Data processing: database and file management or data structures – Database design – Data structure types
Reexamination Certificate
1998-03-30
2001-08-21
Black, Thomas (Department: 2171)
Data processing: database and file management or data structures
Database design
Data structure types
C707S793000, C707S793000
Reexamination Certificate
active
06278996
ABSTRACT:
COPYRIGHT NOTICE
A portion of the disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the xerographic reproduction by anyone of the patent document or the patent disclosure in exactly the form it appears in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights whatsoever.
FIELD OF INVENTION
The present invention is directed to a method and apparatus for automatically recognizing and responding to intent in message text, particularly as used in a large-scale automatic message response system to examine input text and respond appropriately with output text relevant to the writer's intent.
BACKGROUND OF THE INVENTION
Electronic textual interchange is pivotal to businesses, academic institutions and private correspondence, and those recipients of electronic messages are being inundated with messages because of the increased popularity of electronic messaging and the migration to that medium from other messaging media, such as telephone calls, telegrams, physical (postal) mail, newspapers and magazines. This increase is, in part, due to the ease of access, the low cost of electronic exchange and because electronic messaging can be sent asynchronously and received quickly. While this medium allows for large amounts of information exchange, providing quick, relevant, and consistent responses to messages becomes increasingly difficult as the number of messages increases.
To alleviate these problems, some automated response systems have been proposed, with only limited success. In general, an automated response system processes an input message, attempts to “understand” what the writer is saying in the message, formulates an appropriate response and routes that response to the sender. Since many of the input messages are free-form text, a natural language processor and reasoning system is often used to “understand” what the input message is conveying, i.e., the intent of the sender. The term “understand” means the identification of information which corresponds to analogous situations previously identified by humans.
A typical message understanding system identifies words and other patterns in text, combining algorithmic and empirical methods to draw comparisons to known situations. Once the message is understood, a text generation system might be used to generate the response message text so that the entire communication response process is automated. The message understanding system might also include a classifier which understands the content of a message and routes or categorizes the message based on its content.
A number of approaches have been developed for automating text understanding and response. One approach to text understanding is to codify rules of natural language grammar. This approach is problematic because the rules of grammar are complicated, as well as incomplete, so systems based on them are difficult to produce and maintain. Another approach is to use statistical analysis of words within a text corpus, as is used in neural networks. Statistical analysis systems have the advantage that they are less difficult to maintain, but have the disadvantage that they are of limited usefulness where large amounts of relevant training data are not available.
Another approach to the problem of text understanding is to constrain and simplify the input message text. One way to do this is to have the writer of the input text use forms with limited choices and constrained syntax. Computer languages, with rigid and constrained syntax, are examples of how a user can communicate precisely with computers. While this approach greatly reduces the complexity of the process of automatic interpretation, it also requires prolonged, specialized user training.
Whether the messages are constrained or free-form, a message response system must first understand the input message text before it can process and respond to the message. One way to simplify the understanding and response process is to require manual intervention. The manual intervention approach has a number of drawbacks, since the process takes time and labor, requires training for reviewers and might result in inconsistency in responses from different reviewers. A manual review process might have the reviewer read a message and input a set of keywords, a classification, and/or a response. One approach to automating the manual process is to extract keywords from the input message and use them to compose a template response text. The drawback to this approach is that the keywords are chosen indiscriminately and are possibly irrelevant to the central intent of the input message text.
A problem with all the foregoing systems, with the exception of some applications of neural networks, is that system maintenance requires many of the same specialized skills required for original application development. Neural network systems can avoid that requirement, but they are limited to environments having a considerable amount of training data, a requirement which increases commensurately with the desired precision of classification. These systems classify without respect to meaning, i.e., irrelevant words are ineffectively separated from those central to intentionality.
Therefore, what is needed is a set of tools in a message text understanding and response system which reduces the requirement for specialized skills to produce and maintain domain specific applications.
SUMMARY OF THE INVENTION
An improved message text understanding and response system is provided by virtue of the present invention. In one embodiment of an understanding and response system, the system isolates the writer's critical words which relate to prototypical statements stored in the system. These in turn are mapped to models which represent understanding of the writer's possible intents. The writer's intents are then mapped to prototypical response actions as defined by the system operators. Thus the system models typical message writer requests and answers by selecting among typical responses.
To accomplish this, a tokenizing identifier accepts input message text and analyzes the text against a lexicon specific to the system operator's domain, in order to generate a structured data representation of input message text. The structured data and message text is fed to an intent identifier, which uses the structured text and a knowledge base to infer the intent of the writer of the input message. Intent determination can be supplemented by verifying and augmenting input message data against databases outside the system. The identified intents, and possibly the structured data and the message text itself, are passed to a response formulator, which assembles a set of actions according to a set of business rules to formulate a response. The assembled potential actions are evaluated and executed. Examples of potential actions include responding to the writer, giving recommendations for the next step in processing, possibly outside the system, routing the message to a third party for automatic or manual response, or invoking an external system.
The response message text, as well as the input message text, can be transported between the system and the writer of the message using conventional message transport techniques. The input message text can be electronic mail and may include text and other computer formats such as documents, graphic representations, formatted fields, and computer control characters. If the input message contains special formats of structured data, those are flagged for special processing which uses the additional information provided by the structure. The response actions can include: responding to the writer, creating a message summary for later review by system operators, forwarding the message for manual review, output the message to printers and facsimile machines, adding items to workflow systems, etc.
A further understanding of the nature and advantages of the inventions herein may be realized by refe
Aleksandrovsky Boris
Buedel Doug
Greif Jeff
Richardson Keith D.
Black Thomas
Brightware, Inc.
Merchant & Gould P.C.
Rones Charles L.
LandOfFree
System and method for message process and response does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with System and method for message process and response, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System and method for message process and response will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2441594