User-Specific Knowledge Graphs to Support Queries and Predictions

user-specific Knowledge graph system

Sharing is caring!

A recently granted patent from Google is about supporting querying and predictions, and it does this by focusing on user-specific knowledge graphs.

Those User Specific Knowledge Graphs can be specific to particular users.

This means Google can use those graphs to provide results in response to one or more queries submitted by the user, and/or to surface data that might be relevant to the user.

I was reminded of another patent that I recently wrote about when I saw this patent, in the post Answering Questions Using Knowledge Graphs, where Google may perform a search on a question someone asks, and build a knowledge graph from the search results returned, to use to find the answer to their question.

So Google doesn’t just have one knowledge graph but may use many knowledge graphs.

New ones for questions that may be asked, or for different people asking those questions.

This User-Specific Knowledge Graph patent tells us that innovative aspects of the process behind it include:

  1. Receiving user-specific content
  2. The user-specific content can be associated with a user of one or more computer services
  3. That user-specific content is processed using one or more parsers to identify one or more entities and one or more relationships between those entities
  4. A parser being specific to a schema, and the one or more entities and the one or more relationships between entities being identified based on the schema
  5. This processes provides one or more user-specific knowledge graphs
  6. A user-specific knowledge graph being specific to the user, which includes nodes and edges between nodes to define relationships between entities based on the schema
  7. The process includes storing the one or more user-specific knowledge graphs

Optional Features involving providing one or more user-specific knowledge graphs may also include:

  • Determining that a node representing an entity of the one or more entities and an edge representing a relationship associated with the entity are absent from a user-specific knowledge graph
  • Adding the node and the edge to the user-specific knowledge graph
  • The edge connecting the node to another node of the user-specific knowledge graph

Actions further include:

  1. Receiving a query
  2. Receiving one or more user-specific results that are responsive to the query
  3. The one or more user-specific results are provided based on the one or more user-specific knowledge graphs
  4. Providing the one or more user-specific results for display to the user
  5. An edge is associated with a weight
  6. The weight indicating a relevance of a relationship represented by the edge
  7. A value of the weight increases based on reinforcement of the relationship in subsequent user-specific content
  8. A value of the weight decreases based on lack of reinforcement of the relationship in subsequent user-specific content
  9. A number of user-specific knowledge graphs are provided based on the user-specific content
  10. Each user-specific knowledge graph being specific to a respective schema
  11. The user-specific content is provided through use of the one or more computer-implemented services by the user

Advantages of Using the User-Specific Knowledge Graph System

The patent describes the advantages of implementing the process in this patent:

  1. Enables knowledge about individual users to be captured in a structured manner
  2. Enabling results to be provided in response to complex queries, e.g., series of queries, regarding a user
  3. The user-specific knowledge graph may provide a single canonical representation of the user based on user activity inferred from one or more computer-implemented services
  4. User activities could be overlapping, where reconciliation of the user-specific knowledge graph ensures a canonical entry is provided for each activity
  5. Joining these together could lead to a universal knowledge graph, e.g., non-user-specific knowledge graph, and user-specific knowledge graphs

(That Universal Knowledge Graph sounds interesting.)

Information from sources like the following may be used to create User-Specific Knowledge Graphs:

  • A user’s social network
  • Social actions or activities
  • Profession
  • A user’s preferences
  • A user’s current location

This is so that content that could be more relevant to the user is used in those knowledge graphs.

We are told also that “a user’s identity may be treated so that no personally identifiable information can be determined for the user,” and that “a user’s geographic location may be generalized so that a particular location of a user cannot be determined.”

The User-specific Knowledge Graph Patent

This patent can be found at:

Structured user graph to support querying and predictions
Inventors: Pranav Khaitan and Shobha Diwakar
Assignee: Google LLC
US Patent: 10,482,139
Granted: November 19, 2019
Filed: November 5, 2013

Abstract

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for receiving user-specific content, the user-specific content being associated with a user of one or more computer-implemented services, processing the user-specific content using one or more parsers to identify one or more entities and one or more relationships between entities, a parser being specific to a schema, and the one or more entities and the one or more relationships between entities being identified based on the schema, providing one or more user-specific knowledge graphs, a user-specific knowledge graph being specific to the user and including nodes and edges between nodes to define relationships between entities based on the schema, and storing the one or more user-specific knowledge graphs.

What Content is in User-Specific Knowledge Graphs?

