Abstract: A method comprises receiving a session identifier from a streaming system that identifies a user session with the streaming system. The method further includes receiving a first message from a streaming system that is based on a token that is generated based on a combination of the session identifier and a timestamp at which an insertable content item was presented to the user in a content stream by the streaming system. The first message is decrypted using a plurality of timestamps that are within a range of a current time. An identifier is determined for the insertable content item based on the decrypted message. A second message is transmitted to an enabling system, the message including instructions for execution by the enabling system to execute one or more operations with the identified insertable content item.

Abstract: An online system trains machine learning models that, when applied to gathered sensor data, determines personal characteristics (e.g., age, gender, height) of an individual in a non-intrusive manner. Specifically, the online system trains a first machine learning model that analyzes sensor data gathered from a client device associated with the individual. The first machine learning model determines whether a trigger event, such as whether the individual is walking, is currently occurring. A second machine learning model trained by the online system analyzes sensor data corresponding to the trigger event to identify the personal characteristics of the walking individual.

Abstract: An online system tailors audio segments for users accessing a live audio stream such that the audio segments can be presented to users during a break of the live audio stream. The audio segment can include interactive content as well as one or both of an audio clip and a standard message, each of which serves as a type of introduction to the interactive content. For each user, the online system analyzes characteristics of the user to determine whether to include the standard message in addition to the audio clip, or to withhold the standard message from the audio segment. Therefore, different users accessing the live audio stream can be appropriately introduced to the interactive content which can improve the likelihood that the users interact with the interactive content.

Abstract: A content replacement system sends a content item with one or more content replacement parts to be consumed by a user of a client device. The content item is further associated with an interaction pattern that corresponds to a particular user interaction and a defined action that is to be executed by the client device when the particular user interaction is performed on the client device. While the content replacement part is played on the client device, one or more sensors of the client device acquire data regarding user interactions performed by the user on the client device. The acquired data is processed to determine whether the user has performed a user interaction that matches the interaction pattern. Upon detecting a match, the client device sends a notification to the content replacement system indicating that the match has occurred and further executes the defined action.

Abstract: A content replacement system sends a content item with one or more content replacement parts to be consumed by a user of a client device. The content item is further associated with an interaction pattern that corresponds to a particular user interaction and a defined action that is to be executed by the client device when the particular user interaction is performed on the client device. While the content replacement part is played on the client device, one or more sensors of the client device acquire data regarding user interactions performed by the user on the client device. The acquired data is processed to determine whether the user has performed a user interaction that matches the interaction pattern. Upon detecting a match, the client device sends a notification to the content replacement system indicating that the match has occurred and further executes the defined action.

Abstract: An online system tailors audio segments for users accessing a live audio stream such that the audio segments can be presented to users during a break of the live audio stream. The audio segment can include interactive content as well as one or both of an audio clip and a standard message, each of which serves as a type of introduction to the interactive content. For each user, the online system analyzes characteristics of the user to determine whether to include the standard message in addition to the audio clip, or to withhold the standard message from the audio segment. Therefore, different users accessing the live audio stream can be appropriately introduced to the interactive content which can improve the likelihood that the users interact with the interactive content.

Abstract: A content replacement system identifies regular content replacement parts and alternative content replacement parts, each alternative content replacement part being a content item with a duration within one of a plurality of ranges. The system detects a start of a content replacement break in a listener content stream, the listener content stream shifted by a listener delay value from a live content stream. The system inserts the regular content replacement parts into the stream. The system detects an end of the content replacement break for the listener content stream, and selects one of the alternative content replacement parts to insert into the listener content stream, the selected alternative content replacement part having a duration that when inserted into the listener content stream minimizes a difference between the listener delay value and a target delay value.

Abstract: A content replacement system detects a start of a content replacement break in a content stream provided to a client device. The system selects one or more of a plurality of content replacement parts. The system computes a retargeting probability for each of the selected content replacement parts, the retargeting probability based on collected information stored in a profile database, the collected information including a number of times that requests for content replacement parts have been associated with a network address and device identifier of the client device. The system modifies a content score of each of the selected content replacement parts by the corresponding retargeting probability. The system ranks the selected content replacement parts based on the modified content scores, inserts one or more of the highest ranked selected content replacement parts into the content stream, and transmits the modified content stream to the client device.

Abstract: An online system predicts attention scores using a predictive model, each attention score representing a user's perspicaciousness as the user is accessing a live stream. The predictive model describes a relationship between a user's attention score and when the user joined the live stream. Therefore, when the online system receives a request to generate an attention score for a user that joined the live stream at a particular time, the online system applies the predictive model to determine a predicted attention score for the user. The online system can provide the predicted attention score to a content provider server such that the content provider server can select content items to be presented to each user during a break in the live stream. If users do not respond to presented content items as expected, the online system can update the predictive model to account for the unexpected responses.