Data processing: measuring – calibrating – or testing – Measurement system – Time duration or rate
Reexamination Certificate
2001-11-13
2004-08-17
Bhat, Aditya S. (Department: 2863)
Data processing: measuring, calibrating, or testing
Measurement system
Time duration or rate
C709S202000, C709S203000, C709S206000, C709S207000, C715S252000, C715S252000, C707S793000
Reexamination Certificate
active
06778941
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The invention concerns message filtering. Message filtering is the process of determining whether or not a message, one of many in a stream, should be passed to a user. Typically, only a small fraction of the messages in the stream are considered important by the user.
2. Description of the Prior Art
Human users of electronic information systems are faced with finding useful information in the midst of ever-increasing amounts of irrelevant inputs. An illustrative situation is the screening of relevant from irrelevant email messages. Canale et al. have proposed reducing the amount of junk e-mail received (U.S. Pat. No. 5,619,648). Here, each sender of an email system provides a recipient description field as part of a message. The user's mail filter has a model describing user preferences that is accessible by the email sender. Messages with recipient descriptions matching the user preferences are passed. Drawbacks to this approach include the overhead to the sender in the specification of recipient descriptions, and situations where unimportant messages are passed because of a match in the recipient/preference pair. Cobb (U.S. Pat. No. 6,199,102) describes a method for filtering unsolicited email by way of an automated challenge and response scheme. A challenge is initiated when a sender is not on an approved sender list at the recipient site. Two problems with this approach are that it requires users to specify who they will receive messages from, implying that important information comes only from currently known senders, and that unimportant messages from approved senders will be passed to the user.
Rather than filtering messages based solely on sender/receiver characteristic matching, text classification approaches estimate “importance” by analyzing word content of incoming messages. Dumais et al. describe a text classifier (U.S. Pat. No. 6,192,360). In this method, features are computed from the text object. A training phase provides a weight vector and a monotonic function essentially defining a classifier. The weight vector may be found by means of support vector machine training and the monotonic function may be determined by optimization means. Given the weight vector and monotonic function, messages are classified by applying the feature vector from that message to the weight vector and monotonic function. When the result from the monotonic function is above a threshold, the message is passed to the user.
A similar method for text classification is given in the specification of Punch et al. (U.S. Pat. No. 5,924,105). Feature vectors are computed from the text of messages. The feature vectors are input to trained classifiers, resulting in a score for that message. The message is then passed to the user or not based on the relation of the score to a threshold.
Both of these text classification methods rely solely on the text of the incoming message to determine importance. No information outside of message text is used to augment the classification. For example, Rhodes et al., in U.S. Pat. No. 6,236,768, describe a method and system for substantially continuous retrieval of indexed documents related to the user's current context. Indexing involves analyzing the document and storing a representation of that document that is amenable to fast search methods. User context is obtained by monitoring various aspects of the user's computational and physical environment. The “various aspects” are referred to as meta-information, where meta-information is information about the information. Furthermore, meta-information may be explicitly entered by the user, or automatically tagged with context information when the document is created. Examples given by Rhodes include room number, GPS location, time of day, day of the week, people present and so on. Even in traditional desktop environments, meta-information related to time and general subject can provide cues bearing on the relevance. Relevance estimates are improved over methods using text similarity alone by the inclusion of the meta-information. While it is clear that the meta-information is useful for finding relevant documents in the indexed database, the creation of the indexed database can be time consuming and is typically created by running an application overnight. Thus, the invention of Rhodes et al. is useful for finding indexed documents from a previous day that are relevant to a user's current context.
It is clear from the above discussion that it is advantageous for a message filtering method and system to use both message textual content and message attribute information. This is provided by the following invention without a database of indexed documents. Rather, a message arrives, and a substantially instant determination is made regarding whether or not to pass the message to a user by joint analysis of the message's body and attribute information.
SUMMARY OF THE INVENTION
This invention reduces the amount of insignificant message traffic received by the user of an electronic information system by jointly exploiting the body and attributes of incoming messages. Message body refers to the portion of the message which the sender intends the recipient to read as the primary source of information. The message attributes provide information about the message. Attribute information includes the sender of the message, the corporate or academic affiliation(s) of the sender, and so on. A feature vector is computed from the message body and attribute information. The feature vector is then provided to a classification step, where a message discriminant score is computed. The message is passed to or withheld from the user based on the value of the discriminant score. In another embodiment, the context of the user's system is used to preferentially pass messages related to the user's current interests. In yet another embodiment, message body and attributes are used to anticipate significant events in a time series, such as streaming financial data.
REFERENCES:
patent: 5619648 (1997-04-01), Canale et al.
patent: 5905863 (1999-05-01), Knowles et al.
patent: 5924105 (1999-07-01), Punch, III et al.
patent: 6052709 (2000-04-01), Paul
patent: 6073133 (2000-06-01), Chrabaszez
patent: 6073142 (2000-06-01), Geiger et al.
patent: 6192360 (2001-02-01), Dumais et al.
patent: 6199102 (2001-03-01), Cobb
patent: 6202058 (2001-03-01), Rose et al.
patent: 6236768 (2001-05-01), Rhodes et al.
patent: 6356935 (2002-03-01), Gibbs
patent: 6393465 (2002-05-01), Leeds
patent: 6442589 (2002-08-01), Takahashi et al.
patent: 6453327 (2002-09-01), Nielsen
patent: 2002/0016824 (2002-02-01), Leeds
patent: 2002/0198950 (2002-12-01), Leeds
patent: 1 087 315 (2001-03-01), None
patent: WO 01/27828 (2001-04-01), None
Amburn Philip
Kabrisky Matthew
Rogers Steven K.
Worrell Steven W.
Bhat Aditya S.
Qualia Computing, Inc.
Stevens & Showalter LLP
LandOfFree
Message and user attributes in a message filtering method... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Message and user attributes in a message filtering method..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Message and user attributes in a message filtering method... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3296993