The types of services that user-specific knowledge graph information could be pulled from can include:

  • A search service
  • An electronic mail service
  • A chat service
  • A document sharing service
  • A calendar sharing service
  • A photo sharing service
  • A video sharing service
  • Blogging service
  • A micro-blogging service
  • A social networking service
  • A location (location-aware) service
  • A check-in service
  • A ratings and review service

A User-Specific Knowledge Graph System

This patent describes a search system that includes a user-specific knowledge graph system as part of that search system, either directly connected to or connected to search system over a network.

The search system may interact with the user-specific knowledge graph system to create a user-specific knowledge graph.

That user-specific knowledge graph system may provide one or more user-specific knowledge graphs, which can be stored in a data store.

Each user-specific knowledge graph is specific to a user of the one or more computer-implemented services, e.g., search services provided by the search system.

The search system may interact with the user-specific knowledge graph system to provide one or more user-specific search results in response to a search query.

Structured User Graphs For Querying and Predictions

A user-specific knowledge graph is created based on content associated with the user.

These user-specific knowledge graphs include a number of nodes and edges between nodes.

A node represents an entity and an edge represents a relationship between entities.

Nodes and/or entities of a user-specific knowledge graph can be provided based on the content associated with a respective user, to which the user-specific knowledge graph is specific.

User-Specific Knowledge Graphs and Schemas

The user-specific knowledge graphs can be created based on one or more schemas (examples follow). A schema describes how data is structured in the user-specific knowledge graph.

A schema defines a structure for information provided in the graph.

A schema structures data based on domains, types, and properties.

A domain includes one or more types that share a namespace.

A namespace is provided as a directory of uniquely named objects, where each object in the namespace has a unique name or identifier.

For example, a type denotes an “is a” relationship about a topic, and is used to hold a collection of properties.

A topic can represent an entity, such as a person, place or thing.

Each of these topics can have one or more types associated with them.

A property can be associated with a topic and defines a “has a” relationship between the topic and a value of the property.

In some examples, the value of the property can include another topic.

A user-specific knowledge graph can be created based on content associated with a respective user.

That content may be processed by one or more parsers to populate the user-specific structured graph.

A parser may be specific to a particular schema.

Confidence or Weights in Connections

Weights that are assigned between nodes indicate a relative strength in the relationship between nodes.

The weights can be determined based on the content associated with the user, which content underlies provision of the user-specific knowledge graph.

That content can provide a single instance of a relationship between nodes, or multiple instances of a relationship between nodes.

So, there can be a minimum value and a maximum value.

Weights can also be dynamic:

  • Varying over time based on content associated with the user
  • Based on content associated with the user at a first time
  • Based on content or a lack of content associated with the user at a second time
  • The content at the first time can indicate a relationship between nodes
  • Weights can decay over time

Multiple User Specific Knowledge Graphs

More than one user-specific knowledge graph can be provided for a particular user.

Each user-specific knowledge graph may be specific to a particular schema.

Generally, a user-specific knowledge graph includes knowledge about a specific user in a structured manner. (It represents a portion of the user’s world through content associated with the user through one or more services.)

Knowledge captured in the user-specific knowledge graph can include things such as:

  • Activities
  • Films
  • Food
  • Social connections, e.g., real-world and/or virtual
  • Education
  • General likes
  • General dislikes

User-Specific Knowledge Graph Versus User-Specific Social Graph

A social graph contains information about people who someone might be connected to, where a user-specific Knowledge graph also overs knowledge about those connections, such as shared activities between people who might be connected in a knowledge graph.

Examples of Queries and User-Specific Knowledge Graphs

User-specific Knowledge graph example

These are examples from the patent. Note that searches, emails, social network posts may all work together to build a user-specific Knowledge Graph as seen in the combined messages/actions below, taken together, which may cause the weights on edges between nodes to become stronger, and nodes and edges to be added to that knowledge graph.

Example search query: [playing tennis with my kids in mountain view] to a search service

Search results: which may provide information about playing tennis with kids in Mountain View, Calif.

Nodes can be provided, with one representing the entity “Tennis,” one representing “Mountain View,” one representing “Family,” and a couple more each representing “Child.”

An edge can be provided that represents a “/Location/Play_In” relationship between the nodes, another edge may represent a “/Sport/Played_With” relationship between the nodes and other edges may represent “/Family/Member_Of” relationships between the node and the nodes.

