Abstract: Provided are, among other things, systems, methods and techniques for document-based processing. In one implementation, a document is input; features are extracted from it; an index is queried using at least a subset of the extracted features and, in response, identifications for selected document classifiers are received from a larger pool of document classifiers; the document is processed using individual ones of the selected document classifiers, thereby generating corresponding classifier outputs; and then, based on such classifier outputs, (1) the document is categorized within a computer database and/or (2) feedback information is provided to a user.

Abstract: Provided are, among other things, systems, methods and techniques for generating a set of representative components from a set of documents, by selecting at least one component for initial inclusion within a representative component set. An additional component is then identified based on (1) frequency of occurrence of the additional component across the set of documents and (2) strength of negative correlation with the representative component set across the set of documents. The representative component set is supplemented with the additional component, and the identification and supplementing steps are repeated multiple times until a specified end condition is satisfied. The representative component set is output and/or the documents are processed based on the representative component set.

Abstract: For each first entity of a subset of a number of first entities, an expected improvement of a predictive performance of a collaborative filtering model if additional ratings of the first entity in relation to a plurality of second entities were obtained is estimated. Particular first entities from the subset of the first entities of which to obtain the additional ratings in relation to the second entities are selected based at least on the expected improvements that have been determined. The additional ratings of the particular first entities in relation to the second entities are obtained.

Abstract: Techniques for determining classifications from content of data objects (100) are disclosed. Terms from the content of one or more data objects (100) of each of a plurality of classes (200) are used to determine a sub-topic (210) for one of the classes (200).

Abstract: A computing system including a component to perform a function and generates a noise. A microphone to receive an input including the noise. The computing system can monitor a component for an event that produces a noise.

Abstract: An example map matching technique in accordance with the present disclosure includes receiving a plurality of global positioning system (GPS) data points in a dataset, receiving road map data related to a plurality of roads, determining a plurality of paths of minimum Fréchet distance for the GPS dataset, assigning a weight to each path of minimum Fréchet distance by applying a weight function, and outputting the path with the minimum weight.

Abstract: Provided are, among other things, systems, methods and techniques for scanning information across multiple different devices. In one representative system, remote data-processing devices are provided with scanning applications that repeatedly scan information on their respective data-processing devices to identify matching data units that satisfy a specified matching criterion, the specified matching criterion including required matches against a set of screening digests, and then transmit characteristic information regarding the matching data units; and a central processing facility receives the characteristic information from the remote data-processing devices and determines whether the corresponding matching data units satisfy a policy criterion.

Abstract: A technique for determining an obverse weight. A set of cases can be divided into bins. An obverse weight for a bin can be determined based on an importance weight of the bin and a variance of an error estimate of the bin.

Abstract: A computing system including a component to perform a function and generates a noise. A microphone to receive an input including the noise. The computing system can monitor a component for an event that produces a noise.

Abstract: Provided are, among other things, systems, methods and techniques for classifying a collection of documents. A term is identified based on an indication of ability of the term's presence within a given document to predict whether the given document should be classified into an identified category. A document index is then queried using the identified term and, in response, search results that define a candidate set of documents are received. Finally, a classifier is applied to documents within the candidate set to determine which of the documents should be classified into the identified category.

Abstract: A technique to perform cross-validation using a set of randomly selected labeled cases and a set of non-randomly selected labeled cases. A training set for use during cross-validation can include cases from both sets. A test set for use during cross-validation can include cases from the randomly selected set but exclude cases from the non-randomly selected set.

Abstract: A technique for updating road maps is disclosed. A number of GPS traces can be matched with a number of roads in a map. Matched GPS traces may be processed by a matched segment module to produce proposed changes to the map. The map can be updated using a map updating module based on the proposed changes from the matched segment module. Unmatched GPS traces may be processed by an unmatched segment module to produce proposed changes to the map. The map can be updated using a map updating module based on the proposed changes from the unmatched segment module. The proposed changes to the map may include metadata defining new roads in the map, new intersections in the map, updates to turn restrictions in the map, updates to the allowable directional traffic flow on the roads within the map, updates to road closures in the map.

