公开(公告)号：US20240137394A1

公开(公告)日：2024-04-25

申请号：US18049360

申请日：2022-10-24

Applicant: Spotify AB

Inventor： Joseph Cauteruccio ,  Mehdi Ben Ayed ,  Zhenwen Dai

IPC: H04L65/1069 ,  H04L65/613

CPC classification number: H04L65/1069 ,  H04L65/613

Abstract: Simulator augmented content selection is provided by initializing a content selection object according to session initialization parameter values associated with a simulated media content playback session. The content selection object corresponds to a candidate content selection machine learning model trained to predict selectable content media items for at least one simulated user. A simulated session including a sequence of predicted simulated user next actions and one or more predicted sets of selectable content items are generated by applying a simulated user model to content items identified by the initialized content selection object, where the simulated user model is trained to predict a next action of the simulated user in response to a simulated playback input received from the simulated user and each set of the selectable content items are correlated to each next action in the sequence of predicted simulated user next actions.

公开(公告)号：US12050872B2

公开(公告)日：2024-07-30

申请号：US17526845

申请日：2021-11-15

Applicant: Spotify AB

Inventor： Federico Tomasi ,  Zhenwen Dai ,  Mounia Lalmas-Roelleke

IPC: G06F40/279 ,  G06F16/35 ,  G06F40/284 ,  G06F40/295 ,  G06F40/35 ,  G06F40/40

CPC classification number: G06F40/295 ,  G06F16/355 ,  G06F40/40

Abstract: A system implements a dynamic word correlated topic model (DWCTM) to model an evolution of topic popularity, word embedding, and topic correlation within a set of documents, or other dataset, that spans a period of time. For example, the DWCTM receives the set of documents and a quantity of topics for modeling. The DWCTM processes the set computing, for each topic, various distributions to capture a popularity, word embedding, and correlation with other topics across the period of time. In other examples, a dataset of user listening sessions comprised of media content items for modeling by the DWCTM. Media content metadata (e.g., artist or genre) of the media content items, similar to words of a document, can be modeled by the DWCTM.

公开(公告)号：US20220019750A1

公开(公告)日：2022-01-20

申请号：US16932323

申请日：2020-07-17

Applicant: Spotify AB

Inventor： Praveen Chandar Ravichandran ,  Mounia Lalmas-Roelleke ,  Federico Tomasi ,  Zhenwen Dai ,  Gal Levy-Fix

IPC: G06F40/44 ,  G06F16/35 ,  G06F40/295 ,  G06F17/15

Abstract: A system implements a dynamic correlated topic model (DCTM) to model an evolution of topic popularity, topic representation, and topic correlation within a set of documents, or other dataset, that spans a period of time. For example, the DCTM receives the set of documents and a quantity of topics for modeling. The DCTM processes the set by analyzing words of the documents, identifying word clusters representing the topics, and computing, for each topic, various distributions using continuous processes to capture a popularity, representation, and correlation with other topics across the period of time. In other examples, the dataset are user listening sessions comprised of media content items. Media content metadata (e.g., artist or genre) of the media content items, similar to words of a document, can be analyzed and clustered to represent topics for modeling by the DCTM.

公开(公告)号：US20240119279A1

公开(公告)日：2024-04-11

申请号：US18056101

申请日：2022-11-16

Applicant: Spotify AB

Inventor： Zhenwen Dai ,  Ciarán M. Gilligan-Lee ,  Josh C. Tingey

IPC: G06N3/08

CPC classification number: G06N3/08

Abstract: Contrastive learning is used to learn an alternative embedding. A subtree replacement strategy generates structurally similar pairs of samples from an input space for use in contrastive learning. The resulting embedding captures more of the structural proximity relationships of the input space and improves Bayesian optimization performance when applied to tasks such as fitting and optimization.

