Systems and methods for rendering near-real-time embedding models for personalized news recommendations

1 . A computer-implemented method for identifying content to recommend to a user, the method comprising:

creating a user interest profile, by a server processor of a computation device, by aggregating user interaction data associated with a user over a period of time, the user interaction data including one or more user interactions of a first type and one or more user interest features;

training, by the server processor, an embedding model to overlap the user interaction data with the one or more user interaction features and embed text features of the one or more user interest features into a vector space with a fixed length representation in order to overcome cold-start problems, the text features including click or skip labels, based on historical click or skip user feedback, and time decay factors to further train the embedding model to remove old text features;

generating, by a machine learning model processed by the server processor, a content embedding vector based on the embedding model, the content embedding vector including a mean pooling of words extracted from one or more content articles;

generating, by the server processor, an embedding profile vector based on the embedding model, the embedding profile vector including a vector representation of one or more user interaction data, the vector representation including the fixed length representation;

training, by the server processor, a machine learning prediction model based on the embedding profile vector and the content embedding vector;

receiving, by the server processor, a user search query;

determining in real-time, by the trained machine learning prediction model and based on the user search query, a predicted list of one or more content items of interest for recommending to the user and a corresponding relevance score for each of the one or more content items of interest; and

transmitting, over a network, to a device of the user for display, one of the one or more content items of interest based on the relevance score for each of the one or more content items of interest.

2 . The computer-implemented method of claim 1 , wherein:

the one or more user interactions of the first type include consecutive clicks of the user on one or more content articles within a predefined period of time.

3 . The computer-implemented method of claim 2 , wherein:

the predefined period of time includes within a twenty-four hour period of time.

4 . The computer-implemented method of claim 1 , wherein the embedding profile vector is multiplied by a decay factor, the decay factor configured to decrease a value of one or more past interactions over time.

5 . The computer-implemented method of claim 4 , wherein the decay factor is selected by a heuristic approach and is set as a hyperparameter.

6 . The computer-implemented method of claim 1 , wherein the one or more user interaction data of the first type includes: media streaming, search query, menu navigation, electronic messaging, or user application preference setting.

7 . The computer-implemented method of claim 1 , further comprising generating one or more of: a user recommendation, a user notification, and a user profile customization based on the trained machine learning prediction model.

8 . A system for profile partition generation comprising:

at least one processor; and

a storage device that stores a set of instructions, the set of instructions being executable by the at least one processor to cause the at least one processor to implement steps comprising:

creating a user interest profile, by a server processor of a computation device, by aggregating user interaction data associated with a user over a period of time, the user interaction data including one or more user interactions of a first type and one or more user interest features;

training, by the server processor, an embedding model to overlap the user interaction data with the one or more user interaction features and embed text features of the one or more interest features into a vector space with a fixed length representation in order to overcome cold-start problems, the text features including click or skip labels, based on historical click or skip user feedback, and time decay factors to further train the embedding model to remove old text features;

generating, by a machine learning model processed by the server processor, a content embedding vector based on the embedding model, the content embedding vector including a mean pooling of words extracted from one or more content articles;

generating, by the server processor, an embedding profile vector based on the embedding model, the embedding profile vector including a vector representation of one or more user interaction data, the vector representation including the fixed length representation;

training, by the server processor, a machine learning prediction model based on the embedding profile vector and the content embedding vector;

receiving, by the server processor, a user search query;

determining in real-time, by the trained machine learning prediction model and based on the user search query, a predicted list of one or more content items of interest for recommending to the user and a corresponding relevance score for each of the one or more content items of interest; and

transmitting, over a network, to a device of the user for display, one of the one or more content items of interest based on the relevance score for each of the one or more content items of interest.

9 . The system of claim 8 , wherein:

the one or more user interactions of the first type include consecutive clicks of the user on one or more content articles within a predefined period of time.

10 . The system of claim 9 , wherein:

the predefined period of time includes within a twenty-four hour period of time.

11 . The system of claim 8 , wherein the embedding profile vector is multiplied by a decay factor, the decay factor configured to decrease a value of one or more past interactions over time.

12 . The system of claim 11 , wherein the decay factor is selected by a heuristic approach and is set as a hyperparameter.

13 . The system of claim 8 , wherein the one or more user interaction data of the first type includes: media streaming, search query, menu navigation, electronic messaging, or user application preference setting.

14 . The system of claim 8 , further comprising:

generating one or more of: a user recommendation, a user notification, and a user profile customization based on the trained machine learning prediction model.

15 . A non-transitory computer readable medium storing instructions which, when executed by one or more processors, cause the one or more processors to perform operations for profile partition generation, the operations comprising:

creating a user interest profile, by a server processor of a computation device, by aggregating user interaction data associated with a user over a period of time, the user interaction data including one or more user interactions of a first type and one or more user interest features;

training, by the server processor, an embedding model to overlap the user interaction data with the one or more user interaction features and embed text features of the one or more user interest features into a vector space with a fixed length representation in order to overcome cold-start problems, the text features including click or skip labels, based on historical click or skip user feedback, and time decay factors to further train the embedding model to remove old text features;

generating, by a machine learning model processed by the server processor, a content embedding vector based on the embedding model, the content embedding vector including a mean pooling of words extracted from one or more content articles;

generating, by the server processor, an embedding profile vector based on the embedding model, the embedding profile vector including a vector representation of one or more user interaction data, the vector representation including the fixed length representation;

training, by the server processor, a machine learning prediction model based on the embedding profile vector and the content embedding vector;

receiving, by the server processor, a user search query;

determining in real-time, by the trained machine learning prediction model and based on the user search query, a predicted list of one or more content items of interest for recommending to the user and a corresponding relevance score for each of the one or more content items of interest; and

transmitting, over a network, to a device of the user for display, one of the one or more content items of interest based on the relevance score for each of the one or more content items of interest.

16 . The non-transitory computer readable medium of claim 15 , wherein:

the one or more user interactions of the first type include consecutive clicks of the user on one or more content articles within a predefined period of time.

17 . The non-transitory computer readable medium of claim 16 , wherein:

the predefined period of time includes within a twenty-four hour period of time.

18 . The non-transitory computer readable medium of claim 15 , wherein the embedding profile vector is multiplied by a decay factor, the decay factor configured to decrease a value of one or more past interactions over time.

19 . The non-transitory computer readable medium of claim 15 , wherein the one or more user interaction data of the first type includes: media streaming, search query, menu navigation, electronic messaging, or user application preference setting.

20 . The non-transitory computer readable medium of claim 15 , further comprising:

generating one or more of: a user recommendation, a user notification, and a user profile customization based on the trained machine learning prediction model.