Weights may be generated for each of the edges to represent different values as well.

A Person may post the example post “We had a great time playing tennis with our kids today!” in a social networking service, associated with geo-location data indicating Mountain View, Calif.

Nodes may be identified representing tennis, Mountain View, family and children, and edges between those nodes.

Weights may be generated between those edges.

Someone may receive an electronic message from a hotel, which says “Confirming your hotel reservation in Waikiki, Hi. from Oct. 15, 2014, through Oct. 20, 2014. We’re looking forward to making your family’s vacation enjoyable!”

Nodes can be added to the user-specific Knowledge graph, where those nodes represent the entities “Vacation” and “Waikiki”

Edges can be created in the user-specific knowledge graph in response to that email that represents a “/Vacation/Travelled_With” relationship between the nodes, one that represents a “/Vacation/CityTown” relationship between the nodes, and another edge that represents a “/Vacation/CityTown” relationship between the nodes.

Timing nodes may also be associated with the other nodes, such as a timing node representing October 2014, or a node representing a date range of Oct. 15, 2014, through Oct. 20, 2014.

The user can submit the example search query [kids tennis lessons in waikiki] to a search service.

Nodes may be created in the user-specific knowledge graph representing tennis, Waikiki, family, and children, as well as respective edges between at least some of the nodes.

That example search query may reinforce the relevance of the various entities and the relationships between the entities to the particular user.

That reinforcement may cause the respective weights associated with the edges to be increased.

The user can receive an email from a tennis club, which can include “Confirming tennis lessons at The Club of Tennis, Waikiki, Hi.”

Nodes represent tennis, and Waikiki, and the edges between them.

That email reinforces the relevance of the entities and the relationships between the entities to the particular user.

The weights between the entities could be increased, and a node could be added to represent the entity “The Club of Tennis,” which could then be connected to one or more other nodes.

User-Specific Knowledge Graphs Takeaways

This reminds me of personalized search, but tells us that it is looking at more than just our search history – It includes data from sources such as emails that we might send or receive, or posts that we might make to social networks. This knowledge graph may contain information about the social connections we have, but it also contains knowledge information about those connections as well. The patent tells us that personally identifiable information (including location information) will be protected, as well.

And it tells us that User-specific knowledge graph information could be joined together to build a universal knowledge graph, which means that Google is building knowledge graphs to answer specific questions and for specific users that could potentially be joined together, to enable them to avoid the limitations of a knowledge graph based upon human-edited sources like Wikipedia.

Added 11/26/2019: A recent whitepaper from Google covers much of the same territory as this patent, even though this patent was filed originally back in 2013. It’s worth reading that white paper when you read about this patent because they are similar in many ways. The paper is Personal Knowledge Graphs: A Research Agenda by Krisztian Balog and Tom Kenter.

Last updated 11/26/2019

Sharing is caring!

9 thoughts on “User-Specific Knowledge Graphs to Support Queries and Predictions”

  1. Hey!

    StrategicERP here!

    We are Software providers to Real Estates and other B2B businesses

    Thank you for this article!

    I personally liked this article coz I got to know too many things from which I was unaware

    I’ll love to see more articles like this in the upcoming year…

    Regards,

    StrategicERP

  2. Hi StrategicERP,

    I will be trying to find more things to write about this year that are new and not covered by lots of other people.

  3. A well-written article about the user-specific knowledge graphs! Google makes their algorithm better to offer accurate search results for the search phrases by the users. Great to know that this giant is building the knowledge graphs to offer the best results. The example of search query and its explanation are good. Thanks for your great efforts to explain the things in a precise way, keep doing your great job!

  4. Hi Sakthi,

    Thank you. This was interesting to see as a patent that had been granted by Google. I am adding a link to a Google Paper that came out this year about Personal Knowledge graphs to this post because it’s worth reading about that too.

  5. Very informative and well-illustrated user-specific knowledge graph. Very complex topic covered in simple logical elaboration. Thanks and kept for reference.

  6. Hi Andy,

    A little different. A personal search is based upon your search history, where this approach using a user-specific knowledge graph can be based upon queries that you perform, emails that you send and receive, and social posts that you might make, and could possibly expand to things such as calendar events. and many others which I listed above. And it works to understand the entities in those data sources and understand relationships between those entities, such as the place you go to for playing tennis, or that you might play the guitar, or the people who you may know and engage with.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.