Abstract: Computerized data processing and electronic file management methods of organizing and indexing electronic records in an electronic database for categorizing new products that are being added to an existing database of product offerings and computerized digital data processing methods of transferring digital information between a plurality of computers and employing computer instructions to categorize new products that are being added to an existing database of product offerings. Multiple classification models classify a description of a particular product and the classifications are compared, and if found to be equivalent, are added to the existing database of product offerings. If the classifications from the models are not equivalent, then the description is sent to multiple people for classification and the classifications from the people are compared, and if found to be equivalent, are added to the existing database of product offerings.

Abstract: Product records having attributes according to various schema are normalized such that the attributes thereof conform to a canonical schema. Attributes for large numbers of product records are normalized according to a map-reduce framework in which mappers are defined but no reducers. Rules are implemented by a rule engine that is instantiated one time per VM of the map-reduce framework. Likewise, each rule may be implemented by only one object instance per rule engine instance. Generation of rules by analysts may be facilitated by defining an attribute hierarchy. A rule defined for a parent attribute may be presented as a default rule for a child attribute and either accepted or revised. Attributes may be clusters and proposed rules generated. Proposed rules may then be accepted or rejected by an analyst.

Abstract: A method of validating rules configured to be utilized in an information extraction application, including: receiving a plurality of labeled samples in a training database; for each of the rules in the rule database: (a) determining, for each of the data points of the plurality of labeled samples in the training database to which the rule applies, whether applying the rule to the data point has a positive or negative impact on matching an output for the data point based on the rule to the assured output of the labeled sample corresponding to the data point; (b) generating positive impact information for the rule based on the positive voters; (c) generating negative impact information for the rule based on the negative voters; and (d) determining a metric for the rule based on the quantity of the negative voters and the quantity of the positive voters; ranking the rules based on the metrics corresponding to the rules; and sending to a user for refinement one or more flagged rules of the rules that have a lowes

Abstract: A method of determining structured product information for a product from a product description using a product entity graph. The product graph can include a plurality of nodes. Each of the plurality of nodes can include an entity value key, one or more entity names, and an entity name count for each of the one or more entity names. The method can include determining k-grams of the product description. The method also can include, for each k-gram of the product description, determining a matching node of the plurality of nodes of the product entity graph that corresponds to the k-gram and determining a derived entity name for the product from the one or more entity names of the matching node based at least in part on the entity name counts corresponding to the one or more entity names. Other embodiments of related systems and methods are also disclosed.

Abstract: Systems and methods are disclosed herein for rule-based item classification. The methods include receiving, by a computing device, an item record for analysis. The computing device may determine ranked lists of item types using rule-based classifiers and machine learning-based classifiers. Then, the computing device may aggregate the ranked lists of item types to generate a combined ranked list of item types.

Abstract: Systems and methods are disclosed herein for tuning an item classification. In one aspect, a method may include receiving, by a computing device, a classification request. The computing device may determine an item type of the item using a plurality of classifiers, and generate information of item type determination corresponding to the item. In response to a determination that a confidence score associated with the determined item type is less than a predetermined threshold, the computing device may provide the information of item type determination for analysis.

Abstract: Systems and methods are disclosed herein for classifying records, such as product records, using a machine learning algorithm. After training a classification model according to a machine learning algorithm using an initial training set, records are classified and high confidence classifications identified. Remaining classifications are submitted to a crowdsourcing forum that validates or invalidates the classifications or marks them as to unclear to evaluate. Invalidated classifications are automatically analyzed to identify one or both of classification values and categories having a high proportion of invalidated classifications. Requests are transmitted to analysts to generate training data that is added to the training set. The process of classifying records and obtaining crowdsourced validation thereof may then repeat.

Abstract: Systems and methods are disclosed herein for performing classification of documents or performing other tasks based on rules. The rules may include context rules that define a mapping that relates a value and context in a document to an attribute to which the value corresponds. Products are selected for labeling with attributes by identifying patterns, e.g. values and contexts that are not covered by a current rule set. Those products having a highest score are selected for labeling in a crowd sourcing forum, where the score is based on the number of non-covered patterns and a frequency of occurrence of the non-covered patterns in a document corpus. Proposed rules are generated for frequently occurring patterns and submitted to analysts for one or both of completion and validation. Proposed rules may include a proposed attribute for a frequently occurring value and corresponding context.

Abstract: Systems and methods are disclosed herein for generating a normalized record from an import record, the normalized record having attribute-value pairs corresponding to a native schema. In import records, a plurality of attribute-value are identified each having an attribute label not found in a native schema. One or more attribute labels in the native schema having as possible values one or more values corresponding to the values of the plurality of attribute-value pairs are also identified. The computer system generates one or more normalization rules relating one or more attribute labels of the plurality of attribute-value pairs to at least a portion of the one or more attribute labels in the native schema. Normalization rules may be validated by crowdsourcing. Normalization rules may be applied by identifying implicated rules by classifying the import record and identifying rules applicable to the classification.

