Answering entity-seeking queries

kopp-online-marketing.com/patents-papers/answering-entity-seeking-queries

Author: Olaf Kopp

Reading time: 5 Minutes

Topics: Entity based search, Probably in use, Search Query Processing, Semantic
Search

5/5 - (1 vote)

This patent outlines a system designed to enhance how queries are processed to identify
relevant entities more efficiently. The system utilizes a complex method of mapping
queries to a dependency tree, discerning entity types from the root of the tree, and then
identifying specific entities based on their relevance to the query’s terms. This helps to
generate precise and contextually appropriate responses.

Patent ID: US20190370326A1

Countries Published: United States, WIPO
Publication Date: December 5, 2019
Filing Date: May 29, 2018
Applicant/Assignee: Google LLC, Mountain View, CA
Inventors: Mugurel Ionut Andreica, Tatsiana Sakhar, Behshad Behzadi, Marcin M.
Nowak-Przygodzki, Adrian-Marius Dumitran

Background

The background section of the patent discusses the processing of various user queries by
computing systems. These queries can range from simple search terms to specific
commands or questions requiring tailored responses. For example, a query might be a
search for “Isaac Asimov books,” a command like “Turn off lights in the living room,” or a
question such as “What temperature is it outside?” The system is designed to handle
these requests by providing appropriate responses or actions, enhancing the interaction
between users and computing systems.

1/5
The claims of the patent outline a computer-implemented method and system for
processing queries by:

Obtaining a Query: Receiving a query that includes a sequence of terms.

Mapping to a Dependency Tree: Mapping the query based on the sequence of
terms to a dependency tree that illustrates the relationships among the terms.

This process involves converting a user’s query, which consists of a sequence

of terms, into a structured format known as a dependency tree. This tree
visually represents how the terms in the query are related to each other,
highlighting their syntactic or semantic dependencies.
Process Details:
Each term in the query is represented as a node.
Directed edges between nodes indicate dependency relationships,
where one term modifies or relates to another.
The root node typically represents the main focus of the query, which
does not modify other terms but is itself modified by other terms.
Purpose: The dependency tree helps the system understand the grammatical
structure and semantic relationships within the query, which is crucial for
accurately determining what the user is asking for.

2/5
 Determining an Entity Type: Identifying the type of entity sought by the query
based on a term represented by the root of the dependency tree.

Based on the dependency tree, this step involves identifying the type of entity
that the query is seeking. This is derived from the root of the tree or other
significant nodes.
Process Details:
The system examines the term at the root node to determine if it
represents a type of entity (e.g., “restaurant,” “movie”).
It may also consider adjacent nodes or children of the root if they
contribute to specifying the entity type (e.g., “Chinese restaurant”).
The determination can include analyzing the node that represents a type
of entity and also its relationship with action-oriented terms if present.
Purpose: Identifying the correct entity type is critical for narrowing down the
search to relevant entities, thus improving the efficiency and relevance of the
response.

Identifying a Particular Entity: Selecting a specific entity based on the entity type
and its relevance to the terms in the query.

Once the entity type is established, the system identifies one or more specific
entities that match the criteria set out in the query. This involves evaluating
how relevant each potential entity is to the terms used in the query.
Process Details:
The system calculates a relevance score for potential entities based on
how well they match the terms and structure of the query.
It may set a relevance threshold, which must be met or exceeded for an
entity to be considered as a response to the query.
If the query potentially refers to multiple entities, the system may request
additional input from the user to refine the search or decide among
closely scored options.
Purpose: This step ensures that the system not only understands the type of
entity being sought but also precisely identifies the most relevant entity or
entities, thus tailoring the response to the specific needs or context indicated
by the user.

Providing a Response: Delivering a response to the query based on the identified

entity.

Additional claims specify:

The use of a semantic or syntactic dependency tree for mapping.

Processes for adjusting the method based on the query’s complexity or specificity,
such as determining if the query seeks a single entity or multiple entities.
Mechanisms for refining the entity identification process by considering additional
user inputs when multiple relevant entities are identified.

3/5
Figure 2 in the patent document depicts an example of a semantic dependency tree associated with a
specific query. This tree visually represents the structure and relationships of terms within the query:
Root (220): Represents the main entity type sought in the query, in this case, “Chinese restaurant”.
Children (232, 234): Connected to the root by directed edges, these nodes represent additional terms
that modify or specify the root entity. “Call” (232) indicates an action associated with the entity, and
“Street” (234) links to further specific nodes. Subsequent Children (242, 244): These nodes extend
from the “Street” node, further specifying the location with “Piccadilly” (242) and “Fifteen” (244), which
completes the address.

4/5
 Figure 1 in the patent document illustrates a block diagram of the system 100, designed to handle
queries that seek entities. This system includes several key components: Query Mapper (110):
Receives a user’s query and maps it to a semantic dependency tree, analyzing the structure and
relationships among the terms. Entity Type Identifier (120): Determines the type of entity the query is
searching for, based on the semantic dependency tree. Entity Identifier (130): Identifies specific
entities that match the query’s criteria using data from various data stores and the determined entity
type. Subquery Resolver (140): Resolves parts of the query by replacing them with the identified
entities, effectively simplifying the query into a more direct action or response. Query Responder (150):
Provides the final response to the query based on the resolved entities and actions. Data Stores (160-
166): Include web resource data, user location data, user activity data, and entity geographical data,
supporting the entity identification and query resolution processes.

COMMENT ARTICLE

Logged in as aaqib Shah. Edit your profile. Log out? Required fields are marked *

5/5