Abstract: Computer systems and methods incorporate user annotations (metadata) regarding various pages or sites, including annotations by a querying user and by members of a trust network defined for the querying user into search and browsing of a corpus such as the World Wide Web. A trust network is defined for each user, and annotations by any member of a first user's trust network are made visible to the first user during search and/or browsing of the corpus. Users can also limit searches to content annotated by members of their trust networks or by members of a community selected by the user.

Abstract: A method of ranking search results includes producing a relevance score for a document in view of a query. A similarity score is calculated for the query utilizing a feature vector that characterizes attributes and query words associated with the document. A rank value is assigned to the document based upon the relevance score and the similarity score.

Abstract: This invention includes the step of transmitting a query to a set of search engines. Any result lists returned from these search engines is received, and a subset of entries in each result list is selected. Each entry in this subset is assigned a scoring value according to a scoring function, and each result list is then assigned a representative value according to the scoring values assigned to its entries. A merged list of entries is produced based upon the representative value assigned to each result list.

Abstract: A system and method for training an SVM in a scalable manner includes boosting the SVM during training. Specifically, individual SVMs in an ensemble of SVMs are trained using small subsets of a training data set, with data that earlier classifiers in the ensemble incorrectly classified being overrepresented in succeeding subsets. In this way, the speed with which the overall SVM is trained is increased and the memory requirements therefor are reduced, even for relatively large training data sets.

Abstract: The invention is a method, system and computer program for automatically discovering concepts from a corpus of documents and automatically generating a labeled concept hierarchy. The method involves extraction of signatures from the corpus of documents. The similarity between signatures is computed using a statistical measure. The frequency distribution of signatures is refined to alleviate any inaccuracy in the similarity measure. The signatures are also disambiguated to address the polysemy problem. The similarity measure is recomputed based on the refined frequency distribution and disambiguated signatures. The recomputed similarity measure reflects actual similarity between signatures. The recomputed similarity measure is then used for clustering related signatures. The signatures are clustered to generate concepts and concepts are arranged in a concept hierarchy. The concept hierarchy automatically generates query for a particular concept and retrieves relevant documents associated with the concept.

Abstract: The present invention provides a method, system and computer program for naming a cluster, or a hierarchy of clusters, of words and phrases that have been extracted from a set of documents. The invention takes these clusters as the input and generates appropriate labels for the clusters using a lexical database. Naming involves first finding out all possible word senses for all the words in the cluster, using the lexical database; and then augmenting each word sense with words that are semantically similar to that word sense to form respective definition vectors. Thereafter, word sense disambiguation is done to find out the most relevant sense for each word. Definition vectors are clustered into groups. Each group represents a concept. These concepts are thereafter ranked based on their support. Finally, a pre-specified number of words and phrases from the definition vectors of the dominant concepts are selected as labels, based on their generality in the lexical database.

Abstract: A method and apparatus for indexing and searching content in a hardcopy document utilizes a searching assistant computing device (402) with an index table (420) stored in memory (412). The index table (420) is created in memory by scanning a 2-D barcode from a hardcopy document or alternatively by downloading indexing information from a web page via the Internet (430). A search engine (410) in the searching assistant (402) searches the index table (420) to locate a data element found in the content of the hardcopy document. The indexing information corresponding to the data element is displayed to a user as part of the search results to indicate the location of the data element in the hardcopy document.

Abstract: This invention includes the step of transmitting a query to a set of search engines. Any result lists returned from these search engines is received, and a subset of entries in each result list is selected. Each entry in this subset is assigned a scoring value according to a scoring function, and each result list is then assigned a representative value according to the scoring values assigned to its entries. A merged list of entries is produced based upon the representative value assigned to each result list.

Abstract: A method of ranking search results includes producing a relevance score for a document in view of a query. A similarity score is calculated for the query utilizing a feature vector that characterizes attributes and query words associated with the document. A rank value is assigned to the document based upon the relevance score and the similarity score.

Abstract: An application caching system and method are provided wherein one or more applications may be cached throughout a distributed computer network. The system may include a central cache directory server, one or more distributed master application servers and one or more distributed application cache servers. The system may permit a service, such as a search, to be provided to the user more quickly.

Abstract: A method, apparatus, and article of manufacture employing lexicon reduction using key characters and a neural network, for recognizing a line of cursive text. Unambiguous parts of a cursive image, referred to as “key characters,” are identified. If the level of confidence that a segment of a line of cursive text is a particular character is higher than a threshold, and is also sufficiently higher than the level of confidence of neighboring segments, then the character is designated as a key character candidate. Key character candidates are then screened using geometric information. The key character candidates that pass the screening are designated key characters. Two-stages of lexicon reduction are employed. The first stage of lexicon reduction uses a neural network to estimate a lower bound and an upper bound of the number of characters in a line of cursive text. Lexicon entries having a total number of characters outside of the bounds are eliminated.

Abstract: A method, apparatus, and article of manufacture employing lexicon reduction using key characters and a neural network, for recognizing a line of cursive text. Unambiguous parts of a cursive image, referred to as “key characters,” are identified. If the level of confidence that a segment of a line of cursive text is a particular character is higher than a threshold, and is also sufficiently higher than the level of confidence of neighboring segments, then the character is designated as a key character candidate. Key character candidates are then screened using geometric information. The key character candidates that pass the screening are designated key characters. Two-stages of lexicon reduction are employed. The first stage of lexicon reduction uses a neural network to estimate a lower bound and an upper bound of the number of characters in a line of cursive text. Lexicon entries having a total number of characters outside of the bounds are eliminated.

Abstract: A method, apparatus, and article of manufacturing employing lexicon reduction using key characters and a neural network, for recognizing a line of cursive text. Unambiguous parts of a cursive image, referred to as “key characters,” are identified. If the level of confidence that a segment of a line of cursive text is a particular character is higher than a threshold, and is also sufficiently higher than the level of confidence of neighboring segments, then the character is designated as a key character candidate. Key character candidates are then screened using geometric information. The key character candidates that pass the screening are designated key characters. Two-stages of lexicon reduction are employed. The first stage of lexicon reduction uses a neural network to estimate a lower bound and an upper bound of the number of characters in a line of cursive text. Lexicon entries having a total number of characters outside of the bounds are eliminated.

Abstract: Fixed-pitch, fixed-font characters embedded in a noisy gray-scale image of picture elements (pels) within a complex background can be extracted prior to execution of any recognition operations by first deriving a normalized Boolean-coded image from the gray-scale image. Then, a subset of at least three uncontaminated character triples is formed by filtering the Boolean-coded image. Next, an affine transform is approximated from locations in the Boolean-coded image of at least three noncollinear ones of the uncontaminated character triples. Lastly, the locations in a logical matrix array of all possible character triples are estimated according to the affine transform.

Abstract: A multi-stage multi-network character recognition system decomposes the estimation of a posteriori probabilities into coarse-to-fine stages. Classification is then based on the estimated a posteriori probabilities. This classification process is especially suitable for the tasks that involve a large number of categories. The multi-network system is implemented in two stages: a soft pre-classifier and a bank of multiple specialized networks. The pre-classifier performs coarse evaluation of the input character, developing different probabilities that the input character falls into different predefined character groups. The bank of specialized networks, each corresponding to a single group of characters, performs fine evaluation of the input character, where each develops different probabilities that the input character represents each character in that specialized network's respective predefined character group.