Roku, Inc.

Twilio Inc.

Abhishek Bambha

Fei Xiao

Nam Vo

Ronica Jethwa

Rohit Mahto

Zidong Wang

Jose Sanchez

Pulkit Aggarwal

Lian Liu

Jing Ye

Jing Lu

Khaldun Aidarabsah

Claire Electra Longo

Amit Verma

Elizabeth Yee

Atishay Jain

Rameen Mahdavi

Brendon Kyle Villalobos

Liyuan Zhang

Jorge Chang

Jing Xie

Rakesh Ravuru

Anirban Das

Andrey Vlasenko

Mehul Agrawal

Disclosed herein are system, apparatus, article of manufacture, method and/or computer program product aspects, and/or combinations and sub-combinations thereof, for generating short-form content. An example aspect operates by analyzing a media file in a library using a machine learning model. To analyze the media file, the embodiment determines, using the machine learning model, a first portion of the media file that has a feature that satisfies a classification that the machine learning model is configured to identify. The embodiment tags the first portion using one or more position tags indicative of a beginning of the first portion of the media file or an end of the first portion of the media file. The embodiment then generates a segment from the media file based on the one or more position tags. The segment comprises the portion of the media file and excludes one or more second portions of the media file.

Generating short-form content from full-length media using a machine learning model

A set of content items can be accessed by a community of users having a set of interests. A set of interest based clusters for the set of content items correspond to the set of interests. For a user, a recommendation system can determine a group of user interest clusters selected from the set of interest based clusters. A popularity score for each content item of the set of content items with respect to the community of users can be generated, and an interest based popularity score for a content item within the interest based cluster can be generated based on a rank of the content item based on the popularity score of the content item. Recommendation candidates for the user can be generated based on the interest based popularity score of the content item for each content item in the group of user interest clusters.

Interest based recommendation system

Disclosed herein are system, apparatus, article of manufacture, method and/or computer program product embodiments, and/or combinations and sub-combinations thereof, for candidate ranking for content recommendation. An embodiment operates by receiving category candidates over a network, wherein each of the category candidates comprises content candidates associated with one or more applications operating on media devices. The embodiment then ranks the category candidates based on a machine model trained using a learning algorithm based on the time series data, and ranks the content candidates in the each of category candidates based on the time series data. The embodiment then causes the ranked category candidates and the ranked content candidates to be outputted for display.

Candidate ranking for content recommendation

Disclosed herein are system, apparatus, article of manufacture, method and/or computer program product embodiments, and/or combinations and sub-combinations thereof, for generating a scene emotion value for a scene based on a sequence of frame emotion values for a sequence of frames within the scene of a content. The content can include multiple scenes, and a scene can include multiple frames, where a frame emotion value can be generated for each frame. A frame emotion value can be generated based on scene metadata related to the scene, content metadata related to the content, and a frame metadata related to the frame.

Emotion evaluation of contents

Disclosed herein are system, method and/or computer program product embodiments, and/or combinations and sub-combinations thereof, for personalizing a user interface (UI) of a media device and/or content presented thereby. An example embodiment operates by obtaining a first natural language user input, providing the first natural language user input to a personalization language model that is configured to interpret different natural language user inputs to respectively determine different update tasks invoked thereby, the different update tasks including a UI update task and a content update task, receiving from the model a first update task determined thereby based at least on the first natural language user input, generating one or more first application programming interface (API) calls based on the first update task, and placing the one or more first API calls to a service that implements the first update task based on the one or more first API calls.

Media device user interface and content personalization using natural language prompts

Disclosed herein are various embodiments, for a content display and clustering system. An example embodiment operates by receiving a request to display the plurality of content items. At each of multiple levels different pairs of content items are identified and a similarity score is computed for each pair. A subset of pairs for which their similarity score exceeds a similarity threshold for the respective level are identified and clustered. This process is repeated for one or more iterations at the same level, and then the process is repeated for each of the multiple levels. A final clustered subset is identified, and output for display, responsive to the request to display the plurality of content items.

