Justsystems Evans Research, Inc.

Claritech Corporation

Clairvoyance Corporation

Norbert Roma

David A Evans

James G Shanahan

A computer-readable medium comprises data structure for providing information about levels of similarity between pairs of N documents. The data structure comprises a plurality of entries of similarity values representing levels of similarity for a plurality of pairs of the documents. Each of the similarity values represents a level of similarity of one document of a given pair relative to the other document of the given pair. The similarity value of each entry is greater than a threshold similarity value that is greater than zero. The plurality of similarity-value entries are fewer than N&#x2212;N in number if the similarity values are asymmetric with regard to document pairing, and the plurality of similarity-value entries are fewer than in number if the similarity values are symmetric with regard to document pairing. A method and apparatus for generating the data structure are described.

Method and apparatus for constructing a compact similarity structure and for using the same in analyzing document relevance

An information need can be modeled by a binary classifier such as support vector machine (SVM). SVMs can exhibit very conservative precision oriented behavior when modeling information needs. This conservative behavior can be overcome by adjusting the position of the hyperplane, the geometric representation of a SVM. The present invention describes a couple of automatic techniques for adjusting the position of an SVM model based upon a beta-gamma thresholding procedure, cross fold validation and retrofitting. This adjustment technique can also be applied to other types of learning strategies.

Method and apparatus for adjusting the model threshold of a support vector machine for text classification and filtering

The delivery ratio of r (which is a fraction between 0 and 1) partitions a stream of documents into a section of top scoring r-fraction of documents and the remainder. This way a set of successively bigger delivery ratios, r, r, r, . . . sections the stream into tiers. Any given document is assigned to a tier according to how many delivery ratio thresholds it matched or surpassed and how many it failed to reach. This creates a scoring structure which reflects the specificity of the document with respect to a profile in terms of density of relevant documents in the stream. In other words, a document in the ktier is such that it failed to be classified in the top rratio of the stream (thus rfraction of the stream is more relevant to the given profile than the document under consideration). At the same time this document was classified as being in the top rpart of the stream. Thus this mechanism defines a score (let's call it &#x3c3;) for a document depending on how it compares to other documents in the stream when scored against a given profile.

Growing community of inventors

Pittsburgh, PA, United States of America

Norbert Roma