Methods, systems, and apparatuses for improved content recommendations
Methods, systems, and apparatuses for improved content recommendations are described herein. A distribution platform may comprise a system of computing devices, servers, software, etc., that is configured to present media assets (e.g., content) at user devices. In one example embodiment, an analytics subsystem may provide at least one content recommendation to a user device using a classification model. In another example embodiment, the analytics subsystem may train the classification model. In a further example embodiment, the analytics subsystem may provide at least one fallback content recommendation when a recommendation provided by the classification model does not satisfy a threshold level of interest.
1 . A method comprising:
receiving, by an analytics subsystem of a computing device, activity data indicative of a plurality of engagements of a plurality of user devices with a plurality of media assets via a client application and a corresponding plurality of user profiles wherein the client application causes each engagement of the plurality of engagements to be synchronized with a corresponding user profile of the plurality of user profiles;
generating, based on the activity data, a user interest cloud associated with a first user profile that is not among the plurality of user profiles, wherein the user interest cloud is generated based on at least one content feature of a plurality of content features and at least one interest attribute of a plurality of interest attributes associated with each media asset of the plurality of media assets;
determining, by a classification model, based on the plurality of engagements and the user interest cloud associated with the first user profile, a certainty match for at least one content recommendation for the first user profile, wherein the at least one content recommendation comprises at least one media asset of the plurality of media assets, wherein the certainty match is indicative of a predicted level of interest for the at least one content recommendation, and wherein the classification model is associated with the client application;
determining, based on the certainty match, that the predicted level of interest meets or exceeds a threshold level of interest for the at least one content recommendation; and
causing, based on the predicted level of interest meeting or exceeding the threshold level of interest, a first user device associated with the first user profile to output the at least one content recommendation via the client application.
2 . The method of claim 1 , wherein determining the at least one content recommendation comprises determining, based on the activity data, that the first user device has not interacted with the at least one media asset.
3 . The method of claim 1 , wherein the plurality of content features comprises a content type, a content rating, content metadata, a date of creation, a content tag, a content category, a content filter, a language, or one or more words of a content description, and wherein the plurality of interest attributes comprises a numerical indication of a level of interest associated with each media asset of the plurality of media assets or a textual indication of the level of interest associated with each media asset of the plurality of media assets.
4 . The method of claim 1 , wherein the activity data is received in real-time corresponding to the plurality of engagements.
5 . The method of claim 1 , wherein the plurality of engagements comprises at least one of:
a plurality of user interactions with a user interface of the client application during output of the plurality of media assets;
a quantity of time that each media asset of the plurality of media assets was output at the corresponding user device;
a quantity of mute actions performed by the corresponding user device during output of any of the plurality of media assets;
a level of volume associated with output of any of the plurality of media assets at the corresponding user device; or
a duration of inactivity of the client application during output of any of the plurality of media assets.
6 . The method of claim 1 , wherein generating the user interest cloud comprises at least one of:
determining, based on the activity data and the plurality of engagements, a feature vector associated with the first user device, wherein the feature vector comprises the at least one content feature and at least one engagement feature associated with each media asset of the plurality of media assets, wherein the at least one engagement feature of each feature vector comprises at least one of: a quantification of an engagement with each media asset or a numerical weight associated with an engagement feature; or
determining, by a scoring model, the at least one interest attribute, wherein the at least one interest attribute comprises at least one of: a numerical indication of a level of interest associated with each media asset or a textual indication of the level of interest associated with each media asset.
7 . The method of claim 1 , wherein causing the first user device to output the at least one content recommendation comprises at least one of:
causing the first user device, via the client application, to output a user interface object associated with the at least one content recommendation;
causing a notification associated with the at least one content recommendation to be output at the first user device; or
causing the first user device, via the client application, to output the at least one media asset associated with the at least one content recommendation.
