Microsoft Technology Licensing, LLC

Rohan Ramanath

Jeffrey Gee

Deepak Kumar

Gungor Polatkan

Krishnaram Kenthapadi

Kun Liu

Qi Guo

Sahin Cem Geyik

Cagri Ozcaglar

Wen Pu

Liqin Xu

Sanjay Sachdev

Rupesh Gupta

Shipeng Yu

Konstantin Salomatin

Sara Smoot Gerrard

Onkar Anant Dalal

Lingjie Weng

Irina Belousova

Arjun K Kulothungun

Jiaqi Ge

Scott Khamphoune

Vasudeva Nagaraja

Himanshu Khurana

Vijay Ramamurthy

Deepak Dileep Kumar

In some embodiments, a computer system generates a recommendation for a user of an online service based on user actions that have been performed by the user within a threshold amount of time before the generation of the recommendation. For each user action, the computer system determines an intent classification that identifies an activity of the user and that corresponds to different types of user actions, as well as a preference classification that identifies a target of the activity, and then stores these intent and preference classifications as part of indications of the user actions for use in generating different types of recommendations using different types of recommendation models. Additionally, the computer system may use mini-batches of data from an incoming stream of logged data to train an incremental update to one or more recommendation models.

Recommendations using session relevance and incremental learning

The disclosed embodiments provide a system for processing a query for a ranking of candidates for an opportunity. During operation, the system obtains parameters associated with a query for a ranking of candidates for an opportunity, wherein the parameters include a candidate and the opportunity. Next, the system matches one or more of the parameters to a fixed number of quantile thresholds calculated from a distribution of scores for the candidates. The system then estimates, based on the fixed number of quantile thresholds, a quantile for a score of the candidate. Finally, the system outputs a position of the candidate within the ranking based on the estimated quantile.

Efficient percentile estimation for applicant rankings

Methods, systems, and computer programs are presented for embedding user categories into vectors that capture similarities between the user categories. One method includes an operation for building a graph for a category of attributes for users of a social network, the graph including a vertex for each category value. Connections, built between the graph vertices, have a connection value indicating the number of users to which the category values associated with the vertices have been assigned. Further, a first vector for each category value is obtained based on the graph, where a distance between two category values is a function of the connection value between the corresponding vertices. A user vector, based on the first vectors of the category values, is assigned to each user. A search is performed for a given user based on the user vectors, and results are presented to the given user.

Embedding user categories using graphs for enhancing searches based on similarities

Techniques for implementing a learning semantic representations of sparse entities using unsupervised embeddings are disclosed herein. In some embodiments, a computer system accesses corresponding profile data of users indicating at least one entity of a first facet type associated with the user, and generating a graph data structure comprising nodes and edges based on the accessed profile data, with each node corresponding to a different entity indicated by the accessed profile data, and each edge directly connecting a different pair of nodes and indicating a number of users whose profile data indicates both entities of the pair of nodes. The computer system generating a corresponding embedding vector for the entities based on the graph data structure using an unsupervised machine learning algorithm.

Unsupervised learning of entity representations using graphs

Techniques for improving the accuracy, relevancy, and efficiency of a computer system of an online service by providing a user interface to optimize a digital page of a user on the online service are disclosed herein. In some embodiments, a computer system receives a plurality of phrases for a type of job, selects a group of phrases from the plurality of phrases based on a corresponding relevancy measurement and a corresponding diversity measurement for each phrase in the selected group of phrases, and generates a recommendation for a page of a first user based on the selected group of phrases, with the recommendation comprising a suggested addition of the selected group of phrases to the page of the first user.

Phrase extraction for optimizing digital page

Techniques for improving the accuracy, relevancy, and efficiency of a computer system of an online service by providing a user interface to optimize a digital page of a user on the online service are disclosed herein. In some embodiments, a computer system receives a plurality of phrases, and then, for each one of the plurality of phrases, selects a corresponding section of a page of a first user to suggest for placement of the phrase from amongst a plurality of sections using a placement classifier, and generates a corresponding recommendation for the page of a first user based on the phrase and the determined corresponding section of the page of the first user, with the recommendation comprising a suggested addition of the phrase to the determined corresponding section of the page of the first user.

Phrase placement for optimizing digital page

Techniques for improving the accuracy, relevancy, and efficiency of a computer system of an online service by providing a user interface to optimize a digital page of a user on the online service are disclosed herein. In some embodiments, a computer system identifies job postings published on an online service as corresponding to a type of job based on feature data of each one of the job postings, extracts phrases from the identified job postings based on a corresponding relevancy measurement and a corresponding diversity measurement for each one of the phrases, determines a corresponding section of a page of a user to suggest for placement of the extracted phrase using a placement classifier for each one of the extracted phrases, and generates a corresponding recommendation for the page based on the extracted phrase and the determined section of the extracted phrase for each one of the phrases.

User interface for optimizing digital page

Techniques for processing search queries are described. Consistent with some embodiments, a computer system generates a profile vector representation for each of several user profiles based on the user profile data of the user profiles, and then stores the vector representations. A subsequent query is processed to generate a query vector representation for the query. A neural network is used to generate a similarity score for each pairing of the query vector representation and a profile vector representation. Finally, some number of user profiles having the highest similarity scores are provided as search results.

Techniques for querying user profiles using neural networks

Techniques for incremental workflow execution are provided. In one technique, a computing job in a workflow identifies an input path that indicates a first location from which the computing job is to read input data. The computing job identifies an output path that indicates a second location to which the computing job is to write output data. The computing job performs a comparison between the input path and the output path. Based on the comparison, the computing job determines whether to read the input data from the first location. If the input path does not correspond to the output path, then the computing job reads the input data from the first location, generates particular output data based on the input data, and writes the particular output data to the second location. The computing job ceases to execute if the input path corresponds to the output path.

Incremental workflow execution

Techniques related to cache management for multi-node databases are disclosed. In some embodiments, a system comprises one or more computing devices including a training component, data store, cache, filtering component, and listening component. The training component produces a plurality of models based on user interaction data. The plurality of models are stored in the data store, which responds to requests from the cache when the cache experiences cache misses. The cache stores a first subset of the plurality of models. The filtering component selects a second subset of the plurality of models based on one or more criteria. Furthermore, the filtering component sends the second subset of the plurality of models to a messaging service. The listening component retrieves the second subset of the plurality of models from the messaging service. Furthermore, the listening component causes the second subset of the plurality of models to be stored in the cache.

Growing community of inventors

Saratoga, CA, United States of America

Rohan Ramanath