Machine-Learned Disambiguation of Searcher Action Data

What Might a Search Engine Learn From Searcher Action Data?

This patent is about improving the working of a computer.

Searcher action data may represent financial transactions. A credit card statement may include financial transaction data about a particular credit card.

This searcher action data may include:

• A date
• Description
• Amount corresponding to a specific financial transaction

Financial transaction data, [“4/12/2013,” “ITALIAN DELI NEW YORK NY,” “12.08”], may represent a credit card charge for $12.08 for a transaction on Apr. 12, 2013, at “Italian Deli” located in New York, N.Y.

Related Content:

• Technical SEO Agency
• Ecommerce SEO Agency
• Shopify SEO Services
• Franchise SEO Agency
• Enterprise SEO Services

This patent involves searcher action data about many entity types. The process requires using a machine learning-based annotator to:

• Recognize entities
• Annotates the searcher action data
• Receive a query from the searcher
• Query the searcher action data
• Generate an answer to the query from the searcher action data
• Answer the query

The system may receive data taken by a particular searcher about entities of a plurality of entity types. The data for a majority of the entities get determined as ambiguous because the entities cannot get determined. The system disambiguates the information that represents actions taken by a particular searcher about entities of a plurality of entity types.

What Gets Disambiguated From the Searcher Action Data?

The disambiguating can include identifying entities specified for the majority of entities determined ambiguous. It would use a machine learning-based annotator trained to recognize entities and entity attributes in the data. It would also annotate the data representing searcher action data about entities of many entity types with respective entity identifiers.

The system receives a query that is specific to the particular searcher. The question includes terms that state a first entity type of entities and an action taken by the searcher about entities. The system determines entities of the first entity type and queries the data that represents actions taken by the searcher about entities of the first entity type. The system generates an answer to the query from the information that represents the actions taken by the particular searcher about entities of the first entity type and provides the answer to the question specific to the specific searcher.

The data get seen as searcher action data, such as financial transaction data. The system may receive searcher action data that represents a financial transaction. The searcher action data may include:

• Credit card transaction data
• Debit card transaction data
• Financial statement transaction data

The system may analyze financial transaction data to provide information about the financial transactions. For example, the system may categorize financial transactions to view the amount or percentage that the searcher spent in particular categories. A category may include “restaurants,” and the system may index the financial transaction at the restaurant “Italian Deli” under restaurants as the corresponding searcher action data consists of the word “Deli.” But, financial transaction data may get more ambiguous. For example, financial transaction data [“4/11/2013,” “BEST R NEW YORK NY,” “45.78”] may become unclear.

Disambiguating Financial Transaction Data

To disambiguate financial transaction data, the system may use a machine-learning-based annotator that identifies entities associated with financial transaction data. Entities may be a business entity, such as:

• A particular store
• Restaurant
• Office
• Company

For example, the machine-learning-based annotator may identify the “Best Restaurant” entity located in New York, N.Y. It gets associated with the financial transaction data.

Associations Based on Labeled Training Data

The machine-learning-based annotator may get trained to identify entities associated with searcher action data based on labeled training data. For example, the machine-learning-based annotator may get trained to determine that financial transaction data, including locations, e.g., “NEW YORK NY,” “DC,” or “90210,” are likely to get associated with entities associated with the areas.

And, the machine-learning-based annotator may get trained to determine that a single letter, e.g., “R,” maybe an abbreviation for a word that begins with the letter in an entity’s name. The machine-learning-based annotator may determine based on the presence of “NEW YORK NY,” “BEST,” and “R” that the financial transaction data gets associated with the entity named “Best Restaurant” located in New York, N.Y.

Also, the machine-learning-based annotator may get trained only to analyze the description in the financial transaction data or to analyze more information in the financial transaction data, e.g., amount and date. For example, the machine-learning-based annotator may further determine the financial transaction associated with “Best Restaurant” based on considering the amount and the date. The annotator may choose that the amount of $45.78 is consistent with the amount spent at the restaurant “Best Restaurant” and that the searcher that was a party to the financial transaction was in New York, N.Y. on Apr. 11, 2013, based on the searcher’s profile.

Searcher Action Data That is Non-Financial Transaction Data

The patent also tells us that the labeled training data used to train the machine-learning-based annotator may include text and entities identified as becoming associated with the text. For example, the machine-learning-based annotator may get prepared using excerpts of the website for “Best Restaurant” that get identified as becoming associated with the entity “Best Restaurant.” The labeled training data may include other non-financial transaction data, such as:

• Text from a review website, associated with identified entities
• Social network interactions of searchers
• Search terms used by searchers
• Searcher provided training data including explicit confirmations that entities identified are correct or incorrect

Additionally, the labeled training data may include financial transaction data associated with identified entities. For example, the labeled training data may consist of financial transactions representing financial transactions at restaurants, including “Best Restaurant.”

