Abstract: A system uses semantic analysis of text associated with content items to recommend content for display to a user. A subset of representative words from a content description are determined and a content embedding that models the content is generated using a combination of word embeddings associated with each of the representative words. User embeddings are generated using a combination of content embeddings for content that a user has had particular interactions with in a set period of time. Separate user embeddings may be generated to represent user interactions with different categories of content (e.g., travel, photography, apparel, comedy, etc.). The system uses the content embeddings and user embeddings as input to predictive functions which determine a candidate content item that a user is likely to interact with if the candidate content is displayed to the user.

Abstract: Network site users can be selected to receive a communication based on a network site event, such as incomplete registration. A hybrid user interaction machine learning scheme can select a portion of the selected users based on user interaction estimates and network sampling data. The electronic document sent to the users can have portions that undergo two-pass ranking for ordering of content items to be included in the electronic document, such as an email.

Abstract: Highly user-specific data is used to calculate user intent to make a purchase and the value of such a purchase. User activity and information is aggregated, per user, for a set window of time and real-time data on recent site behavior is obtained. Aggregated and/or real-time data is considered by a predictive intent model (calculating the probability that the user will make a purchase) and a predictive value model (calculating the expected revenue such a purchase may generate). Weights, specific to each model, are assigned to predictor features tracked in the aggregated and/or real-time user data. The most highly-weighted features of the intent model relate to users' viewing history, and the most highly-weighted features of the value model relate to price and market. By these means, a user conversion value can be obtained, guiding the application of user acquisition strategies for different home sharing markets.

Abstract: Highly user-specific data is used to calculate user intent to make a purchase and the value of such a purchase. User activity and information is aggregated, per user, for a set window of time and real-time data on recent site behavior is obtained. Aggregated and/or real-time data is considered by a predictive intent model (calculating the probability that the user will make a purchase) and a predictive value model (calculating the expected revenue such a purchase may generate). Weights, specific to each model, are assigned to predictor features tracked in the aggregated and/or real-time user data. The most highly-weighted features of the intent model relate to users' viewing history, and the most highly-weighted features of the value model relate to price and market. By these means, a user conversion value can be obtained, guiding the application of user acquisition strategies for different home sharing markets.

Abstract: Network site users can be selected to receive a communication based on a network site event, such as incomplete registration. A hybrid user interaction machine learning scheme can select a portion of the selected users based on user interaction estimates and network sampling data. The electronic document sent to the users can have portions that undergo two-pass ranking for ordering of content items to be included in the electronic document, such as an email.

Abstract: A cost-focused determination of whether to deliver an electronic advertisement or notice to a particular user can be made through a cumulative consideration of the predicted return on investment over each of a plurality of electronic channels. A plurality of channel-specific budget values are calculated for the user, one for each channel, each setting an upper spending limit for advertisement to the user over that channel based on the user's information and their activity on the channel. A global budget is calculated for the user using a weighted aggregation of the channel-specific values, information about the user and their activity with the advertiser, and consideration of “overlap” effects of advertising to the same user on several channels. When managing whether to deliver an advertisement over a channel, if the channel-specific value is lower than the global budget, the advertisement is delivered, and the global budget is decreased by a complementary amount.

Abstract: Network site users can be selected to receive a communication based on a network site event, such as incomplete registration. A hybrid user interaction machine learning scheme can select a portion of the selected users based on user interaction estimates and network sampling data. The electronic document sent to the users can have portions that undergo two-pass ranking for ordering of content items to be included in the electronic document, such as an email.

Abstract: One of more unique products can be selected for advertisement over a digital marketing channel. The selection is based on the calculation of a base impression budget, calculated per product per day, which calculation considers information related to a product, including supply and demand at the market level, characteristics of the property, and popularity of the listing. A real-time, or current, impression budget is calculated to determine whether a particular product should be recommended to a user. Every time the product is advertised to a user, a user intent value is calculated, indicating the user's likelihood of purchasing the product within a given period of time, along with the user intent of every user to which the product has been advertised. The user intent calculation may consider information specific to the user, such as the user's activity history and profile. These user intent values are subtracted from the base impression budget to obtain a real-time impression budget.

Abstract: A cost-focused determination of whether to deliver an electronic advertisement or notice to a particular user can be made through a cumulative consideration of the predicted return on investment over each of a plurality of electronic channels. A plurality of channel-specific budget values are calculated for the user, one for each channel, each setting an upper spending limit for advertisement to the user over that channel based on the user's information and their activity on the channel. A global budget is calculated for the user using a weighted aggregation of the channel-specific values, information about the user and their activity with the advertiser, and consideration of “overlap” effects of advertising to the same user on several channels. When managing whether to deliver an advertisement over a channel, if the channel-specific value is lower than the global budget, the advertisement is delivered, and the global budget is decreased by a complementary amount.

Abstract: Systems and methods are provided for providing personalized content based on interest levels determined using machine learning. An online marketplace system determines an interest profile for a client user using profile data of the client user as input into an interest extraction model. The online marketplace system selects a content item for the client user based on the interest profile and causes presentation of the content item to the client user via a channel that is separate from the online marketplace system. In response to receiving a request to access the online marketplace system originated from the content item, the online marketplace system selects product recommendations based on interest profile and the content item. The online marketplace system generates a personalized landing webpage including the product recommendations and causes presentation of the personalized landing webpage to the client user.

Abstract: Network site users can be selected to receive a communication based on a network site event, such as incomplete registration. A hybrid user interaction machine learning scheme can select a portion of the selected users based on user interaction estimates and network sampling data. The electronic document sent to the users can have portions that undergo two-pass ranking for ordering of content items to be included in the electronic document, such as an email.

Abstract: An online system generates a content item for a user based on products likely to be of interest to the user. The online system receives information about products associated one or more third party systems accessed by users of the online system. When the online system identifies an opportunity to present to a user, the online system identifies candidate products for inclusion in the content item based on products previously accessed by the users. For example, the online system identifies candidate products based on products accessed by the user and by one or more other users. Based on likelihoods of the user accessing content items including different candidate products, the online system selects a candidate product and includes the content item having information about the selected candidate product in one or more selection processes that select content for presentation to the user.

Abstract: An online system generates a content item for a user based on products likely to be of interest to the user. The online system receives information about products associated one or more third party systems accessed by users of the online system. When the online system identifies an opportunity to present to a user, the online system identifies candidate products for inclusion in the content item based on products previously accessed by the users. For example, the online system identifies candidate products based on products accessed by the user and by one or more other users. The online system may include differing levels of information about a selected candidate product in the content item. In various embodiments, the online system determines a level of information about the selected candidate product based on products previously accessed by the user.

Abstract: An online system generates a content item for a user based on products likely to be of interest to the user. The online system receives information about content provided by one or more third party systems accessed by the user and determines products associated with accessed content. When the online system identifies an opportunity to present to a user, the online system identifies products for inclusion in the content item and identifies candidate products for inclusion in the content item based on products previously accessed by the user. The online system selects a product of the candidate products based on probabilities of the user accessing content items including different candidate products. The online system includes the content item having information about the selected product in one or more selection processes that select content for presentation to the user.