Yahoo! Inc.

Excalibur Ip, LLC

Pavel Berkhin

Andrew Stephen Tomkins

Rajesh Girish Parekh

Usama M Fayyad

Scott J Gaffney

John Canny

Shanmugasundaram Ravikumar

George H John

Chad Brower

Boris Klots

Prabhakar Raghavan

John Anthony Tomlin

Richard Tao-Hwa Chow

Elena Eneva

Nicolas Eddy Mayoraz

Shi Zhong

Zhichen Xu

Qi Lu

Eckart Walther

Jianchang Mao

Daniel Eric Rose

Jan O Pedersen

Jignashu Parikh

David Ku

Chung-Man Tam

Ye Chen

Dmitry Pavlov

Jignesh Parmar

Bassel Ojjeh

David Wellspring Sculley, Ii

Zoltan Istvan Gyongyi

Shi Zhonog

A method of targeting receives several granular events and preprocesses the received granular events thereby generating preprocessed data to facilitate construction of a model based on the granular events. The method generates a predictive model by using the preprocessed data. The predictive model is for determining a likelihood of a user action. The method trains the predictive model. A system for targeting includes granular events, a preprocessor for receiving the granular events, a model generator, and a model. The preprocessor has one or more modules for at least one of pruning, aggregation, clustering, and/or filtering. The model generator is for constructing a model based on the granular events, and the model is for determining a likelihood of a user action. The system of some embodiments further includes several users, a selector for selecting a particular set of users from among the several users, a trained model, and a scoring module.

Granular data for behavioral targeting

Techniques are described for generating an authority value of a first one of a plurality of documents. A first component of the authority value is generated with reference to outbound links associated with the first document. The outbound links enable access to a first subset of the plurality of documents. A second component of the authority value is generated with reference to a second subset of the plurality of documents. Each of the second subset of documents represents a potential starting point for a user session. A third component of the authority value is generated representing a likelihood that a user session initiated by any of a population of users will end with the first document. The first, second, and third components of the authority value are combined to generate the authority value. At least one of the first, second, and third components of the authority value is computed with reference to user data relating to at least some of the outbound links and the second subset of documents.

User-sensitive PageRank

Embodiments of the invention relate to methods of presenting personalized search results pages to users, and to search engine systems and servers configured to implement such methods. For example, a method of presenting such a page to a user of a search engine includes steps of computing an engagement index of the user based on the distribution in time of that user's interactions with the search engine then presenting, in response to a query by the user, a personalized search results page to the user.

Personalizing sponsored search advertising layout using user behavior history

Methods and apparatus are described for optimally allocating an online advertising budget for a search engine marketing (SEM) campaign among a fixed set of keywords.

Bid optimization in search engine marketing

An improved system and method for web destination profiling for online population-targeted advertising is provided. A web destination profiler may be provided for generating web destination profiles. Traffic may be analyzed at a particular web destination in order to understand the population visiting the web destination. The analysis of user traffic, including differentiated clickstream data, may be applied for determining known characteristics of a web destination profile. Moreover, unknown characteristics of a web destination profile may be determined using a variety of techniques including inferring characteristics by modeling traffic flow through other web destinations, estimating characteristics from other web destination profiles by predicting traffic flow through other web destinations, propagating characteristics to a web destination profile by smoothing a joint distribution of characteristics of other web destination profiles, and so forth. Web destination profiles may be used by applications such as an online application for population-targeted advertising.

System and method for web destination profiling

Methods and apparatus are described for evaluating a binary classification system operable to classify each of a plurality of events as a first event type or a second event type. At least some of the events of the first event type are independently verifiable with reference to verification data. The binary classification system is susceptible to a first error type in which events of the first event type are classified as the second event type, and a second error type in which events of the second event type are classified as the first event type. Operation of a first configuration of the binary classification system is evaluated with reference to an objective function. The objective function is derived by expressing a number of errors of the second error type in terms of a number of errors of the first error type with reference to the verification data, and by assuming relative proportions of the first and second event types within the plurality of events.

Evaluating performance of click fraud detection systems

Evaluating performance of binary classification systems

System and method for population-targeted advertising

Method and system for generating a linear machine learning model for predicting online user input actions

A method and a system are provided for large-scale behavioral targeting for advertising over a network, such as the Internet. In one example, the system receives training data that is processed raw data of user behavior. The system generates selected features by performing feature selection on the training data. The system generates feature vectors from the selected features. The system initializes weights of a behavioral targeting model by scanning the feature vectors once. The system then updates the weights of the behavioral targeting model by scanning iteratively the feature vectors using a multiplicative recurrence.

Large-scale behavioral targeting for advertising over a network

Embodiments of the present invention provide systems and methods for ranking a result set. The method according to one embodiment comprises selecting an item from the result set, selecting a user profile from one or more user profiles and selecting one or more items of personalized information from the selected user profile. A rank is calculated for the item on the basis of the selected one or more items of personalized information and the selected item in the result set is ranked in accordance with the calculated rank.

Search engine with augmented relevance ranking by community participation

A method of targeting receives several granular events and preprocesses the received granular events thereby generating preprocessed data to facilitate construction of a model based on the granular events. The method generates a predictive model by using the pre-processed data. The predictive model is for determining a likelihood of a user action. The method trains the predictive mode. A system for targeting includes granular events, a preprocessor for receiving the granular events, a model generator, and a model. The preprocessor has one or more modules for at least one of pruning, aggregation, clustering, and/or filtering. The model generator is for constructing a model based on the granular events, and the model is for determining a likelihood of a user action. The system of some embodiments further includes several users, a selector for selecting a particular set of users from among the several users, a trained model, and a scoring module.

Granular data for behavioral targeting using predictive models

A variety of techniques are described by which keyword sets and target audience profiles may be generalized in a systematic and effective way with reference to relationships between keywords, profiles, and the data of an underlying user population.

Keyword set and target audience profile generalization techniques

A first component of an authority value is generated with reference to outbound links associated with a document and corresponding to a first subset of a plurality of documents. A second component of the authority value is generated with reference to a second subset of the plurality of documents that represent potential starting points for user sessions. A third component of the authority value is generated representing a likelihood that a user session initiated by any of a population of users will end with the document. At least one of the first, second, and third components of the authority value is computed with reference to user data relating to at least some of the outbound links and the second subset of documents.

User-sensitive pagerank

Methods and apparatus are described for storing a plurality of objects in a plurality of storage options. An importance index is generated for each of the plurality of objects with reference to importance data associated with each object. At least a portion of the importance data represents relevance of the associated object relative to a population of users interacting with the plurality of objects. Each of the objects is stored in a selected one of the storage options with reference to the corresponding importance index and a hierarchy of the storage options. The hierarchy of storage options represents at least partial ordering of the storage options with reference to economic costs and efficiency of retrieval.

Multi-tiered storage

Growing community of inventors

Sunnyvale, CA, United States of America

Pavel Berkhin