Abstract: Described are systems and methods that solve localization problems using Machine Learning models to compute country vectors for each linked content item and present content items in response to requests based on the country vectors. For example, a request from a user in Country A may be processed to determine candidate content items responsive to the request and to determine Country A as the country corresponding to the request. The candidate content items may then be processed to determine, for each candidate content item, a country vector corresponding to Country A as indicative of the relevance of the content item to Country A. Content items that are more likely than not to be relevant to the country of the request (e.g., Country A), as indicated by the respective country vector, may be considered as responsive and all other candidate content items discarded.

Abstract: Systems and methods for de-serializing one or more data elements of a serialized structured data record are presented. In response to a request to de-serialize one or more data elements of a serialized structured data record of a first type, the location of the serialized structured data record is determined within a data file containing a plurality of serialized structured data records. Locations of the data of the one or more data elements are determined and the data of the one or more data elements is retrieved. The data is de-serialized and stored in corresponding data elements in an instantiated structured data record of the first type. The one or more data elements that are de-serialized are fewer than all of the data elements of the serialized structured data record.

Abstract: The described implementations are operable to determine potential objects of interest to a user based on a blend of the user's long-term behavior and short-term interests. Long term user behavior may be determined for the user over a period of time and represented as continuous data. Short-term interest may be determined based on objects with which the user has recently interacted, and attributes of those objects may be represented together as continuous data corresponding to the short-term user interest. The continuous data of the short-term user interest and long-term user behavior may be blended to produce a user embedding. The user embedding may then be compared with objects to determine objects that are of potential interest to the user.

Abstract: Described are systems and methods that solve localization problems using Machine Learning models to compute country vectors for each linked content item and present content items in response to requests based on the country vectors. For example, a request from a user in Country A may be processed to determine candidate content items responsive to the request and to determine Country A as the country corresponding to the request. The candidate content items may then be processed to determine, for each candidate content item, a country vector corresponding to Country A as indicative of the relevance of the content item to Country A. Content items that are more likely than not to be relevant to the country of the request (e.g., Country A), as indicated by the respective country vector, may be considered as responsive and all other candidate content items discarded.

Abstract: The described implementations are operable to determine potential objects of interest to a user based on a blend of the user's long-term behavior and short-term interests. Long term user behavior may be determined for the user over a period of time and represented as continuous data. Short-term interest may be determined based on objects with which the user has recently interacted and attributes of those objects may be represented together as continuous data corresponding to the short-term user interest. The continuous data of the short-term interest and long-term user behavior may be blended to produce a user embedding. The user embedding may then be compared with objects to determine objects that are of potential interest to the user.

Abstract: Described are systems and methods that solve localization problems using Machine Learning models to compute country vectors for each linked content item and present content items in response to requests based on the country vectors. For example, a request from a user in Country A may be processed to determine candidate content items responsive to the request and to determine Country A as the country corresponding to the request. The candidate content items may then be processed to determine, for each candidate content item, a country vector corresponding to Country A as indicative of the relevance of the content item to Country A. Content items that are more likely than not to be relevant to the country of the request (e.g., Country A), as indicated by the respective country vector, may be considered as responsive and all other candidate content items discarded.

Abstract: The subject matter of this specification can be implemented in, among other things, a computer-implemented method for application lifecycle management. The method includes receiving crash reports from a first version of an application at multiple initial computing devices. The first version of the application is from a first group of versions. The method further includes determining that the crash reports include more than a threshold amount of crash reports. The method further includes determining that a second version of the application from a second group of versions corrects a cause of the crash reports. The method further includes providing instructions to the initial computing devices to retrieve updates to the application from the second group of versions rather than the first group of versions.

Abstract: The subject matter of this specification can be implemented in, among other things, a computer-implemented method for application lifecycle management including providing a first application version to initial computing devices. The method includes receiving first crash reports from the first application version. The method includes determining that the first crash reports include more than a first threshold of reports. The method includes preventing provision of the first application version to additional computing devices in response to determining that the first crash reports include more than the first threshold of reports. The method includes receiving second crash reports from the first application version. The method includes determining that the second crash reports include less than a second threshold of reports.

Abstract: The subject matter of this specification can be implemented in, among other things, a computer-implemented method for identifying causes of application crashes. The method includes receiving crash reports from multiple computing devices resulting from a problem that occurs in an application in operation at the computing devices. The crash reports include sets of properties that describe system states of the computing devices where the problem occurred. The method further includes identifying similarities between one or more properties in the sets of properties from the crash reports. The method further includes identifying differences between one or more properties in the sets of properties from the crash reports. The method further includes identifying one or more candidates for a cause of the problem in the application based on a clustering of the similarities and the differences. The method further includes generating a report that includes the one or more candidates.