8 . A method comprising:
receiving, by an analytics subsystem of a computing device, activity data indicative of a plurality of engagements of a plurality of user devices with at least one media asset of a plurality of media assets via a client application and a corresponding plurality of user profiles, wherein the client application causes each engagement of the plurality of engagements to be synchronized with a corresponding user profile of the plurality of user profiles;
updating, based on the activity data, a user interest cloud associated with a first user profile that is not among the plurality of user profiles, wherein the user interest cloud is generated based on the activity data and at least one content feature, of a plurality of content features, and at least one interest attribute, of a plurality of interest attributes, associated with each media asset of the plurality of media assets;
determining at least one triggering event associated with the client application; and
retraining, based on the at least one triggering event, a trained classification model, wherein the trained classification model is associated with the client application.
9 . The method of claim 8 , wherein the activity data is received in real-time corresponding to the plurality of engagements.
10 . The method of claim 8 , wherein the plurality of engagements comprises at least one of:
a user interaction with a user interface of the client application during output of the at least one media asset;
a quantity of time that the at least one media asset was output at a user device;
a quantity of mute actions performed by a user device during output of the at least one media asset;
a level of volume associated with output of the at least one media asset at a user device; or
a duration of inactivity of the client application during output of the at least one media asset.
11 . The method of claim 8 , wherein determining the at least one triggering event associated with the client application comprises:
determining, based on a threshold quantity of time that the client application is inactive at a user device, the at least one triggering event.
12 . The method of claim 8 , wherein determining the at least one triggering event associated with the client application comprises:
determining, based on an expiration of a quantity of time associated with training the classification model, the at least one triggering event.
13 . The method of claim 8 , wherein determining the at least one triggering event associated with the client application comprises at least one of:
determining, based on a threshold quantity of new media assets associated with the client application, the at least one triggering event.
14 . The method of claim 8 , wherein determining the at least one triggering event associated with the client application comprises:
determining, based on a plurality of further activity data, the at least one triggering event.
15 . A method comprising:
receiving, by an analytics subsystem of a computing device, activity data indicative of a plurality of interactions of a plurality of user devices with a plurality of media assets via a client application and a corresponding plurality of user profiles, wherein the client application causes each interaction of the plurality of interactions to be synchronized with a corresponding user profile of the plurality of user profiles;
generating, based on the activity data, a user interest cloud comprising a plurality of content features and corresponding interest attributes associated with the plurality of media assets, wherein the user interest cloud is associated with a first user profile that is not among the plurality of user profiles, wherein the user interest cloud is generated based on at least one content feature of the plurality of content features and at least one interest attribute of the corresponding interest attributes associated with the plurality of media assets;
determining, by a classification model, based on the user interest cloud associated with the first user profile, a certainty match for at least one content recommendation for the first user profile, wherein the certainty match is indicative of a predicted level of interest for the at least one content recommendation, wherein the classification model is associated with the client application;
determining, based on the certainty match, that the predicted level of interest does not meet or exceed a threshold level of interest for the at least one content recommendation; and
causing, based on the predicted level of interest not meeting or exceeding the threshold level of interest, a first user device to output at least one fallback content recommendation, wherein the first user device is associated with the first user profile.
16 . The method of claim 15 , wherein the corresponding interest attributes comprise at least one of: a numerical indication of a level of interest associated with each media asset of the plurality of media assets or a textual indication of the level of interest associated with each media asset of the plurality of media assets.
17 . The method of claim 15 , wherein the at least one content recommendation comprises at least one media asset.
18 . The method of claim 17 , wherein determining the certainty match for the at least one content recommendation comprises determining, by the classification model, based on the plurality of content features and the corresponding interest attributes, and based on at least one content feature associated with the at least one media asset, the certainty match for the at least one content recommendation.
19 . The method of claim 15 , wherein the threshold level of interest for the at least one content recommendation is less than each of the corresponding interest attributes.
20 . The method of claim 15 , wherein the at least one fallback content recommendation comprises at least one of: a searchable library of media assets, a curated plurality of media assets, a media asset associated with a high interest attribute for a plurality of other user devices, at least one media asset with which the first user device has not previously interacted, or at least one external media asset.