Abstract: One aspect of the present invention includes a method and device for incremental updating of a training set of documents used for automatic categorization. Particular aspects of the present invention are described in the claims, specification and drawings.

Abstract: Meta-content analysis and annotation upon the body of email documents, and other electronic documents, and to create a displayable index of these instances of meta-content, which is sorted and annotated by type are provided. In addition, the electronic document is enhanced by providing links for the semantic foci to external documents containing related information.

Abstract: Phrases in a corpus of documents including stopwords are found using a data processor arranged to execute phrase queries. Memory stores an index structure which maps entries in the index structure to documents in the corpus. Entries in the index structure represent words and other entries represent stopwords found in the corpus coalesced with prefixes of respective adjacent words adjacent to the stopwords. The prefixes comprise one or more leading characters of the respective adjacent words. A query processor forms a modified query by substituting a stopword with a search token representing the stopword coalesced with a prefix of the next word in the query. The processor executes the modified query. Also, index structures including coalesced stopwords are created and maintained.

Abstract: Words having selected characteristics in a corpus of documents are found using a data processor arranged to execute queries. Memory stores an index structure in which entries in the index structure map words and marks for words having the selected characteristics to locations within documents in the corpus. Entries in the index structure represent words and other entries represent marks with the location information of a marked word. The entries for the marks can be tokens coalesced with prefixes of respective marked words or adjacent. A query processor forms a modified query by adding a mark for a word to the query. The processor executes the modified query.

Abstract: A system and method is provided for accessing information from a plurality of searchable information sources. The method includes the steps of: analyzing a user search query to determine a subject matter of the query; and selecting a sub-set of information from the plurality of information sources based upon the determined subject matter of the query. In further detailed embodiment, the analyzing step combines at least two methods of deriving the subject matter from the search query; and the method further includes the step of searching the information source(s) in the sub-set of information sources, substantially in parallel, for documents relevant to the search query. A system and method is also provided for searching a plurality of searchable information sources, where the information sources include at least one secure source.

Abstract: The present invention includes methods and systems to audit and identify potential errors and/or omissions in a training set. Particular aspects of the present invention are described in the claims, specification and drawings.

Abstract: A system and method is provided for accessing information from a plurality of searchable information sources. The method includes the steps of: analyzing a user search query to determine a subject matter of the query; and selecting a sub-set of information from the plurality of information sources based upon the determined subject matter of the query. In further detailed embodiment, the analyzing step combines at least two methods of deriving the subject matter from the search query; and the method further includes the step of searching the information source(s) in the sub-set of information sources, substantially in parallel, for documents relevant to the search query. A system and method is also provided for searching a plurality of searchable information sources, where the information sources include at least one secure source.

Abstract: Disclosed are methods and for incrementally updating the accuracy provided by documents in training set of used for automatic categorization. A k-nearest neighbor database includes the documents in the training set, categories, category assignments of the documents and category scores for the documents. A list made up of the nearest neighbors of the documents and corresponding similarity scores contains is maintained by the method. On adding or deleting documents or category assignments, the documents influenced by the changed documents or category assignments are identified. The category scores of the identified documents are updated to be consistent for the updated training set and a new precision and recall curves are computed for the categories including updated category scores. The precision and recall curves may be used to determine an optimal number of documents to maximize the return of relevant documents while minimizing the total number of documents.

Abstract: A node-link structure is displayed within a display area having a narrow rectangular shape with an edge along one side acting as a horizon of a hyperbolic space half-plane. Lower level node features that share a parent node feature have centers of area positioned on the display in order along a line parallel with the horizon, with sufficiently similar spacings along an axis perpendicular to the horizon from the region around a parent node feature, and with sufficiently similar spacings in a dimension parallel to the horizon from adjacent node features along the line, that the lower level node features sharing the parent node feature are perceptible as a group of related node features. The half-plane model with compression is used for layout of the node-link data, and the hyperbolic layout data is mapped to a Euclidean space for display.

Abstract: A node-link structure is displayed within a display area, having a narrow rectangular shape with an edge along one side acting as a horizon of a hyperbolic space half-plane. Lower level node features that share a parent node feature have centers of area positioned on the display in order along a line parallel with the horizon, with sufficiently similar spacings along an axis perpendicular to the horizon from the region around a parent node feature, and with sufficiently similar spacings in a dimension parallel to the horizon from adjacent node features along the line, that the lower level node features sharing the parent node feature are perceptible as a group of related node features. The half-plane model with compression is used for layout of the node-link data, and the hyperbolic layout data is mapped to a Euclidean space for display.

Abstract: A method is provided for information extraction and classification which combines aspects of local regularities formulation with global regularities formulation. A candidate subset is identified. Then tentative labels are created so they can be associated with elements in the subset that have the global regularities, and the initial tentative labels are attached onto the identified elements of the candidate subset. The attached tentative labels are employed to formulate or “learn” initial local regularities. Further tentative labels are created so they can be associated with elements in the subset that have a combination of global and local regularities, and the further tentative labels are attached onto the identified elements of the candidate subset. Each new dataset is processed with reference to an increasingly-refined set of global regularities, and the output data with their associated confidence labels can be readily evaluated as to import and relevance.

Abstract: In a series of iterations, each of which automatically traverses a set of elements of a node-link structure, an iteration modifies the part of node-link data defining the structure that is in memory. The modified part of node-link data is more likely to define the set of elements that will be traversed by the next iteration. The iteration can determine whether to remove any of the part of node-link data, and then apply a criterion, such as a navigation history criterion or a least recently traversed criterion, to determine which of the part to remove. The iteration can also determine whether its set of elements includes an element with children that are not defined by the part of node-link data in memory, and can obtain a modified part that defines those children. Each iteration can, for example, prepare and present a representation of the node-link structure.

Abstract: A software visualization tool consistent with the present invention integrates OLAP functionality with focus+context based techniques for navigation through and inspection of large multidimensional datasets. Focus+context based navigation techniques are used to increase the clarity and information content provided to the user. The visualization tool supports a number of operations of the data set, including: select-slice, aggregation, promote/demote, repeat-variables, and sort.

Abstract: A sequence of steps presents node-link representations on a display. The sequence includes a first step, a last step, and at least one intermediate step. The first step presents a first representation that represents a first node-link structure. The last step presents a last representation that represents a second node-link structure that is a modified version of the first node-link structure. Both node-link structures include a set of shared elements that include a moving element. The moving element is represented by features that have different positions in the first and last representations. Each intermediate step presents an intermediate representation that includes features representing a subset of the shared elements including the moving element, and each element in the subset is also represented by features in the first and last representations. The feature representing the moving element has object constancy through the sequence of steps.

Abstract: A node-link structure laid out in a space with negative curvature, such as the hyperbolic plane, is mapped to a rendering space, such as the unit disk, beginning from a starting element that has both a parent and a child. Data identifying the starting element are obtained, such as based on user signals. The starting element can be located at a starting position in the unit disk. Then a set of other elements can be mapped, each reachable from the starting element by a path that leads only through elements in the set. Each element's position in the negatively curved space relative to a preceding element on its path and the position of the preceding element in the rendering space can be used to obtain the element's position in the rendering space. The positions in the unit disk can then be used to present a representation of the node-link structure.