International Business Machines Corporation

Frank J Oles

David E Johnson

Tong Zhang

Rudolf Maarten Bolle

Norman C Haas

Kathryn K Lam

Thilo W Goetz

A method (and structure) of extracting information from text, includes parsing an input sample of text to form a parse tree and using user inputs to define a machine-labeled learning pattern from the parse tree.

Method and system for extracting information from unstructured text using symbolic machine learning

This invention is a business system and method to perform categorization (classification) of multimedia items and to make business decisions based on the categorization of the item. The multimedia items are comprised of a multitude of disparate information sources, in particular, visual information and textual information. Classifiers are induced based on combining textual and visual feature vectors. Textual features are the traditional ones. Visual features include, but are not limited to, color properties of key intervals and motion properties of key intervals. The visual feature vectors are determined in such a fashion that the vectors are sparse. The text and the visual representation vectors are combined in a systematic and coherent fashion. This vector representation of a media item lends itself to well-established learning techniques and can be used for multimedia item categorization. The resulting business system, subject of this invention, can be used for many purposes. An example here are enforcement of copyright, trademark, intellectual property, parental guidance and common decency restrictions. Other uses are multimedia item classifier to determine routing of incoming items or building user profiles based on user multimedia preferences.

Business method and apparatus for employing induced multimedia classifiers based on unified representation of features reflecting disparate modalities

Supervised learning is used to develop a computer program that can automatically route or respond to electronic communications assuming the existence of an appropriate formal scheme for categorizing incoming electronic communications. A method is described by which such a categorization scheme for electronic communications can be constructed. The method is based on an analysis of factors having an impact on the categorization scheme from both the business domain and the technology domain. The problem solved by this method is a new one that is only now emerging as automated methods of routing communications based on supervised learning are becoming feasible. Among other uses, this method may be practically employed as a disciplined way of carrying out consulting engagements that call for setting up and maintaining categorization schemes for electronic communications.

Methodology for creating and maintaining a scheme for categorizing electronic communications

This invention is a system and method to perform categorization (classification) of multimedia items. These items are comprised of a multitude of disparate information sources, in particular, visual information and textual information. Classifiers are induced based on combining textual and visual feature vectors. Textual features are the traditional ones, such as, word count vectors. Visual features include, but are not limited to, color properties of key intervals and motion properties of key intervals. The visual feature vectors are determined in such a fashion that the vectors are sparse. The vector components are features such as the absence or presence of the color green in spatial regions and the absence or the amount of visual flow in spatial regions of the media items. The text and the visual representation vectors are combined in a systematic and coherent fashion. This vector representation of a media item lends itself to well-established learning techniques. The resulting system, subject of this invention, categorizes (or classifies) media items based both on textual features and visual features.

Method and apparatus for inducing classifiers for multimedia based on unified representation of features reflecting disparate modalities

A method and apparatus make keyword selection and/or weighting as a function of a session history of user input in order to answer queries submitted by the user to a computer system by providing answers based on stored documents. The aim is to find the best answers by matching stored natural language documents both to the most recent query and to the most latest query in a context that captures the recent history interaction. To do this, answers are matched against a set of keywords extracted from the most recent query as well as a set of keywords extracted from those queries received since the last topic switch was detected. A central feature of the method is for the computer system implementing this method to maintain a session history for each user session history. Keywords are extracted from each query by a system implementing this method. A graded keyword list is a list of keywords paired with ages, which are indicators of how long ago in the session the user employed this keyword in a query. Graded keyword lists are maintained in the session history so the system can assign weights to keywords, with more recently received keywords being assigned higher weights than keywords with comparatively greater ages. The weights assigned to keywords are used in computing scores that indicate closeness of match of a document to a list of keywords.

Session-history-based recency-biased natural language document search

A method for iteratively drilling-down on a user's textual free-form natural language query uses a session history to interpret successive queries in the context of previous queries on a topic or topics and to detect an implicit switch in topic. By maintaining a session history of the user's free-form natural language input and by automatically determining whether there is a topic or context switch, the search process is substantially simplified and is more effective; that is, more accurate answers to a user's queries are found faster. In addition, as the system operates on free-form natural language input, automatically constructing the actual search expressions, the complexity of constructing successive search expressions is obviated. If the system determines the user is, according to its session history and tests, asking successive questions within a given topic or context, the system keeps searching within a previously determined given set of previous responses on that context or topic. This effectively narrows the documents found allowing the user to quickly and accurately find just the documents of interest. If the system determines the user has implicitly changed context or topic, based on its session history and tests, it searches all the information at its disposal; i.e., all of the collections of documents.

Automatic topic identification and switch for natural language search of textual document collections

A method to automatically categorize messages or documents containing text. The method of solution fits in the general framework of supervised learning, in which a rule or rules for categorizing data is automatically constructed by a computer on the basis of training data that has been labeled beforehand. More specifically, the method involves the construction of a linear separator: training data is used to construct for each category a weight vector w and a threshold t, and the decision of whether a hitherto unseen document d is in the category will depend on the outcome of the test w x&gE;t, where x is a vector derived from the document d. The method also uses a set L of features selected from the training data in order to construct the numerical vector representation x of a document. The preferred method uses an algorithm based on Gauss-Seidel iteration to determine the weight factor w that is determined by a regularized convex optimization problem derived from the principle of minimizing modified training error.

Text categorizers based on regularizing adaptations of the problem of computing linear separators

A computerized system responds, not just to a single query issued by a user, but to a query in the context of a dialog with the user. The system, which is referred to as an interactive automated response system, consists of three principal components or subsystems, which are a text categorizer that assigns categories to text extracted from a dialog, a search system whose purpose is to match text extracted from a dialog with answers, and a dialog manager whose purpose is to maintain a user's session history, to decide what text should be sent to the text categorizer and to the search system, to make use of a partially ordered category scheme to categorize each stage of the dialog based on the results returned by the other components, and to use the results of dialog categorization, as well as the results returned by the other components to create suitable responses to the user's query in the context of his or her earlier queries.

Interactive automated response system

A method to automatically categorize messages or documents containing text. The method of solution fits in the general framework of supervised learning, in which a rule or rules for categorizing data is automatically constructed by a computer on the basis of training data that has beforehand been categorized, i.e., each training data item has been labeled with the categories to which it belongs. More specifically, the method for rule induction involves the novel combination of (1) inducing from the training data a decision tree for each category, (2) automated construction from each decision tree of a simplified symbolic rule set that is logically equivalent overall to the decision tree, and which is to be used for categorization instead of the decision tree, and (3) determination of a confidence level for each rule. The method covers both decision-tree-based symbolic rule induction and the use for the purpose of document categorization of rules in the logical format of those generated by the rule induction procedure described herein.

Decision-tree-based symbolic rule induction system for text categorization

A method and apparatus for interacting with incoming text information, e.g., a user query. The method of the present invention categorizes the incoming text information, and may also provide associated confidence levels with the categorization feature. The categorized information is used to query a database having associated category information with associated threshold values. The confidence levels may also be used when querying the database. The categorized information and corresponding confidence levels are then compared to the threshold value. If the categorized information and corresponding confidence levels equals or exceeds the threshold value then a response is provided by the system and method of the present invention. If the categorized information and corresponding confidence levels does not equal or exceed the threshold value then the system of the present invention will request further information of the user, and the method will repeat until (i) the categorized information and corresponding confidence levels equals or exceeds the threshold value or (ii) the system of the present invention times out.

Growing community of inventors

Peekskill, NY, United States of America

Frank J Oles