The system may associate the financial transactions with entities identified by the machine-learning-based annotator. The method may annotate financial transaction data with data that includes an identifier that represents the recognized entity. For example, the system may annotate the financial transaction data [“4/11/2013,” “BEST R NEW YORK NY,” “45.78”] with an entity identifier “00542687” to result in annotated financial transaction data [“4/11/2013,” “BEST R NEW YORK NY,” “45.78,” “00542687”] where “00542687” is a unique identifier for the entity “Best Restaurant.”

The System May Store Entity Data About The Entity

The system may store entity data about the entity. For example, entity data for an entity that is a restaurant may include:

• A restaurant name
• Address
• Phone number
• Hours
• Dining style
• Cuisine type
• Executive chef name
• Price range

An entity for a clothing store may include:

• A store name
• Type of clothing
• Target audience
• Hours
• Price range

Other types of entities may include different values for other properties.

Searcher Action Data Can Provide Information About Financial Transactions

Using the annotated financial transaction data, the system may provide information about financial transactions. For example, if a searcher requests that the system identify what restaurant the searcher ate at on Apr. 11, 2013, the system may identify financial transactions in April based on the annotated financial transaction data. 11, 2013, and that get associated with an entity that is a restaurant and returns the restaurant’s result “Best Restaurant.”

The subject matter described in this patent may include detailed financial transaction data about a specific financial transaction. It may provide the particular financial transaction data to a machine-learning-based annotator that identifies entities associated with financial transaction data. More actions may include associating the specific financial transaction with an entity identified by the machine-learning-based annotator for the detailed financial data.

And, the subject matter described in this specification may get embodied in methods that may include the actions of receiving a query from a searcher that was a party to a particular financial transaction and determining an entity type associated with the question. More measures may include determining that an entity associated with the specific financial transaction matches the entity type associated with the query. Further actions may include providing, in response to the query, a reply that identifies the entity.

Other versions include corresponding systems, apparatus, and computer programs, configured to perform the actions of the methods encoded on computer storage devices.

Other Types of Financial Transactions

These and other versions may each include the following features. The financial transaction data may include:

• Credit card transaction data
• Debit card transaction data
• Financial statement transaction data

In certain aspects, associating the particular financial transaction with an entity identified by the machine-learning-based annotator may include annotating the specific financial transaction data with an identifier representing the recognized entity.

The actions may include receiving data identifying candidate entities from the machine-learning-based annotator and selecting the entity associated with the particular financial transaction from the identified candidate entities based on a searcher profile of a searcher that was a party to the specific financial transaction.

Selecting the entity that gets associated with the particular financial transaction from the candidate entities based on a searcher profile may include determining that, on a date indicated by the financial transaction data, the searcher profile suggests that the searcher gets located in a location associated with the entity.

The Machine-Learned Disambiguation of Searcher Ation Data Patent

The machine-learning-based annotator gets trained using labeled training data.

The actions in the patented Searcher Action Data process may include:

• Receiving a query from a searcher that was a party to the particular financial transaction
• Determining an entity type associated with the query
• Deciding that the entity associated with the particular financial transaction matches the entity type associated with the query
• Providing, in response to the query, a response that identifies the entity

The patent is here:

Machine-learned disambiguation of searcher action data
ventors: Mathew Cowan
Assignee: Google LLC
US Patent: 11,151,198
Granted: October 19, 2021
Filed: January 22, 2020

Abstract

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for receiving data that represents actions taken by a particular searcher about entities of a plurality of entity types, disambiguating the data by identifying entities specified in the data using a machine learning-based annotator that gets trained to recognize entities and annotating the data, receiving a query-specific from the particular searcher, querying the data that represents actions taken by the particular searcher, generating an answer to the query from the data representing the actions taken by the particular searcher, and providing the answer to the query.

About Bill Slawski

With more than 26 years of SEO experience and a Juris Doctor Degree, Bill Slawski is the foremost expert on Google’s patents as related to SEO. Patent Exploration is one of the quickest and most detailed ways to find new information about SEO. Bill is the Editor of SEO by the Sea, a prominent search engine optimization blog, where he is the author of over 1,300 posts. Bill’s experience includes Fortune 500 brands and some of the largest websites in the world. Bill is a contributing author for Moz, Search Engine Land, and Search Engine Journal. In 2014-2021, he spoke at industry-leading international conferences about topics including search engine algorithms, universal and blended search, personalization in search, search and social, and duplicate content problems, structured data, and schema