Abstract: Implementing query classification includes executing a reductionist module on a query to extract a core term, which term is used to search a hash table that maps core terms to corresponding categories, deriving a first result including one of the categories from the search, and executing an enrichment module on the query to yield a second result. The enrichment module includes searching an index of terms that are mapped to documents and corresponding categories. Upon determining the core term is present in the hash table, a weighted average is calculated for values of the first and second results based on training data. Upon determining the core term from the query is not in the hash table, and also that a probability score of the category in the index for the second result meets a minimum confidence value, the core term and the corresponding categories are stored in the hash table.

Abstract: A system and method is provided which may comprise parsing an unstructured geographic web-search query into a field-based format, by utilizing conditional random fields, learned by semi-supervised automated learning, to parse structured information from the unstructured geographic web-search query. The system and method may also comprise establishing semi-supervised conditional random fields utilizing one of a rule-based finite state machine model and a statistics-based conditional random field model. Systematic geographic parsing may be used with the one of the rule-based finite state machine model and the statistics-based conditional random field model. Parsing an unstructured local geographical web-based query in local domain may be done by applying a learned model parser to the query, using at least one class-based query log from a form-based query system. The learned model parser may comprise at least one class-level n-gram language model-based feature harvested from a structured query log.

Abstract: Implementing query classification includes executing a reductionist module on a query to extract a core term, which term is used to search a hash table that maps core terms to corresponding categories; deriving a first result including one of the categories from the search; and executing an enrichment module on the query to yield a second result. The enrichment module includes searching an index of terms that are mapped to documents and corresponding categories. Upon determining the core term is present in the hash table, a weighted average is calculated for values of the first and second results based on training data. Upon determining the core term from the query is not in the hash table, and also that a probability score of the category in the index for the second result meets a minimum confidence value, the core term and the corresponding categories are stored in the hash table.

Abstract: Implementing query classification includes receiving a core term from a search query responsive to execution of a first module. The first module searches a table for the core term and yields a first result. The query classification also includes receiving a second result from the search query responsive to execution of a second module. The second module searches an index of terms that are mapped to documents and corresponding categories in the index. The second result is indicative of one of the corresponding categories in the index based on a probability score. Upon determining the first result is a category associated with the core term in the table, the query classification also includes calculating a weighted average for the first result and the second result. The calculation yields a third result. The query classification further includes transmitting the third result to a computer device that generated the query.

Abstract: A method and system for applying a task to a crowdsourcing platform for processing are disclosed. For example, a method forwards the task having a question to the crowdsourcing platform, and receives a plurality of answers to the question, wherein the plurality of answers is provided by a plurality of workers associated with the crowdsourcing platform. The method analyzes the plurality of answers on a per-worker basis to determine an adjustment, and applies the adjustment to the plurality of answers.

Abstract: Systems and computer-implemented methods include a validation phase in which validation projects are submitted to an online crowdsourcing system. The validation projects each have different combinations of the two validation control parameters, payment per human intelligence task (“HIT”) and number of solutions per HIT. By analyzing the solutions for the validation projects, final control parameters can be selected from the validation project having the best combination of high quality, low cost, and short time to completion. The final control parameters can be submitted along with a final project to the online crowdsourcing system. In addition, a novel outlier detection algorithm enables the system to detect outlier workers based on the distribution of non-experts' agreement probabilities, reject outlier workers' results, and thus save labeling cost.

Abstract: In one or more embodiments, one or more methods and/or systems described can perform producing a lattice of object hypotheses based on multiple reference objects from image information; receiving input speech information that includes a request for information associated with at least one reference object of the multiple reference objects; producing a lattice of speech hypotheses based on at least a first possible description included in the speech information; producing a lattice of scored semantic hypotheses based on at least the lattice of object hypotheses and the lattice of speech hypotheses; determining that a single semantic interpretation score of the lattice of scored semantic hypotheses exceeds a predetermined value; and providing requested information associated with the at least the first reference object of the plurality of reference objects.

Abstract: Implementing query classification includes receiving a core term from a search query responsive to execution of a first module. The first module searches a table for the core term and yields a first result. The query classification also includes receiving a second result from the search query responsive to execution of a second module. The second module searches an index of terms that are mapped to documents and corresponding categories in the index. The second result is indicative of one of the corresponding categories in the index based on a probability score. Upon determining the first result is a category associated with the core term in the table, the query classification also includes calculating a weighted average for the first result and the second result. The calculation yields a third result. The query classification further includes transmitting the third result to a computer device that generated the query.

Abstract: A system and method is provided which may comprise parsing an unstructured geographic web-search query into a field-based format, by utilizing conditional random fields, learned by semi-supervised automated learning, to parse structured information from the unstructured geographic web-search query. The system and method may also comprise establishing semi-supervised conditional random fields utilizing one of a rule-based finite state machine model and a statistics-based conditional random field model. Systematic geographic parsing may be used with the one of the rule-based finite state machine model and the statistics-based conditional random field model. Parsing an unstructured local geographical web-based query in local domain may be done by applying a learned model parser to the query, using at least one class-based query log from a form-based query system. The learned model parser may comprise at least one class-level n-gram language model-based feature harvested from a structured query log.

Abstract: A method and system for applying a task to a crowdsourcing platform for processing are disclosed. For example, a method forwards the task having a question to the crowdsourcing platform, and receives a plurality of answers to the question, wherein the plurality of answers is provided by a plurality of workers associated with the crowdsourcing platform. The method analyzes the plurality of answers on a per-worker basis to determine an adjustment, and applies the adjustment to the plurality of answers.

Abstract: In one or more embodiments, one or more methods and/or systems described can perform producing a lattice of object hypotheses based on multiple reference objects from image information; receiving input speech information that includes a request for information associated with at least one reference object of the multiple reference objects; producing a lattice of speech hypotheses based on at least a first possible description included in the speech information; producing a lattice of scored semantic hypotheses based on at least the lattice of object hypotheses and the lattice of speech hypotheses; determining that a single semantic interpretation score of the lattice of scored semantic hypotheses exceeds a predetermined value; and providing requested information associated with the at least the first reference object of the plurality of reference objects.

Abstract: Systems and computer-implemented methods include a validation phase in which validation projects are submitted to an online crowdsourcing system. The validation projects each have different combinations of the two validation control parameters, payment per human intelligence task (“HIT”) and number of solutions per HIT. By analyzing the solutions for the validation projects, final control parameters can be selected from the validation project having the best combination of high quality, low cost, and short time to completion. The final control parameters can be submitted along with a final project to the online crowdsourcing system. In addition, a novel outlier detection algorithm enables the system to detect outlier workers based on the distribution of non-experts' agreement probabilities, reject outlier workers' results, and thus save labeling cost.