Microsoft Technology Licensing, LLC

Terrapin Technologies, Inc.

Stuart MacDonald Ambler

Krishnaram Kenthapadi

Parul Jain

Lawrence M Kauvar

Liang Zhang

Edoardo M Airoldi

Deepak Agarwal

Bo Zhao

Ryan Wade Sandler

In an example, each of a plurality of members of social networking service is mapped to a weighted skill vector, each weighted skill vector including a list of skills for the member with an associated weight indicating strength of the skill. Members of the social networking service that belong to an industry are aggregated to obtain a weighted matrix of members and skills along with compensation vectors indicating compensation for each of the members in the matrix. The weighted matrix of users and skills and corresponding compensation vectors is used to train a machine learning skill monetary value prediction model to output a predicted monetary value for one or more skills contained in a candidate vector fed to the machine learning skill monetary value prediction model.

Machine learning technique for recommendation of courses in a social networking service based on confidential data

In an example, one or more member profiles and corresponding elapsed times indicating, for each of the one or more member profiles, how long the corresponding member of a social networking service took to respond to a request for confidential data with a submission of confidential data are obtained. Then a first set of one or more features are extracted from the one or more member profiles. The first set of one or more features and corresponding elapsed times are fed into a machine learning algorithm to train a confidential data response time prediction model to output a predicted time to respond to a request for confidential data for a candidate member profile. A second set of one or more features are obtained from a candidate member profile and fed to the confidential data response time prediction model, outputting the predicted time to respond to a request for confidential data.

Estimating the conditional response time to a request for confidential data

In an example, one or more job postings, as well as corresponding confidential data values, are obtained from a social networking service. A first set of one or more features are extracted from the one or more job postings. The first set of one or more features and corresponding confidential data values are fed into a machine learning algorithm to train a confidential data value prediction model to output a predicted confidential data value for a candidate job posting. Then, the candidate job posting is obtained and a second set of one or more features are extracted from the candidate job posting. The extracted second set of one or more features is fed to the confidential data value prediction model, outputting the predicted confidential data value.

Machine learning model to estimate confidential data values based on job posting

In an example, a method includes requesting cohort data for the cohort, computing a plurality of cohort data first accuracy metrics, adding a threshold percentage of noise data points to the cohort data, computing a plurality of cohort data second accuracy metrics, repeating the adding and computing the second accuracy metrics until a mathematical difference between one or more of the first accuracy metrics and the second accuracy metrics exceed a threshold value, and suppressing displaying the cohort data in response to the mathematical difference exceeding the threshold value.

Mapped dataset error tolerance

In an example, an anonymized set of confidential data data values of a first confidential data type is obtained. Then an anonymized set of confidential data data values of a second confidential data type is also obtained. A multiplier following a log-normal distribution is determined for the anonymized set of confidential data data values of the first confidential data type. Then smoothing is performed independently for the anonymized set of confidential data data values of the first confidential data type and the multiplier. Percentiles for the anonymized set of confidential data data values of the second confidential data type are then determined using the smoothed anonymized set of confidential data data values of the first confidential data type and the smoothed multiplier.

Ensuring consistency between confidential data value types

In an example, a method includes determining an external quality metric for cohort data by comparing the cohort data with externally available data for persons who are members of the cohort; determining a confidence metric for the cohort data according to a variability between the cohort data and ground truth data; determining a member accuracy metric by requesting evaluation of the cohort data by members of the online social networking service; normalizing each of the metrics to a predetermined numerical range; retrieving a weight for each of the external quality metric, the confidence metric, and the member accuracy metric; filtering the cohort data according to a convex combination of the external quality metric, the confidence metric, and the member accuracy metric with their respective weights; and suppressing display of the cohort data in response to the filtered information for the cohort indicating that the convex combination traverses a threshold value.

Filtering mapped datasets

In an example, one or more member profiles and corresponding Boolean attributes indicating, for each of the one or more member profiles, whether the corresponding member of a social networking service interacted with a request for confidential data, are obtained. A first set of one or more features are extracted from the one or more member profiles. The first set of one or more features and corresponding Boolean attributes are fed into a machine learning algorithm to train a confidential data response propensity prediction model to output a predicted propensity to interact with a request for confidential data for a candidate member profile. A second set of one or more features are extracted from the candidate member profile. The extracted second set of one or more features are fed to the confidential data response propensity prediction model, outputting the predicted propensity to interact with a request for confidential data.

Machine learning model for estimating confidential information response

In an example, a set of cohort types and an anonymized set of confidential data data values for a plurality of cohorts having cohort types in the set of cohort types are obtained. Then it is determined, from a set of candidate data transformations, a best fitting data transformation for the anonmyized set of confidential data data values. The anonymized set of confidential data data values is transformed using the best fitting data transformation. Optimal smoothing parameters are computed for each cohort type. Then, for each cohort in the set of cohort types having a small sample size, a best parent for the cohort is determined and a posterior distribution for the cohort is determined based on the best parent for the cohort and the optimal smoothing parameters for a cohort type for the cohort.

