Abstract: Disclosed are various embodiments for using contributor-provided attributes in conjunction with an electronic item catalog. A computer reconciles a plurality of attributes for an item in an electronic catalog using feedback data related to the item. Each of the attributes is provided by one of a plurality of users. The reconciled attribute of the item is updated in the electronic catalog.

Abstract: Embodiments of systems and methods can determine evaluations of the quality of task results provided by judges. Certain embodiments can determine the quality evaluations with relatively little overlap of judges (e.g., two or three judges). The quality evaluations may include determining an error rate or a confidence index for a judge or for a particular judgment made by the judge. Certain embodiments may determine the quality evaluations by intercomparing judging results from multiple judges without requiring human intervention, comparison to evaluation data known (or assumed) to be accurate, or input from an external judge review panel. In some implementations, task results can be combined with a confidence score (based at least in part on the quality evaluation of the judge who provided the task result), and this information can be used as training data to improve performance of machine learning algorithms.

Abstract: Disclosed are various embodiments for using contributor-provided attributes in conjunction with an electronic item catalog. A computer reconciles a plurality of attributes for an item in an electronic catalog using feedback data related to the item. Each of the attributes is provided by one of a plurality of merchants. The reconciled attribute of the item is updated in the electronic catalog.

Abstract: Candidate suggestions for correcting misspelled query terms input into a search application are automatically generated. A score for each candidate suggestion can be generated using a first decoding pass and paths through the suggestions can be ranked in a second decoding pass. Candidate suggestions can be generated based on typographical errors, phonetic mistakes and/or compounding mistakes. Furthermore, a ranking model can be developed to rank candidate suggestions to be presented to a user.

Abstract: Systems and methods for dispersing search engine results by category. A search engine application queries a searchable index of document data associated with a plurality of electronic documents in response to a search request to identify one or more electronic documents having document data matching data included in the search request. The search engine application disperses identified electronic documents according to category data included in the document data for display to a user.

Abstract: An electronic ink object is preprocessed for improving handwriting recognition results. An electronic ink object, possibly having multiple lines of non-horizontal ink, is preprocessed by segmenting or separating the electronic ink object into one or more lines, computing a writing angle of each line, and rotating each line so that it is horizontal.

Abstract: An improved method and mechanism for recognizing chirographs (handwritten characters) input into a computer system. A primary recognizer is provided for converting chirographs to code points, and secondary recognizers such as binary CART trees are developed and trained to differentiate chirographs which produce certain code points at the primary recognizer. Each such secondary recognizer is associated with each selected code point. When a chirograph is received, the chirograph is provided to the primary recognizer whereby a code point corresponding thereto is received. If the code point corresponds to one of the secondary recognizers, the chirograph is passed to the secondary recognizer, and a code point is returned from the secondary recognizer. If not, the code point provided by the primary recognizer is returned.

Abstract: Candidate suggestions for correcting misspelled query terms input into a search application are automatically generated. A score for each candidate suggestion can be generated using a first decoding pass and paths through the suggestions can be ranked in a second decoding pass. Candidate suggestions can be generated based on typographical errors, phonetic mistakes and/or compounding mistakes. Furthermore, a ranking model can be developed to rank candidate suggestions to be presented to a user.

Abstract: A method and improved data structure for attaching information to words in a trie of nodes. Each node of a trie includes a single tag bit which is interpreted according to information specified in a header of the trie. The information may specify that a node may be tagged with multiple tags, whereby if the tag bit is set in a given node, the node further includes a bitmask indicating which one or ones of the tags apply to that node. A value mask may be provided in the header indicating which of the tags (if any) have values associated therewith, whereby information representative of the value such as the value itself or a pointer thereto is stored in each node tagged with at least one tag having an associated value. Partial enumeration of tagged nodes may be provided by storing a count of the number of tagged words under a node, wherein if a trie has multiple tags, each tag may be selectively and separately enumerated as specified in header information.

Abstract: In computerized recognition having multiple experts, a method and system is described that obtains an optimum value for an expert tuning parameter in a single pass over sample tuning data. Each tuning sample is applied to two experts, resulting in scores from which ranges of parameters that correct incorrect recognition errors without changing correct results for that sample are determined. To determine the range data for a given sample, the experts return scores for each prototype in a database, the scores separated into matching and non-matching scores. The matching and non-matching scores from each expert are compared, providing upper and lower bounds defining ranges. Maxima and minima histograms track upper and lower bound range data, respectively. An analysis of the histograms based on the full set of tuning samples provides the optimum value. For tuning multiple parameters, each parameter may be optimized by this method in isolation, and then iterated.

Abstract: An improved method and mechanism for recognizing chirographs (handwritten characters) input into a computer system. A primary recognizer is provided for converting chirographs to code points, and secondary recognizers such as binary CART trees are developed and trained to differentiate chirographs which produce certain code points at the primary recognizer. Each such secondary recognizer is associated with each selected code point. When a chirograph is received, the chirograph is provided to the primary recognizer whereby a code point corresponding thereto is received. If the code point corresponds to one of the secondary recognizers, the chirograph is passed to the secondary recognizer, and a code point is returned from the secondary recognizer. If not, the code point provided by the primary recognizer is returned.

Abstract: In computerized recognition having multiple experts, a method and system is described that obtains an optimum value for an expert tuning parameter in a single pass over sample tuning data. Each tuning sample is applied to two experts, resulting in scores from which ranges of parameters that correct incorrect recognition errors without changing correct results for that sample are determined. To determine the range data for a given sample, the experts return scores for each prototype in a database, the scores separated into matching and non-matching scores. The matching and non-matching scores from each expert are compared, providing upper and lower bounds defining ranges. Maxima and minima histograms track upper and lower bound range data, respectively. An analysis of the histograms based on the full set of tuning samples provides the optimum value. For tuning multiple parameters, each parameter may be optimized by this method in isolation, and then iterated.