Abstract: A method of validating rules configured to be utilized in an information extraction application, including: receiving a plurality of labeled samples in a training database; for each of the rules in the rule database: (a) determining, for each of the data points of the plurality of labeled samples in the training database to which the rule applies, whether applying the rule to the data point has a positive or negative impact on matching an output for the data point based on the rule to the assured output of the labeled sample corresponding to the data point; (b) generating positive impact information for the rule based on the positive voters; (c) generating negative impact information for the rule based on the negative voters; and (d) determining a metric for the rule based on the quantity of the negative voters and the quantity of the positive voters; ranking the rules based on the metrics corresponding to the rules; and sending to a user for refinement one or more flagged rules of the rules that have a lowes

Abstract: Product records having attributes according to various schema are normalized such that the attributes thereof conform to a canonical schema. Attributes for large numbers of product records are normalized according to a map-reduce framework in which mappers are defined but no reducers. Rules are implemented by a rule engine that is instantiated one time per VM of the map-reduce framework. Likewise, each rule may be implemented by only one object instance per rule engine instance. Generation of rules by analysts may be facilitated by defining an attribute hierarchy. A rule defined for a parent attribute may be presented as a default rule for a child attribute and either accepted or revised. Attributes may be clusters and proposed rules generated. Proposed rules may then be accepted or rejected by an analyst.

Abstract: Computerized data processing and electronic file management methods of organizing and indexing electronic records in an electronic database for categorizing new products that are being added to an existing database of product offerings and computerized digital data processing methods of transferring digital information between a plurality of computers and employing computer instructions to categorize new products that are being added to an existing database of product offerings. Multiple classification models classify a description of a particular product and the classifications are compared, and if found to be equivalent, are added to the existing database of product offerings. If the classifications from the models are not equivalent, then the description is sent to multiple people for classification and the classifications from the people are compared, and if found to be equivalent, are added to the existing database of product offerings.

Abstract: Systems and methods are disclosed herein for obtaining a structured listing of attributes and corresponding values based on an unstructured document, such as a product description in a product record. Putative values are identified in the document and corresponding candidate attributes are identified in a taxonomy. Attribute-value pairs are then evaluated with respect to a plurality of rules. Attribute-value pairs and outputs of the one or more rules are evaluated using a machine-learning algorithm, such as a decision tree, in order to determine which attribute-value pairs to retain. Retained attribute-value pairs are stored and used to respond to search queries and facilitate comparison of products. Attributes selected may also be used to update a product template.

Abstract: A method of determining structured product information for a product from a product description using a product entity graph. The product graph can include a plurality of nodes. Each of the plurality of nodes can include an entity value key, one or more entity names, and an entity name count for each of the one or more entity names. The method can include determining k-grams of the product description. The method also can include, for each k-gram of the product description, determining a matching node of the plurality of nodes of the product entity graph that corresponds to the k-gram and determining a derived entity name for the product from the one or more entity names of the matching node based at least in part on the entity name counts corresponding to the one or more entity names. Other embodiments of related systems and methods are also disclosed.

Abstract: Systems and methods are disclosed herein for classifying records, such as product records, using a machine learning algorithm. After training a classification model according to a machine learning algorithm using an initial training set, records are classified and high confidence classifications identified. Remaining classifications are submitted to a crowdsourcing forum that validates or invalidates the classifications or marks them as to unclear to evaluate. Invalidated classifications are automatically analyzed to identify one or both of classification values and categories having a high proportion of invalidated classifications. Requests are transmitted to analysts to generate training data that is added to the training set. The process of classifying records and obtaining crowdsourced validation thereof may then repeat. High confidence classifications may be identified using an accuracy model trained to relate an accuracy percentage to a confidence score output by the classification model.

Abstract: The present disclosure extends to methods, systems, and computer program products for automatically determining key words within item information with product types, and classifying new items within product types within a merchant's database.

Abstract: Systems and methods are disclosed herein for tuning an item classification. In one aspect, a method may include receiving, by a computing device, a classification request. The computing device may determine an item type of the item using a plurality of classifiers, and generate information of item type determination corresponding to the item. In response to a determination that a confidence score associated with the determined item type is less than a predetermined threshold, the computing device may provide the information of item type determination for analysis.

Abstract: Systems and methods are disclosed herein for rule-based item classification. The methods include receiving, by a computing device, an item record for analysis. The computing device may determine ranked lists of item types using rule-based classifiers and machine learning-based classifiers. Then, the computing device may aggregate the ranked lists of item types to generate a combined ranked list of item types.

Abstract: Systems and methods are disclosed herein for generating a normalized record from an import record, the normalized record having attribute-value pairs corresponding to a native schema. In import records, a plurality of attribute-value are identified each having an attribute label not found in a native schema. One or more attribute labels in the native schema having as possible values one or more values corresponding to the values of the plurality of attribute-value pairs are also identified. The computer system generates one or more normalization rules relating one or more attribute labels of the plurality of attribute-value pairs to at least a portion of the one or more attribute labels in the native schema. Normalization rules may be validated by crowdsourcing. Normalization rules may be applied by identifying implicated rules by classifying the import record and identifying rules applicable to the classification.

Abstract: Systems and methods are disclosed herein for obtaining a structured listing of attributes and corresponding values based on an unstructured document, such as a product description in a product record. Putative values are identified in the document and corresponding candidate attributes are identified in a taxonomy. Attribute-value pairs are then evaluated with respect to a plurality of rules. Attribute-value pairs and outputs of the one or more rules are evaluated using a machine-learning algorithm, such as a decision tree, in order to determine which attribute-value pairs to retain. Retained attribute-value pairs are stored and used to respond to search queries and facilitate comparison of products. Attributes selected may also be used to update a product template.