公开(公告)号：US12259950B2

公开(公告)日：2025-03-25

申请号：US17063606

申请日：2020-10-05

Applicant: Spotify AB

Inventor： Zhenwen Dai ,  Praveen Chandar Ravichandran ,  Ghazal Fazelnia ,  Benjamin Carterette ,  Mounia Lalmas-Roelleke

IPC: G06N20/00 ,  G06F18/20 ,  G06F18/21 ,  G06F18/2113 ,  G06N7/01

Abstract: Disclosed examples include an automated online experimentation mechanism that can perform model selection from a large pool of models with a relatively small number of online experiments. The probability distribution of the metric of interest that contains the model uncertainty is derived from a Bayesian surrogate model trained using historical logs. Disclosed techniques can be applied to identify a superior model by sequentially selecting and deploying a list of models from the candidate set that balance exploration-exploitation.

公开(公告)号：US20240236155A9

公开(公告)日：2024-07-11

申请号：US18049360

申请日：2022-10-25

Applicant: Spotify AB

Inventor： Joseph Cauteruccio ,  Mehdi Ben Ayed ,  Zhenwen Dai

IPC: H04L65/1069 ,  H04L65/613

CPC classification number: H04L65/1069 ,  H04L65/613

Abstract: Simulator augmented content selection is provided by initializing a content selection object according to session initialization parameter values associated with a simulated media content playback session. The content selection object corresponds to a candidate content selection machine learning model trained to predict selectable content media items for at least one simulated user. A simulated session including a sequence of predicted simulated user next actions and one or more predicted sets of selectable content items are generated by applying a simulated user model to content items identified by the initialized content selection object, where the simulated user model is trained to predict a next action of the simulated user in response to a simulated playback input received from the simulated user and each set of the selectable content items are correlated to each next action in the sequence of predicted simulated user next actions.

公开(公告)号：US20230419187A1

公开(公告)日：2023-12-28

申请号：US17851426

申请日：2022-06-28

Applicant: Spotify AB

Inventor： Zhenwen Dai ,  Joseph Cauteruccio ,  Federico Tomasi ,  Mehdi Ben Ayed

IPC: G06N20/10

CPC classification number: G06N20/10

Abstract: Methods, systems and computer program products are provided for content generation. A distribution of policies is defined based on an action space. Distribution parameters are received from a reinforcement learning (RL) algorithm. In turn, a policy is randomly sampled from the distribution of policies. A candidate content item is generated using the sampled policy. A quality of the candidate content item is measured based on a predefined quality criteria and a parameter model is adjusted as specified by the reinforcement learning algorithm to obtain a plurality of updated distribution parameters. Environment settings are passed to a trained parameter model to obtain a plurality of policy distribution parameters. A predetermined number of policies from the distribution of policies are then sampled and the plurality of environment settings are passed to the predetermined number of sampled policies to obtain at least one content item.

公开(公告)号：US11727221B2

公开(公告)日：2023-08-15

申请号：US16932323

申请日：2020-07-17

Applicant: Spotify AB

Inventor： Praveen Chandar Ravichandran ,  Mounia Lalmas-Roelleke ,  Federico Tomasi ,  Zhenwen Dai ,  Gal Levy-Fix

IPC: G06F40/44 ,  G06F40/295 ,  G06F16/35 ,  G06F40/20 ,  G06F40/30 ,  G06F17/15

CPC classification number: G06F40/44 ,  G06F16/355 ,  G06F17/15 ,  G06F40/20 ,  G06F40/295 ,  G06F40/30

Abstract: A system implements a dynamic correlated topic model (DCTM) to model an evolution of topic popularity, topic representation, and topic correlation within a set of documents, or other dataset, that spans a period of time. For example, the DCTM receives the set of documents and a quantity of topics for modeling. The DCTM processes the set by analyzing words of the documents, identifying word clusters representing the topics, and computing, for each topic, various distributions using continuous processes to capture a popularity, representation, and correlation with other topics across the period of time. In other examples, the dataset are user listening sessions comprised of media content items. Media content metadata (e.g., artist or genre) of the media content items, similar to words of a document, can be analyzed and clustered to represent topics for modeling by the DCTM.