Google Inc.

Yahoo! Inc.

Excalibur Ip, LLC

Sergei Vassilvitskii

Mohammad Mahdian

Shanmugasundaram Ravikumar

Andrew Stephen Tomkins

Umar Syed

Prabhakar Raghavan

Bo Pang

Jayavel Shanmugasundaram

Erik Nathan Vee

Alessandro Epasto

Hossein Esfandiari

Andres Munoz Medina

Vahab Seyed Mirrokni

Eric Theodore Bax

Andrei Z Broder

Marcus Felipe Fontoura

Jianchang Mao

Sihem Amer-Yahia

John Anthony Tomlin

Cong Yu

Jian Yang

Tasos Anastasakos

Seyed Vahab Mirrokni Banadaki

Silvio Lattanzi

R Preston McAfee

Raimondas Kiveris

Suhas Sadanandan

Jason Zien

Bahman Bahmani

Ashton Anderson

Oliver Kennedy

Srihari Venkatesan

Vibhor Rastogi

A computer-implemented method for k-anonymizing a dataset to provide privacy guarantees for all columns in the dataset can include obtaining, by a computing system including one or more computing devices, a dataset comprising data indicative of a plurality of entities and at least one data item respective to at least one of the plurality of entities. The computer-implemented method can include clustering, by the computing system, the plurality of entities into at least one entity cluster. The computer-implemented method can include determining, by the computing system, a majority condition for the at least one entity cluster, the majority condition indicating that the at least one data item is respective to at least a majority of the plurality of entities. The computer-implemented method can include assigning, by the computing system, the at least one data item to the plurality of entities in an anonymized dataset based at least in part on the majority condition.

Systems and methods for anonymizing large scale datasets

The subject matter of this specification can be embodied in, among other things, a system includes a map store, a map server, and a content server. The map server identifies a set of content items in which each content item is associated with a geographic location that is presented in a portion of a map, determines, for a given content item, an adjusted value of a given content item based at least in part on a base value for the content given item and a set of adjustment factors for one or more other content items in the set of content items, selects content items from the set of content items based on the adjusted values, and provides data that present the selected content items on the portion of the map presented at the device.

Valuing advertisements on a map

The present disclosure provides systems and methods for in-database generation of generalized linear models within a relational database. Generalized linear models form the basis of many machine learning algorithms and applications. In particular, in some implementations, the database commands that enable generation and use of the models include only pure SQL queries, thereby eliminating the need for user defined aggregates (UDAs), which are not offered by many cloud database service providers. For example, a set of client-side driver scripts can implement respective sets of pure SQL queries to import training data, generate and train the generalized linear model, and employ the model to generate inferences.

Large-scale in-database machine learning with pure SQL

A system and method for implementing bid adjustments in an online advertisement exchange are disclosed. A plurality of bids from a plurality of advertisers may be received for an online advertisement opportunity. A first advertiser may be associated with a first bid bias value. In some embodiments, if the first advertiser is associated with a first bid bias value, then the bid received from the first advertiser may be adjusted based on the first bid bias value. For example, an adjusted bid may be calculated based on the bid received from the first advertiser and the first bid bias value. The online advertisement exchange may conduct an online auction with the adjusted bid from the first advertiser. If the first advertiser wins the auction based on the adjusted bid, then the first advertiser may be charged the value of the bid received from the first advertiser instead of the adjusted bid value.

Systems and methods for implementing bid adjustments in an online advertisement exchange

Methods and systems for generating a dense graph are described. One of the methods includes receiving a graph and computing a threshold to apply to the graph. The method further includes determining whether the graph includes a first set of at least one node, determining whether a second set of at least one node from the first set meets the threshold, and removing the at least one node of the second set concurrently from the graph upon determining that the at least one node of the second set meets the threshold. The operation of removing is performed to generate an updated graph. The method includes determining whether a density of the updated graph is greater than a density of the graph and replacing the graph within the updated graph upon determining that the density of the updated graph is greater than the density of the density of the graph.

Systems and methods for generating a dense graph

Systems and methods for improving the time and cost to calculate connected components in a distributed graph are disclosed. One method includes reducing a quantity of map-reduce rounds used to determine a cluster assignment for a node in a large distributed graph by alternating between two hashing functions in the map stage of a map-reduce round and storing the cluster assignment for the node in a memory. Another method includes reducing a quantity of messages sent during map-reduce rounds by performing a predetermined quantity of rounds to generate, for each node, a set of potential cluster assignments, generating a data structure in memory to store a mapping between each node and its potential cluster assignment, and using the data structure during remaining map-reduce rounds, wherein the remaining map-reduce rounds do not send messages between nodes. The method can also include storing the cluster assignment for the node in a memory.

