Michael Perkowitz

Bio: Michael Perkowitz is an academic researcher from Amazon.com. The author has contributed to research in topics: Context (language use) & Computer user satisfaction. The author has an hindex of 10, co-authored 16 publications receiving 455 citations.

Calendaring location-based events and associated travel

Abstract: A user interface for an electronic calendar represents different locations or different users or different user calendars in different portions of the display. Calendar entries can be associated with one or more locations, one or more users, and with one or more user calendars. The different locations may reside in different time zones and a timeline for each time zone is displayed. The position of the calendar entry provides a visual identifier of the timeline with which the event is associated. Travel time to and from events in the calendar are calculated for calendared events and shown adjacent to the beginning and end of the event. A user's future location at a point in time is inferred from patterns in the user's locations and by analyzing the user's calendared events and correspondence in order to calculate travel time to calendared events.

Contextual information provider

Abstract: A user's context history is used to help select contextual information to provide to the user. Context data describing the user's current context is received and a plurality of information items corresponding to the user's current context are identified from a contextual information corpus. A personalized user behavior model for the user is applied to determine the likelihood that each of the identified information items will be of value to the user. One or more of the information items are selected based on the corresponding likelihoods and the selected information items are provided for presentation to the user.

Social Network Model for Semantic Processing

Abstract: A social network model, based on data relevant to a user, is used for semantic processing to enable improved entity recognition among text accessed by the user. An entity extraction module of the server, with reference to a general training corpus, general gazetteers, user-specific gazetteers, and entity models, parses text to identify entities. The entities may be, for example, people, organizations, or locations. A social network module of the server builds the social network model implicit in the data accessed by the user. The social network model includes the relationships between entities and an indication of the strength of each relationship. The social network module is also used to disambiguate names and unify entities based on the social network model.

Annotating Large Email Datasets for Named Entity Recognition with Mechanical Turk

Abstract: Amazon's Mechanical Turk service has been successfully applied to many natural language processing tasks. However, the task of named entity recognition presents unique challenges. In a large annotation task involving over 20,000 emails, we demonstrate that a competitive bonus system and inter-annotator agreement can be used to improve the quality of named entity annotations from Mechanical Turk. We also build several statistical named entity recognition models trained with these annotations, which compare favorably to similar models trained on expert annotations.

Electronic message prioritization

Abstract: A system, method, and computer program product dynamically prioritizes electronic messages. An electronic message having one or more properties is received. These message properties can include, a particular sender or body text. Information describing past activity of a recipient user of the electronic message is accessed. A priority is determined for the electronic message, where the determining is based at least in part on a comparison of a property of the electronic message with the accessed information. The priority determination may include detecting the presence of a request in the electronic message, determining the social weight of the sender of the electronic message, determining the temporal urgency of the electronic message, or determining the relevance of the electronic message, for example. An indication of the priority of the message is presented to the recipient user.

A Survey of Crowdsourcing Systems

Abstract: Crowd sourcing is evolving as a distributed problem-solving and business production model in recent years. In crowd sourcing paradigm, tasks are distributed to networked people to complete such that a company's production cost can be greatly reduced. In 2003, Luis von Ahn and his colleagues pioneered the concept of "human computation", which utilizes human abilities to perform computation tasks that are difficult for computers to process. Later, the term "crowdsourcing" was coined by Jeff Howe in 2006. Since then, a lot of work in crowd sourcing has focused on different aspects of crowd sourcing, such as computational techniques and performance analysis. In this paper, we give a survey on the literature on crowd sourcing which are categorized according to their applications, algorithms, performances and datasets. This paper provides a structured view of the research on crowd sourcing to date.

Electronic Devices with Voice Command and Contextual Data Processing Capabilities

Abstract: An electronic device may capture a voice command from a user. The electronic device may store contextual information about the state of the electronic device when the voice command is received. The electronic device may transmit the voice command and the contextual information to computing equipment such as a desktop computer or a remote server. The computing equipment may perform a speech recognition operation on the voice command and may process the contextual information. The computing equipment may respond to the voice command. The computing equipment may also transmit information to the electronic device that allows the electronic device to respond to the voice command.

Voice trigger for a digital assistant

Abstract: A method for operating a voice trigger is provided. In some implementations, the method is performed at an electronic device including one or more processors and memory storing instructions for execution by the one or more processors. The method includes receiving a sound input. The sound input may correspond to a spoken word or phrase, or a portion thereof. The method includes determining whether at least a portion of the sound input corresponds to a predetermined type of sound, such as a human voice. The method includes, upon a determination that at least a portion of the sound input corresponds to the predetermined type, determining whether the sound input includes predetermined content, such as a predetermined trigger word or phrase. The method also includes, upon a determination that the sound input includes the predetermined content, initiating a speech-based service, such as a voice-based digital assistant.

Creating Speech and Language Data With Amazon's Mechanical Turk

Abstract: In this paper we give an introduction to using Amazon's Mechanical Turk crowdsourcing platform for the purpose of collecting data for human language technologies. We survey the papers published in the NAACL-2010 Workshop. 24 researchers participated in the workshop's shared task to create data for speech and language applications with $100.