Google Inc.

Lars Kare Engebretsen

Ian James Leader

Jennifer J Huang

Wojciech W Skut

Puneet Chopra

Methods, systems, and apparatus, including computer programs encoded on a computer-readable storage medium, for evaluating a creative. Methods can include identifying one or more selection keywords that are associated with a creative. Matching criteria can be identified for comparing the selection keywords and the request keywords. The matching criteria can be defined by a range from an exact match to a broad match and include a matching threshold in the range. A bid modifier associated with the matching criteria can be established. The bid modifier can modify a bid associated with the creative based on an extent of a match determined between the selection keywords and the request keywords for a given content request. The creative responsive to received content requests can be evaluated based on the matching criteria and the bid modifier.

Broad match control

Systems and methods of this disclosure are directed to optimizing a content selector executing on content selection infrastructure. An entity engine retrieves a search query report with queries corresponding to selected content items of a content campaign and a performance metric for each query. The entity engine determines an entity for each of the queries. A cluster engine generates a first subset of the queries and a second subset of the queries based on the entity for each query. The cluster engine generates a first performance metric for the first subset and a second performance metric for the second subset. An interface displays the first performance metric and the second performance metric. A campaign generator receives a selection of a semantic criterion based on the first subset and updates the content campaign to include the semantic criterion.

Systems and methods of generating semantic traffic reports

This specification describes technologies relating to displaying online content. In general, one aspect of the subject matter described in this specification can be embodied in methods that include receiving a first query associated with a user request. The methods may further include determining a score for a word in the first query based at least in part on user interaction with a content item served for display in response to a past query that includes the word. The methods may further include selecting a keyword derived from the first query based at least in part on the score. The methods may further include identifying candidate content items using the selected keyword. Other embodiments of this aspect include corresponding systems, apparatus, and computer program products.

Selection and display of online content items

A query classification system classifies queries based on query features of search queries and a query classification model. The query classification model can be trained to classify a search query based on the query features independent of (e.g., without accessing) search results or other search log data for the search query. The query classification model can be trained based on query data and search log data corresponding to a plurality of labeled queries.

Query classification

One embodiment of the present invention provides a system that produces an object signature for an object, wherein the object comprises a set of features. During operation, the system first initializes a k-dimensional state vector {s, s, . . . , s}containing floating-point numbers. Next, for each occurrence of each feature in the object, the system (1) computes a random-number seed from the feature; (2) generates k pseudo-random floating-point numbers {X, X, . . . , X} using the random-number seed, wherein each X(i&#x3b5;[1, k]) is generated in accordance with an &#x3b1;-stable distribution, wherein 1&#x2266;&#x3b1;&#x3c;2; and (3) updates each floating-point number sin the k-dimensional state vector using an associated pseudo-random floating-point number X. The system then produces the object signature for the object by condensing the k-dimensional state vector. Note that using an &#x3b1;-stable distribution with 1&#x2266;&#x3b1;&#x3c;2 to generate the k pseudo-random floating-point numbers for each feature reduces the influence of multiple occurrences of a given feature on the object signature.

Growing community of inventors

Zurich, Switzerland

Lars Kare Engebretsen