Amazon Technologies, Inc.

Sven Daehne

Janet Ellen Galore

Dominick Khanh Pham

Mike Dodge

Damon Jon Ganem

Charles Shearer Dorner

Robert Yuji Haitani

Techniques described herein include systems and methods for evaluating an unsupervised machine learning model. In some embodiments, the system identifies item-to-item similarity values based on historical transaction data. The system may also generate collection data for a number of users based on the historical transaction data. Similarity matrices may be created for each pair of users that include rows associated with a first collection and columns associated with a second collection. Each data field in the similarity matrix may indicate an item-to-item similarity value as identified by the system. In some embodiments, a similarity score may be calculated for the user pair based on the item-to-item similarity values included in the similarity matrix. In some embodiments, the system may generate a graphical summary representation of the similarity matrix.

Evaluating unsupervised learning models

Embodiments herein describe a video editor that can identify and track objects (e.g., products) in a video. The video editor identifies a particular object in one frame of the video and tracks the location of the object in the video. The video editor can update a position of an indicator that tracks the location of the object in the video. In addition, the video editor can identify an identification (ID) of the object which the editor can use to suggest annotations that provide additional information about the object. Once modified, the video is displayed on a user device, and when the viewer sees an object she can is interested in, she can pause the video which causes the indicator to appear. The user can select the indicator which prompts the user device to display the annotations corresponding to the object.

Tagging tracked objects in a video with metadata

A video segmenting system identifies a product for sale in a video and determines one or more attributes of audio and video content within the video. The video segmenting system determines a video segment within the video that is associated with the product for sale, based on the attributes. The video segmenting system generates a tag that associates the product for sale with the video segment and sends an indication of the tag to a user device. Once the video is played on a user device, the user device detects a search query about the product for sale. Using the tag, the user device can display a marker on the user device corresponding to the location of the video segment within the video.

Generating video segments based on video metadata

A video tagging system that can generate tags corresponding to associations of object-related keywords mentioned in a video to time instances in the video is described. The video tagging system identifies a particular object associated with a video. Using a transcription of audio content within the video, the video tagging system determines a keyword mentioned in the audio content that is associated with the object and a time instance within a timeline of the video when the keyword is mentioned. The video tagging system generates a tag that associates the keyword with the time instance and sends an indication of the tag to a user device. Once the video is displayed on the user device, the user can search for the keyword. This prompts the user device to display a marker indicating the time instance when the keyword is mentioned.

Growing community of inventors

Seattle, WA, United States of America

Sven Daehne