Abstract: A method and improved data structure for attaching information to words in a trie of nodes. Each node of a trie includes a single tag bit which is interpreted according to information specified in a header of the trie. The information may specify that a node may be tagged with multiple tags, whereby if the tag bit is set in a given node, the node further includes a bitmask indicating which one or ones of the tags apply to that node. A value mask may be provided in the header indicating which of the tags (if any) have values associated therewith, whereby information representative of the value such as the value itself or a pointer thereto is stored in each node tagged with at least one tag having an associated value. Partial enumeration of tagged nodes may be provided by storing a count of the number of tagged words under a node, wherein if a trie has multiple tags, each tag may be selectively and separately enumerated as specified in header information.

Abstract: A method and improved data structure for attaching information to words in a trie of nodes. Each node of a trie includes a single tag bit which is interpreted according to information specified in a header of the trie. The information may specify that a node may be tagged with multiple tags, whereby if the tag bit is set in a given node, the node further includes a bitmask indicating which one or ones of the tags apply to that node. A value mask may be provided in the header indicating which of the tags (if any) have values associated therewith, whereby information representative of the value such as the value itself or a pointer thereto is stored in each node tagged with at least one tag having an associated value. Partial enumeration of tagged nodes may be provided by storing a count of the number of tagged words under a node, wherein if a trie has multiple tags, each tag may be selectively and separately enumerated as specified in header information.

Abstract: In computerized recognition having multiple experts, a method and system is described that obtains an optimum value for an expert tuning parameter in a single pass over sample tuning data. Each tuning sample is applied to two experts, resulting in scores from which ranges of parameters that correct incorrect recognition errors without changing correct results for that sample are determined. To determine the range data for a given sample, the experts return scores for each prototype in a database, the scores separated into matching and non-matching scores. The matching and non-matching scores from each expert are compared, providing upper and lower bounds defining ranges. Maxima and minima histograms track upper and lower bound range data, respectively. An analysis of the histograms based on the full set of tuning samples provides the optimum value. For tuning multiple parameters, each parameter may be optimized by this method in isolation, and then iterated.

Abstract: A method and mechanism for displaying partial results of full context handwriting recognition. As handwritten characters are entered into a system, a shape matcher associates the character with a plurality of alternate code points, with each alternate code point having probability information associated therewith. The alternate code points are placed at the end of a queue, and a cost is determined from each alternate code point to any immediately preceding alternate in the queue. The cost is based on the probability information of the alternates and a transition cost therebetween. Then, the lowest cost path back from each of the alternates at the end of the queue to an alternate at the beginning of the queue is determined. If each lowest cost path back converges to a common alternate in the queue, the common alternate and any previous alternates on the path back are recognized as the code points for each of the handwritten characters associated therewith.

Abstract: A method and system for enumerating a trie of states of nodes. A node near the middle of a state of alphabetically ordered nodes is selected as a skip node and moved to the logical beginning of the state. The skip node is provided with a pointer to a jump node that is at the skip node's former position, and the node immediately to the left of the node's former position is marked as a soft terminal node. As a result of the alphabetic ordering, the segment of nodes before the jump node are alphabetically before the skip node, while the segment of nodes after the jump node are alphabetically after the skip node. The segments of the state may be recursively split into further segments via further skip nodes, jump nodes and soft terminal nodes, and, once the segments are split as desired, the nodes within the segment may be sorted.

Abstract: An improved trie compression method that compresses by merging partially identical subtrees. States of the trie are selected, and the nodes of those states examined find nodes that are identical to one another. The most frequently occurring identical node is selected as a substate, and the states are separated into a first group of states that have the substate node therein and a second group of states that do not. The nodes in the first group of states are reordered such that the substate is at the end thereof. Then, the substate of each state is merged into a single node, replaced by a pointer from each state. Compression is performed recursively by choosing a new substate for the remaining nodes of the first group, and for subsequently separated groups, until no further identical nodes are available for merging.

Abstract: A method and mechanism for displaying partial results of full context handwriting recognition. As handwritten characters are entered into a system, a shape matcher associates the character with a plurality of alternate code points, with each alternate code point having probability information associated therewith. The alternate code points are placed at the end of a queue, and a cost is determined from each alternate code point to any immediately preceding alternate in the queue. The cost is based on the probability information of the alternates and a transition cost therebetween. Then, the lowest cost path back from each of the alternates at the end of the queue to an alternate at the beginning of the queue is determined. If each lowest cost path back converges to a common alternate in the queue, the common alternate and any previous alternates on the path back are recognized as the code points for each of the handwritten characters associated therewith.

Abstract: A computer system and method for generating probability tables for use in recognizing handwritten characters. The system receives handwritten characters as a sequence of strokes. The system characterizes each stroke by its shape and position. The position is represented as a pair of coordinates specifying the starting and ending points of the stroke. The system represents an input handwritten character by a string representing the shapes of the strokes and by a string of the pairs of coordinates of the strokes. The strings are ordered by the sequence in which the strokes are written. The system compares all possible pairs of a large sample of handwritten characters one to another by comparing the shape and position of each stroke in one character with those of the corresponding stroke of the other character. When comparing the two strokes, the system generates a pair of shapes and the sum of the squared distances between the starting points and the ending points of the two strokes.