Content display and clustering system

Disclosed are mechanisms for selecting a recommended item for a current item being viewed by a user account based on a view history of the user account with reduced bias. For a current item being viewed by the user account represented by a current node of a co-watch graph, embodiments can select a recommended item represented by an associated node in the co-watch graph likely being viewed by the user account, and determine a probability of the recommended item likely being viewed. The co-watch graph can be generated based on a view history of the user account. An edge between a first node and a second node of the co-watch graph can have a weight representing a number of co-occurrence times when the first item represented by the first node and the second item represented by the second node are viewed in sequence within a predetermined time interval.

Recommendation system with reduced bias based on a view history

Disclosed herein are system, method and/or computer program product embodiments, and/or combinations thereof, for training a conversational recommendation system. An embodiment generates a probabilistic pseudo-user neural network model based on at least one interest probability distribution corresponding to a pseudo-user profile. The embodiment trains, using the pseudo-user neural network model, the conversational recommendation system to learn a recommendation policy, where the conversational recommendation system includes an interest-exploration engine and a prompt-decision engine. The training includes performing an iterative learning process that includes selecting an interest-exploration strategy based on one or more of the following: an interest-exploration policy, an earlier pseudo-user response generated by the pseudo-user neural network model, content data, and pseudo-user interaction history. The embodiment then generates, using the trained conversational recommendation system, a real-time recommendation having high play probability based on the minimal number of iterations of conversation between a user and the trained conversational recommendation system.

Interest-based conversational recommendation system

Disclosed herein are system, apparatus, article of manufacture, method, and/or computer program product embodiments, and/or combinations and sub-combinations thereof, for demographic predictions for content items. An example embodiment operates by assigning weights representing demographics to a first plurality of nodes of a predictive model and assigning predictive values representing predicted demographics to a second plurality of nodes of the model. Pairwise distances between the predictive values for the nodes of the second plurality of nodes and the weighted values of the first plurality of nodes may be calculated and the shortest calculated pairwise distances may be used to assign demographics for content items corresponding to nodes of the first plurality of nodes to content items corresponding nodes of the second plurality of nodes. When content is requested, a content item for which the same demographic has been assigned may be recommended to the requestor.

Demographic predictions for content items

Disclosed herein are system, apparatus, article of manufacture, method and/or computer program product aspects, and/or combinations and sub-combinations thereof, for generating trailers (previews) for multimedia content. An example aspect operates by generating an initial set of candidate points to generate a trailer for a media content; determining conversion data for each of the initial set of candidate points; determining an updated set of candidate points based on the conversion data; determining an estimated mean and upper bound for each of the updated set of candidate points; computing a value for each of the updated set of candidate points; generating a ranked list based on the value computed for each of the updated set of candidate points; and repeating the process until an optimal candidate point is converged upon.

Automated trailer generation

Disclosed herein are system, apparatus, article of manufacture, method and/or computer program product embodiments, and/or combinations and sub-combinations thereof, for a content acquisition system to recommend for acquisition a subset of content items selected from a set of content items available for purchase in relation to a content recommendation system currently used in a media environment. The content acquisition system may include a content recommendation system simulator to estimate an impact function value for a potential subset of content items of the set of content items available for purchase based on the currently used content recommendation system. Afterwards, an acquisition recommender can recommend for acquisition a subset of content items based on an optimized objective function value calculated based on an optimization model while meeting one or more budget constraints.

Content acquisition system

Methods, systems, and computer programs are presented for the determination of optimal communication scheduling. One method includes an operation for training a machine-learning program to generate a frequency model that determines a frequency for sending communications to users. The training utilizes training data defined by features related to user information and responses of users to previous communications to the users. The method further includes determining, by the frequency model and based on information about a first user, a first frequency for the first user. The first frequency identifies the number of communications to transmit to the first user per period of time. Further, the method includes operations for receiving a communication request to send one or more communications to the first user and determining send times for the one or more communications to the first user based on the first frequency. The communications are sent at the determined send times.

Growing community of inventors

Burlingame, CA, United States of America

Abhishek Bambha