Techniques for modifying training data for video response quality optimization are provided. In one technique, training data is identified that is generated based on video presentation data that indicates multiple video items were presented to multiple entities. The training data comprises multiple training instances, each indicating a presentation of at least a portion of a video item to an entity. For each training instance in a subset of the training instances, a quality metric of the presentation of the video item indicated in said each training instance is computed and that training instance is modified based on the quality metric. After modifying one or more of the training instances, the model is trained using one or more machine learning techniques. In response to a content request, the model is used to determine whether to transmit a particular video item over a network to a computing device of a particular entity.

Systems and methods are described herein for providing content item recommendations based on a video. Using feature vectors corresponding to at least one frame of a video (e.g., generated based on texture and shape intensity of a frame), a recommendation system improves content recommendation using analytic and quantitative characteristics derived from a frame of a content item rather than merely manually labeled bibliographic data (e.g., a genre or producer). The recommendation system may generate a feature vector based on a texture, a shape intensity (e.g., generated from a Generalized Hough Transform), and temporal data corresponding to at least one frame of a video. The feature vector is analyzed using a machine learning model (e.g., a neural network) to produce a machine learning model output. The recommendation system causes a recommended content item to be provided based on the machine learning model output.

Systems and methods are described herein for providing content item recommendations based on a video. Using feature vectors corresponding to at least one frame of a video (e.g., generated based on texture and shape intensity of a frame), a recommendation system improves content recommendation using analytic and quantitative characteristics derived from a frame of a content item rather than merely manually labeled bibliographic data (e.g., a genre or producer). The recommendation system may generate a feature vector based on a texture, a shape intensity (e.g., generated from a Generalized Hough Transform), and temporal data corresponding to at least one frame of a video. The feature vector is analyzed using a machine learning model (e.g., a neural network) to produce a machine learning model output. The recommendation system causes a recommended content item to be provided based on the machine learning model output.

Aspects of the present disclosure aim to improve upon methods and systems for the incorporation of additional material into source video data. In particular, the method of the present disclosure may use a pre-existing corpus of source video data to produce, test and refine a prediction model for enabling the prediction of the characteristics of placement opportunities. The model may be created using video analysis techniques which obtain metadata regarding placement opportunities and also through the identification of categorical characteristics relating to the source video which may be provided as metadata with the source video, or obtaining through image processing techniques described below. Using the model, the method and system may then be used to create a prediction of insertion zone characteristics for projects for which source video is not yet available, but for which information corresponding to the identified categorical characteristics is known.

This disclosure relates to methods, non-transitory computer readable media, and systems that determine multiple personas corresponding to a user account for digital content and train a persona classifier to predict a given persona (from among the multiple personas) for content requests associated with the user account. By using the persona classifier, the disclosed methods, non-transitory computer readable media, and systems accurately detect a given persona for a content request upon initiation of the request. Based on determining the given persona, in some implementations, the methods, non-transitory computer readable media, and systems generate a digital-content recommendation for presentation on a client device associated with the user account.

Example methods, apparatus, systems and articles of manufacture (e.g., physical storage media) to perform probabilistic modeling for anonymized data integration and measurement of sparse and weakly-labeled datasets are disclosed. An apparatus includes a training controller to train a neural network to produce a trained neural network to output model parameters of a probability model, a model evaluator to execute the trained neural network on input data specifying a time of day, a media source, and at least one feature different from the time of day and the media source to determine one or more first model parameters of the probability model, and a ratings metric generator to evaluate the probability model based on input census data to determine a ratings metric corresponding to the time of day, the media source, and the at least one feature, the probability model configured with the one or more first model parameters.

An example to predict demographics for impressions includes a prediction manager to: determine that first demographic data corresponding to a first database proprietor subscriber does not match second demographic data corresponding to a media delivery device, both the first and second demographic data corresponding to an impression; obtain third demographic data corresponding to an Internet protocol address, the third demographic data obtained from a second database proprietor; and generate matched demographic data based on comparing the third demographic data to the first demographic data; and a modeler to generate a prediction model based on the matched demographic data, the prediction model to predict fourth demographic data for the impression.

An intelligent return path data (iRPD) system enables transmission of return path data via a communication network for a television connected to a non-hybrid set top box (STB). The iRPD system is configured to receive the key codes of the keys pressed on a remote control device along with the date time stamps and the location information. The iRPD system analyzes the keypress data along with the date time stamps to recognize the channels accessed in programming operations and the non-programming control operations executed by a viewer operating the remote control device. The viewer's behavior pattern is thus recorded and analyzed to identify the viewer. Upon identifying the viewer, various functions such as collecting the viewership statistics, implementing metered usage billing or ecommerce activities are enabled.

Methods, apparatus, systems and articles of manufacture to assign viewers to media meter data are disclosed. An apparatus includes memory, and a processor to execute instructions to: determine first probabilities for first panelists in a first household based on a first number of minutes of first media presented by a first media presentation device monitored by a first meter, determine second probabilities for second panelists in a plurality of second households based on a second number of minutes of second media presented by second media presentation devices monitored by a plurality of second meters, compare the first probabilities and the second probabilities to identify a candidate household from the plurality of second households to associated with the first household, and impute respective portions of the first number of minutes to corresponding ones of the first panelists when monitored behavior of the candidate household matches monitored behavior of the first household.

Methods and apparatus to predict demographics are disclosed. An example apparatus includes a prediction manager to access matched demographic impression data when a database proprietor subscriber does not match user data corresponding to a media delivery device, the matched demographic impression data including first demographic data stored by a database proprietor; a linear scaler to linearly scale the matched demographic impression data when the matched demographic impression data covers all demographic categories under consideration; a differential scaler to differentially scale the matched demographic impression data when the matched demographic impression data does not cover at least one of the demographic categories; and a modeler to: generate a model representative of the matched demographic impression data linearly scaled or the matched demographic impression data differentially scaled; and generate predicted demographic impression data by applying the model to second impression data to correct for the database proprietor not storing second demographic data.