Computing smoothed posterior distribution of confidential data

In an example embodiment, each of a plurality of members of a social networking service is mapped to a weighted skill vector, each weighted skill vector including a list of skills for the member with an associated weight indicating a strength of the skill. Members of the social networking service who belong to an industry are aggregated to obtain a weighted matrix of members and skills along with compensation vectors indicating compensation for each of the members in the matrix. The weighted matrix of members and skills and corresponding compensation vectors are used to train a machine learning skill monetary value prediction model to output a predicted monetary value for a skill contained in a candidate vector fed to the machine learning skill monetary value prediction model. A recommendation is provided to a member of one or more skills to add based on output of the model.

Machine learning technique for recommendation of skills in a social networking service based on confidential data

In an example, a method includes generating a targeted communication to respective computing devices of one or more members of an online social networking service, electronically collecting the responses to the targeted communication, mapping the responses to a cohort by updating a record in a database of members, the record identifying the cohort for the respective member, filtering information corresponding to a cohort to provide filtered cohort, and suppressing displaying of information corresponding to the cohort in response to the filtered information for the cohort indicating one or more of the biases being above a bias threshold value.

In an example, a query on a plurality of previously submitted confidential data values for a first cohort having one or more attributes is obtained, and a level in a hierarchy corresponding to an attribute type for the attribute is determined for each attribute. One or more additional cohorts corresponding to different combinations of generalizations of the one or more attributes up one or more levels in each hierarchy corresponding to an attribute type for each attribute are formed. For each cohort, a confidence score and a granularity score are calculated, and then a cohort score is calculated based on a weighted combination of the confidence score and the granularity score. A statistical function is performed on previously submitted confidential data values for a cohort having the highest cohort score, and a response to the query including a result from the statistical function is formed.

Dynamic hierarchical generalization of confidential data in a computer system

In an example embodiment, posterior distribution based percentiles for confidential data submitted to a computer system are computed. Then empirical percentiles are computed for the confidential data. A convex combination factor is computed based on a ratio between a number of valid entries in a cohort of the confidential data values and a combination of the number of valid entries in the cohort and the number of valid entries in a parent cohort of the cohort. Then, for each percentile of interest, a convex combination of the empirical percentile and the posterior distribution based percentile is calculated, using the convex combination factor to weight the empirical percentile.

Combining the smoothed posterior distribution with empirical distribution for cohorts with small sample size

In an example embodiment, an anonymized set of confidential data values is obtained for a plurality of combinations of cohorts having a first attribute type and a second attribute type. A matrix of the confidential data values having the first attribute type as a first axis and the second attribute type as the second axis is constructed. A set of candidate low rank approximations of the matrix is calculated using an objective function evaluated using a set of candidate data transformation functions, the objective function having one or more parameters and an error function. One or more parameters that minimize the error function of the objective function are minimized to select one of the candidate low rank approximations of the matrix. Then one or more cells that are missing data, of the selected one of the candidate low rank approximations of the matrix, are inferred.

Computerized matrix factorization and completion to infer median/mean confidential values

In an example, a submission of a confidential data value of a first confidential data type is received from a first user with one or more attributes. A plurality of previously submitted confidential data values of a first confidential data type for a cohort matching the one or more attributes of the first user are retrieved. Then, one or more intermediate cohorts are derived by generalizing each of the one or more attributes of the cohort up at least one level in a different taxonomy corresponding to each of the one or more attributes. One or more of the intermediate cohorts are selected, and a parameterized distribution is fitted to the previously submitted confidential data values that are contained within the selected one or more of the intermediate cohorts, outputting one or more estimated parameters for each of the selected one or more of the intermediate cohorts. A lower limit for the first confidential data type is then set based on the one or more estimated parameters.

Outlier detection based on distribution fitness

In an example, a granularity of title similarity is determined, wherein the granularity of title similarity is a level at which social networking data should be filtered to identify titles similar to a target title. Then a weighted graph of titles is constructed at the granularity of title similarity, wherein each node in the weighted graph is a title and a directed edge exists in the weighted graph between a first node and a second node if the social networking data, at the granularity of title similarity, indicates that a transition occurred wherein a member who held a position with a title corresponding to the first node transitioned to a position with a title corresponding to the second node, wherein each directed edge contains a weight indicating a strength of relationship between nodes. The weighted graph of locations is traversed from a node corresponding to the target title in order to identify titles similar to the target title.

Growing community of inventors

Longmont, CO, United States of America

Stuart MacDonald Ambler