Computing connected components in large graphs

Methods and apparatus related to associating a quality measure with a given location. For example, an anticipated distance value for a given location may be identified that is indicative of anticipated time and/or distance to reach the given location. At least one actual distance may be identified that is indicative of actual time for the one or more members to reach the given location. In some implementations, the anticipated/actual distance values may include one or more distributions. A quality measure is then determined based on a comparison of the anticipated distance value and the identified actual distance value. The quality measure is associated with the given location. The quality measure may be further based on additional factors.

Determining the quality of locations based on travel time investment

Methods and apparatus are described herein for recommending candidates for consumption. A first set of one or more candidates for first time consumption by a user and associated utility scores may be identified. A second set of one or more candidates for reconsumption by the user may be determined. Utility scores of the second set of one or more candidates for reconsumption may be determined based on consumption history of the user and aggregate consumption history of a population of users of which the user is associated. A candidate to be recommended to the user for consumption may be selected from the first or second set. The candidate may be selected based at least in part on the utility scores of the first and second sets.

Recommending candidates for consumption

Methods, systems, and apparatus include computer programs encoded on a computer-readable storage medium, including a method for providing content. A request is received for content to be displayed on a resource to a user. The user is identified. Candidate content items are identified from an inventory of content items that are responsive to the request. The candidate content items are evaluated including determining bid-eligible content items including content items that satisfy an eligibility threshold based on an eligibility status associated with a respective candidate content item. The eligibility status is based at least in part on results associated with past opportunities to present a content item to identified user. An auction is conducted to identify a winner from among the bid-eligible content items. The winning content item is provided responsive to the request. An eligibility status is updated for losing and winning content items in the auction.

Limiting bid selection to eligible content items

Methods and apparatus related to determining a characteristic of a location based on visit data associated with the location. Visit data indicative of travel of users to a first location from one or more other locations and/or of travel of users from the first location to one or more other locations may be utilized to determine information about the location. In some implementations visit data may be utilized to determine a connectedness measure and associate the second location with the first location based on the connectedness measure. A characteristic of the first location may optionally be determined based on a characteristic of the second location and/or an attribute associated with the visit data.

Determining information about a location based on travel related to the location

An advertisement impression distribution system includes a data processing system operable to generate an allocation plan for serving advertisement impressions. The allocation plan allocates a first portion of advertisement impressions to satisfy guaranteed demand and a second portion of advertisement impressions to satisfy nonguaranteed demand. The data processing system includes an optimizer configured to establish a relationship between the first portion of advertisement impressions and the second portion of advertisement impressions. The relationship defines a range of proportions of allocation of the first portion and the second portion. The optimizer generates a solution maximizing guaranteed demand fairness, non-guaranteed demand revenue, and click value. The solution identifies a determined proportion of the first portion of advertisement impressions to serve and a determined proportion of the second portion of advertisement impressions to serve. The data processing system outputs the allocation plan to control serving of the advertisement impressions.

System for display advertising optimization using click or conversion performance

An item of inventory is described as a Boolean expression, which is converted into a multi-level, alternating AND/OR impression tree representation with leaf nodes representing conjuncts. Processing the conjuncts of the tree through a contract index results in retrieving a set of candidate contracts that match at least some but not necessarily all impression tree leaf node predicates. Next, an AND/OR contract tree representation is constructed with each contract tree leaf node having a label representing a projection onto a discrete set of ordered symbols. Contracts with projections that cover the entire range of discrete set of ordered symbols are deemed to satisfy the item of inventory. Implementation of the contract index includes retrieval techniques to support multi-valued predicates as well as confidence threshold functions using a multi-level tree representation of multi-valued predicates.

System and method for automatic matching of contracts using a fixed-length predicate representation

An improved method for information retrieval in web query and recommendation systems, where items that are likely unfamiliar to the users of the system, but potentially relevant, are recommended. In a recommendation system having ratings by a plurality of users for a plurality of items, items are assigned to one or more data regions based on item attributes or user activity. Source regions are identified for each of the data regions. For a given user, data regions with which both the user and the user's social network are unfamiliar are identified. Within a given data region, the relevance of items to the user within such regions is evaluated using ratings provided by other users who have entered ratings similar to the user in source regions for the data region. Items receiving the highest relevance score are recommended to the user.

Growing community of inventors

New York, NY, United States of America

Sergei Vassilvitskii