Abstract: A method for classifying cases includes receiving a pool of unlabeled cases with associated per-case symmetrical importance scores, applying a selection algorithm with a classifier to a training set and the pool, but without the per-case symmetrical importance scores, to determine selection scores for the unlabeled case, and combining the selection scores and the corresponding per-case symmetrical importance scores to form combined scores for the unlabeled cases. The method further includes providing a high scoring unlabeled case to an oracle to label, receiving a labeled case back from the oracle and augmenting the training set with the labeled case, training the classifier with the augmented training set, and applying the classifier to an additional unlabeled case.

Abstract: A method and a system of making a recommendation to a user that is currently generating events based on a subset of historical event data are provided. Historical event data, which is segmented into a set of sessions, is received. Each session includes events. The sessions are associated with clusters that represent the users that generated the historical event data. Each of the associated sessions is associated with one cluster and the number of the clusters is the same as the number of the users. A determination as to which cluster is associated with events currently being generated by a current user's behavior is made. The determining does not require identification of the current user. A recommendation is made to the current user based on the cluster that is associated with the events currently being generated.

Abstract: A method for classifying cases includes receiving a pool of unlabeled cases with associated per-case symmetrical importance scores, applying a selection algorithm with a classifier to a training set and the pool, but without the per-case symmetrical importance scores, to determine selection scores for the unlabeled case, and combining the selection scores and the corresponding per-case symmetrical importance scores to form combined scores for the unlabeled cases. The method further includes providing a high scoring unlabeled case to an oracle to label, receiving a labeled case back from the oracle and augmenting the training set with the labeled case, training the classifier with the augmented training set, and applying the classifier to an additional unlabeled case.

Abstract: Provided are, among other things, systems, methods and techniques for determining applicability of a policy defined by reference to a source document. A first sketch that was generated based on content of the source document is obtained, and a matching criterion is defined based on the first sketch. Also obtained is a second sketch that was generated based on content of a potential target document. A determination is made as to whether the policy applies to the potential target document based on whether the second sketch satisfies the matching criterion and, if the policy applies, a notification regarding applicability of the policy automatically is provided and/or an action automatically is blocked so as to prevent a violation of the policy.

Abstract: Locations of word images corresponding to words in a document image are ascertained. The word images are grouped into clusters. For each of multiple of the clusters, a respective compressed word image cluster is determined based on a joint compression of respective ones of the word images that are grouped into the cluster. The positions of the word images in the document image are associated with the respective ones of the compressed word image clusters corresponding to the clusters respectively containing the word images.

Abstract: A method and a system of making a recommendation to a user that is currently generating events based on a subset of historical event data are provided. Historical event data, which is segmented into a set of sessions, is received. Each session includes events. The sessions are associated with clusters that represent the users that generated the historical event data. Each of the associated sessions is associated with one cluster and the number of the clusters is the same as the number of the users. A determination as to which cluster is associated with events currently being generated by a current user's behavior is made. The determining does not require identification of the current user. A recommendation is made to the current user based on the cluster that is associated with the events currently being generated.

Abstract: An exemplary embodiment of the present invention provides a computer implemented method of developing a classifier. The method includes receiving input for a case, the case comprising a plurality of instances and an example, the example comprising a plurality of data fields each corresponding to one of the plurality of instances, wherein the input indicates which, if any, of the instances includes a data field belonging to a target class. The method also includes training the classifier based, at least in part, on the input from the trainer.

Abstract: Methods, systems, and computer program products are provided for personalizing web sites. A model based on mining web usage data is accessed. The model defines associations between web sites. Interest associations extracted from web interactions are stored. The interest associations comprise interest indications and web sites associated with the interest indications. An interest indication from the interest associations is selected. The interest indication is associated with an associated web site. The associated web site has an association with a target web site as defined by the model. The interest indication is sent to the target web site.