Latency-efficient multi-stage tagging mechanism

Patent metadata

Field	Value
Publication number	US-9659259-B2
Application number	US-201414578424-A
Country	US
Kind code	B2
Filing date	Dec 20, 2014
Priority date	Dec 20, 2014
Publication date	May 23, 2017
Grant date	May 23, 2017

Abstract

Functionality is described herein for analyzing an input linguistic item, such as a query, in a series of stages. The linguistic item includes one or more candidate items. In a first stage, a brand classifier component determines whether the linguistic item specifies at least one brand, to provide a classifier output result. In a second stage, a tagging component generates a set of tags for at least some of the candidate items in the linguistic item, based, in part, on the classifier output result, to generate a tagging output result. An action-taking component then generates at least one result item based on the tagging output result. Functionality is also described herein for producing the brand classifier component and the tagging component using machine-learning training techniques. The training techniques may include provisions to address the later appearance of new brands that do not appear in a brand dictionary.

First claim

What is claimed is:

1. One or more computing devices that implement a latency-efficient system for assigning tags to a linguistic item, comprising:
a feature generation component configured to generate a plurality of feature values associated with a linguistic item that has been received from a user,
a machine-trained brand classifier component configured to determine, based on a first subset of the plurality of feature values, whether the linguistic item specifies at least one brand, to provide a classifier output result, the brand classifier component including a plurality of classifier subcomponents;
a machine-trained tagging component configured to determine, based on a second subset of the plurality of feature values, a tag associated with at least one candidate item within the linguistic item, to produce a tagging output result, the tagging component including a plurality of tagging subcomponents,
wherein the plurality of classifier subcomponents and the plurality of tagging subcomponents form a plurality of pairs of classifier subcomponents and tagging subcomponents that are associated with respective product domains,
wherein each classifier subcomponent and tagging subcomponent in a particular pair produces, respectively, a partial classifier result and a partial tagging result, the partial tagging result being based, in part, on the partial classifier result,
wherein the tagging output result encompasses a plurality of partial tagging results produced by the respective pairs of classifier subcomponents and tagging subcomponents;
a post-tagging interpretation component configured to analyze the plurality of partial tagging results, to determine which partial tagging result is most appropriate based on the user's intent, and to produce a final output result corresponding to the partial tagging result that is determined to be most appropriate; and
a user interface component to electronically communicate the final output result to the user.

2. The one or more computing devices of claim 1 , further comprising an action-taking component that is configured to generate a result item based, at least in part, on the tagging output result.

3. The one or more computing devices of claim 2 ,
wherein the input linguistic item is a query submitted by the user to a search engine,
wherein the tagging output result reflects an intent of the user in submitting the query, and
wherein the result item is an item, provided by the search engine to the user in response to the query, that is selected based on the intent of the user.

4. The one or more computing devices of claim 3 , wherein the final output result is a product page, provided by the search engine, that pertains to a product selected based on the user's intent.

5. The one or more computing devices of claim 1 ,
wherein the first subset of feature values and the second feature values include a shared set of common feature values, and
wherein the feature generation component is configured to generate the common feature values once, so as to reduce latency in generating the common feature values.

6. A method, implemented by one or more computing devices, for assigning tags to a linguistic item, and performing an action based on the tags, comprising:
receiving an input linguistic item from a user that includes one or more candidate items;
determining, using a machine-trained brand classifier component, whether the linguistic item specifies at least one brand, to provide a classifier output result, wherein the brand classifier component includes a plurality of classifier subcomponents;
determining, using a machine-trained tagging component, a tag associated with at least one candidate item within the linguistic item, to produce a tagging output result, said determining of the tag being based, in part, on the classifier output result, wherein the tagging component includes a plurality of tagging subcomponents;
generating a result item based, at least in part, on the tagging output result;
said receiving, determining whether the linguistic items specifies at least one brand, determining a tag, generating, and electronically communicating being performed by said one or more computing devices,
wherein the plurality of classifier subcomponents and the plurality of tagging subcomponents form a plurality of pairs of classifier subcomponents and tagging subcomponents that are associated with respective product domains,
wherein each classifier subcomponent and tagging subcomponent in a particular pair produces, respectively, a partial classifier result and a partial tagging result, the partial tagging result being based, in part, on the partial classifier result, and
wherein the tagging output result encompasses a plurality of partial tagging results produced by the respective pairs of classifier subcomponents and tagging subcomponents;
analyzing the plurality of partial tagging results in the tagging output result to determine which partial tagging result is most appropriate based on the user's intent;
producing a final output result corresponding to the partial tagging result that is determined to be most appropriate; and
electronically communicating the final output result to the user.

7. The method of claim 6 , wherein the input linguistic item is a query submitted by the user to a search engine, and
wherein the tagging output result reflects an intent of the user in submitting the query.

8. The method of claim 7 , wherein the result item is a product page, provided by the search engine, that pertains to a product that is selected based on the intent of the user.

9. The method of claim 7 , wherein the result item is a digital advertisement, provided by the search engine, that is selected based on the intent of the user.

10. The method of claim 7 , wherein the result item corresponds to a compilation of search result items, provided by the search engine, that is selected based on the intent of the user.

11. The method of claim 6 , further comprising generating a plurality of feature values associated with the linguistic item,
wherein the brand classifier component operates based on a first subset of the plurality of feature values,
wherein the tagging component operates based on a second subset of the plurality of feature values, and
wherein the first subset of feature values and the second subset of feature values include a shared set of common feature values.

12. The method of claim 11 , wherein the final output result is a product page, provided by the search engine, that pertains to a product selected based on the user's intent.

13. The method of claim 11 ,
wherein said generating of the plurality of feature values includes generating the common feature values once, so as to reduce latency in generating the common feature values.

14. The method of claim 6 , further comprising, in a machine-learning training process:
generating at least one brand classifier model based on a first group of feature values that are generated based on a set of training examples, the brand classifier component being implemented based on said at least one brand classifier model; and
generating at least one tagging model based on a second group of feature values that are generated based on the training examples, the tagging component being implemented based on said at least one tagging model,
the second group of feature values also including feature values that are generated based on a classifier output result that is produced by the brand classifier component, in response to processing the training examples.

15. The method of claim 14 , wherein said generating o

Assignees

• Microsoft Technology Licensing Llc
• Microsoft Corp

Inventors

• Li Min
• Zhang Ruofei
• Alam Muhammad Adnan

Classifications

• G06F17/30864

  Physics · mapped topic

• G06Q30/0277

  Online advertisement · CPC title

• G06N99/005 Primary

  Physics · mapped topic

• G06F16/906 Primary

  Clustering; Classification · CPC title

• G06N20/00

  Machine learning · CPC title