Abstract: Disclosed herein are various embodiments for planogram management system. An embodiment operates by receiving a planogram of a retail establishment. The planogram indicates a relative location of a plurality of items offered for sale by the retail establishment placed across one or more retail locations within the retail establishment. An image of a first one of the retail locations of the retail establishment is received. The image is compared to the portion of the planogram corresponding to the first one of the retail locations, and one or more differences between the image and the portion of the planogram for the first retail location are identified. An employee of the retail establishment is notified of the one or more differences.

Abstract: Disclosed herein are system, method, and computer program product embodiments for providing an image search training system. An embodiment operates by determining a query image on which to train an image search system, and a positive image visually similar to the query image. A set of negative images from a plurality of negative images visually dissimilar to the query image are selected, where the selected set of negative images includes both a first negative image and a second negative image. A first similarity measure between the first negative image and the positive image, and a second similarity measure between the second negative image and the positive image are calculated. The first negative image is selected based on the first similarity measure being less than the second similarity measure. The query image, the positive image, and the first negative image are provided to the image search system for training.

Abstract: Disclosed herein are system, method, and computer program product embodiments for robust key value extraction. In an embodiment, one or more hierarchical concepts units (HCUs) may be configured to extract key value and hierarchical information from text inputs. The HCUs may use a convolutional neural network, a recurrent neural network, and feature selectors to analyze the text inputs using machine learning techniques to extract the key value and hierarchical information. Multiple HCUs may be used together and configured to identify different categories of hierarchical information. While multiple HCUs may be used, each may use a skip connection to transmit extracted information to a feature concatenation layer. This allows an HCU to directly send a concept that has been identified as important to the feature concatenation layer and bypass other HCUs.

Abstract: A method, apparatus, and computer program product are therefore provided for disambiguating non-deterministic spoken dialog commands as they relate to points-of-interest and information regarding points-of-interest within or near the field of view of a user. An apparatus may include a processor and a memory including program code instructions. The program code instructions are configured to, when executed, cause the apparatus to at least: receive an indication of a user location; receive an indication of a field of view of the user at the location; receive a natural language utterance; generate, from the natural language utterance, a query, where the query includes at least one request determined from the natural language utterance, the location, and the field of view of the user; determine a point-of-interest associated with the query by conducting a search of a database using the generated query; and provide information associated with the point-of-interest via a user interface.

Abstract: Disclosed herein are system, method, and computer program product embodiments for providing an image search training system. An embodiment operates by determining a query image on which to train an image search system, and a positive image visually similar to the query image. A set of negative images from a plurality of negative images visually dissimilar to the query image are selected, where the selected set of negative images includes both a first negative image and a second negative image. A first similarity measure between the first negative image and the positive image, and a second similarity measure between the second negative image and the positive image are calculated. The first negative image is selected based on the first similarity measure being less than the second similarity measure. The query image, the positive image, and the first negative image are provided to the image search system for training.

Abstract: An approach is provided for mobility-based language model adaptation for navigational speech interfaces. The approach involves receiving a word lattice resulting from a speech recognition process applied on a speech input received from a user via a navigational speech interface. The word lattice includes a respective speech recognition score for each word in the word lattice. The approach further involves adapting a language model comprising a plurality of location entities according to a mobility relevance and a search relevance of the plurality of location entities to the user. In one embodiment, the mobility relevance and the search relevance are determined by computing a probability of the user visiting the location entities associated with said each word after the user queries for said each word. The approach further involves initiating a re-scoring of said each word in the word lattice using the adapted language model to generate a speech recognition output.

Abstract: Disclosed herein are system, method, and computer program product embodiments for robust key value extraction. In an embodiment, one or more hierarchical concepts units (HCUs) may be configured to extract key value and hierarchical information from text inputs. The HCUs may use a convolutional neural network, a recurrent neural network, and feature selectors to analyze the text inputs using machine learning techniques to extract the key value and hierarchical information. Multiple HCUs may be used together and configured to identify different categories of hierarchical information. While multiple HCUs may be used, each may use a skip connection to transmit extracted information to a feature concatenation layer. This allows an HCU to directly send a concept that has been identified as important to the feature concatenation layer and bypass other HCUs.

Abstract: A method, apparatus, and computer program product are therefore provided for disambiguating non-deterministic spoken dialog commands as they relate to points-of-interest and information regarding points-of-interest within or near the field of view of a user. An apparatus may include a processor and a memory including program code instructions.

Abstract: A method, apparatus, and computer program product are therefore provided for disambiguating non-deterministic spoken dialog commands as they relate to points-of-interest and information regarding points-of-interest within or near the field of view of a user. An apparatus may include a processor and a memory including program code instructions. The program code instructions are configured to, when executed, cause the apparatus to at least: receive an indication of a user location; receive an indication of a field of view of the user at the location; receive a natural language utterance; generate, from the natural language utterance, a query, where the query includes at least one request determined from the natural language utterance, the location, and the field of view of the user; determine a point-of-interest associated with the query by conducting a search of a database using the generated query; and provide information associated with the point-of-interest via a user interface.

Abstract: An approach is provided for global voice-based entry of location information. The approach involves partitioning a global speech decoding graph into spatial partitions. The approach also involves determining key entities occurring in each spatial partitions to construct a combined set of key entities. The approach further involves creating a retrieval index to map the key entities in the combined set of key entities to a corresponding partition. A first partition, the combined set of key entities, and the retrieved index are stored in a memory of a device for processing a voice input signal. A second partition that is not in the memory of the device is retrieved based on the combined set of key entities and the retrieval index to automatically re-process the voice input signal when an out-of-vocabulary result is obtained from the first partition.

Abstract: A method, apparatus, and computer program product are therefore provided for disambiguating non-deterministic spoken dialog commands as they relate to points-of-interest and information regarding points-of-interest within or near the field of view of a user. An apparatus may include a processor and a memory including program code instructions.

Abstract: An approach is provided for mobility-based language model adaptation for navigational speech interfaces. The approach involves receiving a word lattice resulting from a speech recognition process applied on a speech input received from a user via a navigational speech interface. The word lattice includes a respective speech recognition score for each word in the word lattice. The approach further involves adapting a language model comprising a plurality of location entities according to a mobility relevance and a search relevance of the plurality of location entities to the user. In one embodiment, the mobility relevance and the search relevance are determined by computing a probability of the user visiting the location entities associated with said each word after the user queries for said each word. The approach further involves initiating a re-scoring of said each word in the word lattice using the adapted language model to generate a speech recognition output.

Abstract: An approach is provided for global voice-based entry of location information. The approach involves partitioning a global speech decoding graph into spatial partitions. The approach also involves determining key entities occurring in each spatial partitions to construct a combined set of key entities. The approach further involves creating a retrieval index to map the key entities in the combined set of key entities to a corresponding partition. A first partition, the combined set of key entities, and the retrieved index are stored in a memory of a device for processing a voice input signal. A second partition that is not in the memory of the device is retrieved based on the combined set of key entities and the retrieval index to automatically re-process the voice input signal when an out-of-vocabulary result is obtained from the first partition.

Abstract: According to some aspects, a method of processing user input received from a user is provided. The method comprises generating a plurality of segmentation hypotheses from content of the user input based, at least in part, on a set of parameters, querying a domain-specific database using each of the plurality of segmentation hypotheses to obtain at least one result, and modifying at least one of the set of parameters based, at least in part, on the at least one result.