Abstract: Embodiments relate to a method, a computer program and a system for optimizing execution of natural language to structured query language. The method comprises the steps of—receiving a natural language text input and performing natural language processing on the natural language text input to generate a plurality of encoded question tokens. Next, the method comprises performing natural language processing on a plurality of table schema stored in a database to generate a plurality of encoded table schema tokens for each table schema. Further, the method comprises determining a similarity between the plurality of encoded question tokens and the plurality of encoded table schema tokens for at least two table schemas of the plurality of table schema. Furthermore, the method comprises determining an output table schema from the plurality of table schema based on the similarity and outputting a natural language string based on the output table schema.

Abstract: A method for providing responsive actions to user inputs in a multi-domain context includes receiving, by a speech-based user interface, a first speech input from a user and converting said first speech input into a text-based representation of the first speech input. A natural language processor processes the text-based representation to determine an intent, entity and internal state of the first speech input. The method further includes determining, by a model-based module based on the intent, entity and internal state, a first data processing policy to apply to the first speech input, wherein the first data processing policy is either a rules-based data processing policy applied by a rules-based module or a statistical model-based data processing policy applied by the model-based module. The first responsive action is generated by the determined first data processing module, and outputted via the speech-based user interface and/or a machine interface.

Abstract: The present disclosure relates to a concept for machine-learning-based prediction of a destination for a user. Based on historic search data associated with the user, at least one candidate destination is determined based on the user and a given context. A plurality of embedding vectors are determined from an embedding matrix, wherein the embedding vectors are associated with the at least one candidate destination, the user, and the given context. The embedding matrix comprising embedding vectors for different components of the historic search data. The plurality of embedding vectors are fed into one or more first neural network layers to generate a semantic embedding for the candidate destination. The semantic embedding is into one or more second neural network layers to generate a probability score for the candidate destination.

Abstract: Embodiments relate to a method, a computer program and a system for optimizing execution of natural language to structured query language. The method for optimizing execution of natural language to structured query language, comprises the steps of receiving a natural language text input and performing natural language processing on the natural language text input to generate a plurality of encoded question tokens. Further the method comprises performing natural language processing on a plurality of table schema stored in a database, to generate a plurality of encoded table schema tokens for each table schema of the plurality of table schema and determining a similarity between the plurality of encoded question tokens and the plurality of encoded table schema tokens for at least two table schemas of the plurality of table schema.

Abstract: The present disclosure relates to a concept for machine-learning-based prediction of a destination for a user. Based on historic search data associated with the user, at least one candidate destination is determined based on the user and a given context. A plurality of embedding vectors are determined from an embedding matrix, wherein the embedding vectors are associated with the at least one candidate destination, the user, and the given context. The embedding matrix comprising embedding vectors for different components of the historic search data. The plurality of embedding vectors are fed into one or more first neural network layers to generate a semantic embedding for the candidate destination. The semantic embedding is into one or more second neural network layers to generate a probability score for the candidate destination.

Abstract: Examples relate to apparatuses, methods and computer programs for a user device and for a server, e.g. to provide a personalized service enabled by converting human language to processing logic. An apparatus for a user device comprises processing circuitry configured to obtain a user command from the user via a user interface of the user device, the user command being obtained in natural language. The processing circuitry is configured to provide the user command in natural language to the server via an interface. The processing circuitry is configured to obtain a response to the user command from the server via the interface, the response to the user command comprising an identifier that is assigned to the user command by the server. The processing circuitry is configured to assign the user command to an element of a graphical user interface of an application being executed by the user device, and to link the identifier to the element of the graphical user interface.

Abstract: A system, method and non-transitory computer-readable medium for predicting a trip destination of a user based on user behavior learning are provided. Historical behaviors and a target behavior of the user are received from a feature processing layer, and the received historical behaviors and the target behavior are embedded with features including a time and a location to produce a context modeling layer. A user modeling layer is produced by embedding the context modeling layer. A trip destination is predicted based on historical trip data and target trip data in the user modeling layer.

Abstract: A system, method and non-transitory computer-readable medium are provided for deep user modeling of user behavior. According to the deep user modeling, user behavior vectors that represent historical user behaviors of a user are determined. Based on a concatenation of the user behavior vectors, a variable-length user behavior matrix is determined. The variable-length user behavior matrix is converted into a fixed-length embedding vector via a long short term memory network, and the fixed-length embedding vector is outputted to the user as a predicted target behavior.

Abstract: A system, method and non-transitory computer-readable medium provided an algorithmic framework for context modeling of user behavior and machine learning of the user behavior in order to optimize user behavior across users, context, and content with different kinds of behaviors. According to the algorithmic framework, context and content modeling optimizes user behavior across users, context, content with different kind of behaviors based on a user behavior matrix.

Abstract: A method for providing responsive actions to user inputs in a multi-domain context includes receiving, by a speech-based user interface, a first speech input from a user and converting said first speech input into a text-based representation of the first speech input. A natural language processor processes the text-based representation to determine an intent, entity and internal state of the first speech input. The method further includes determining, by a model-based module based on the intent, entity and internal state, a first data processing policy to apply to the first speech input, wherein the first data processing policy is either a rules-based data processing policy applied by a rules-based module or a statistical model-based data processing policy applied by the model-based module. The first responsive action is generated by the determined first data processing module, and outputted via the speech-based user interface and/or a machine interface.