the present invention. The invention should, therefore, be measured in terms of the claims which follow.
This application is a continuation of U.S. patent application Ser. No. 09/224,378, filed Dec. 31, 1998 now U.S. Pat. No. 6,654,787.
Number | Name | Date | Kind |
---|---|---|---|
5377354 | Scannell et al. | Dec 1994 | A |
5438433 | Reifman et al. | Aug 1995 | A |
5557789 | Mase et al. | Sep 1996 | A |
5619648 | Canale et al. | Apr 1997 | A |
5634005 | Matsuo | May 1997 | A |
5649182 | Reitz | Jul 1997 | A |
5675507 | Bobo, II | Oct 1997 | A |
5678041 | Baker et al. | Oct 1997 | A |
5696898 | Baker et al. | Dec 1997 | A |
5734826 | Olnowich et al. | Mar 1998 | A |
5790789 | Suarez | Aug 1998 | A |
5796948 | Cohen | Aug 1998 | A |
5809242 | Shaw et al. | Sep 1998 | A |
5822527 | Post | Oct 1998 | A |
5826022 | Nielsen | Oct 1998 | A |
5835087 | Herz et al. | Nov 1998 | A |
5845263 | Camaisa et al. | Dec 1998 | A |
5862325 | Reed et al. | Jan 1999 | A |
5864684 | Nielsen | Jan 1999 | A |
5870548 | Nielsen | Feb 1999 | A |
5874955 | Rogowitz et al. | Feb 1999 | A |
5884033 | Duvall et al. | Mar 1999 | A |
5889943 | Ji et al. | Mar 1999 | A |
5905863 | Knowles et al. | May 1999 | A |
5930479 | Hall | Jul 1999 | A |
5968117 | Schuetze | Oct 1999 | A |
5978837 | Foldare et al. | Nov 1999 | A |
5999932 | Paul | Dec 1999 | A |
5999967 | Sundsted | Dec 1999 | A |
6023700 | Owens et al. | Feb 2000 | A |
6023723 | McCormick et al. | Feb 2000 | A |
6052709 | Paul | Apr 2000 | A |
6073165 | Narashimhan et al. | Jun 2000 | A |
6112227 | Heiner | Aug 2000 | A |
6146026 | Ushiku | Nov 2000 | A |
6157630 | Adler et al. | Dec 2000 | A |
6158031 | Mack et al. | Dec 2000 | A |
6161130 | Horvitz et al. | Dec 2000 | A |
6173364 | Zenchelsky et al. | Jan 2001 | B1 |
6182118 | Finney et al. | Jan 2001 | B1 |
6189026 | Birrell et al. | Feb 2001 | B1 |
6192360 | Dumais et al. | Feb 2001 | B1 |
6195686 | Moon et al. | Feb 2001 | B1 |
6199102 | Cobb | Mar 2001 | B1 |
6199103 | Sakaguchi et al. | Mar 2001 | B1 |
6216165 | Woltz et al. | Apr 2001 | B1 |
6226630 | Billmers | May 2001 | B1 |
6230156 | Hussey | May 2001 | B1 |
6314454 | Wang et al. | Nov 2001 | B1 |
6327610 | Uchida | Dec 2001 | B2 |
6334140 | Kawamata | Dec 2001 | B1 |
6360254 | Linden et al. | Mar 2002 | B1 |
6377949 | Gilmour | Apr 2002 | B1 |
6411947 | Rice et al. | Jun 2002 | B1 |
6421709 | McCormick et al. | Jul 2002 | B1 |
6438608 | Biliris et al. | Aug 2002 | B2 |
6505237 | Beyda et al. | Jan 2003 | B2 |
6546416 | Kirsch | Apr 2003 | B1 |
6549957 | Hanson et al. | Apr 2003 | B1 |
6571275 | Dong et al. | May 2003 | B1 |
6718321 | Birrell et al. | Apr 2004 | B2 |
6732157 | Gordon et al. | May 2004 | B1 |
6757713 | Ogilvie et al. | Jun 2004 | B1 |
20020007301 | Reuning | Jan 2002 | A1 |
20030023722 | Vinberg | Jan 2003 | A1 |
20050081059 | Bandini et al. | Apr 2005 | A1 |
Number | Date | Country |
---|---|---|
41020649 | Sep 1998 | JP |
WO 9837680 | Aug 1998 | WO |
Number | Date | Country | |
---|---|---|---|
Parent | 09224378 | Dec 1998 | US |
Child | 10117065 | US |