Abstract: Removing bias when matching job candidates to open positions by obtaining candidates’ data including information about the job candidates and a likelihood rate that the candidate matches the open position, identifying protected characteristics from the candidates’ data, generating a training data set that does not bias within groups of candidates having different protected characteristics, where the training data set includes a portion of the job candidates, training a model based on the training data set, applying the trained model on a test set, where the test set is different from the training data set, and determining a fairness measurement value of the trained model using the results of the model on the test set and protected characteristics of candidates of the test set.

Abstract: Identifying personalized content recommendations for users in an electronic environment is disclosed. User data comprising information relating to web-based content consumption of multiple users is collected. Multiple user cluster types associated with one or more interest categories are established. A feature vector is generated for each user for each of the multiple user cluster types. Based on the generated feature vectors, the user are grouped into multiple clusters. A grade is generated for each of a plurality of candidate recommendations for each of the clusters. Based on the generated grades, one or more personalized content recommendations for each of the clusters are identified.

Abstract: A server computer system identifies a user and a destination document. The server computer system identifies recommendations that correspond to the user and the destination document. The server computer system determines grades for the recommendations based on the user and/or the destination document. The server computer system determines serving probabilities for the recommendations based on the proportions of the grades and provides the serving probabilities to serve the recommendations.

Abstract: Identifying impressions relating to a target publisher that are related to or derived from a user interaction with a content recommendation source. User activity data for multiple users is collected during an activity window. Based on the collected user activity data, an initial interaction by a user with a source is identified and used to establish a source-related user session beginning at a time of the initial interaction and ending after a session period. A set of impressions (e.g., page views) by the user relating to the target publisher occurring during the user session is identified. The identified set of impressions is associated with the user session. A source-related user activity measurement is calculated based on the identified user sessions and associated impressions occurring during the activity window.

Abstract: Identifying impressions relating to a target publisher that are related to or derived from a user interaction with a content recommendation source. User activity data for multiple users is collected during an activity window. Based on the collected user activity data, an initial interaction by a user with a source is identified and used to establish a source-related user session beginning at a time of the initial interaction and ending after a session period. A set of impressions (e.g., page views) by the user relating to the target publisher occurring during the user session is identified. The identified set of impressions is associated with the user session. A source-related user activity measurement is calculated based on the identified user sessions and associated impressions occurring during the activity window.

Abstract: Identifying collaboratively filtered content recommendations based on user activity data collected from users in an electronic environment is disclosed. The user activity data includes information relating to user-engagement indications of the users. Multiple connection pairs are identified based on the user activity data, with each connection pair including a potential target document and a candidate recommendation document. A connection strength is determined for each of the identified connection pairs. An overall grade is generated for a set of candidate connection pairs of the plurality of connection pairs, wherein the overall grade is based on the respective connection strength associated with each of the set of candidate connection pairs. One or more filtered recommendations for provisioning in connection with the target document based on a comparison of the overall grade associated with each of the set of candidate connection pairs.