Understanding Knowledge Graphs and Their Uses
Key Points
- Knowledge graphs power virtual assistants by storing semantic relationships—e.g., “Ottawa” linked to “Canada” via a “capital” edge—enabling quick factual answers.
- They are composed of nodes (entities) and edges (relationships), allowing multiple, diverse connections between the same entities (e.g., Paris → France as “capital” and Paris → Roman Empire as “city of”).
- By integrating multiple data sources (census data, online reviews, etc.) into a unified graph, statistical inference can fill gaps and reveal more accurate insights, such as estimating the true number of Chinese restaurants in New York City.
- Natural language processing is used for semantic enrichment, converting unstructured text into linked nodes and edges, which supports not only question‑answering but a wide range of commercial applications.
Sections
- Explaining Knowledge Graph Basics - The speaker introduces knowledge graphs, showing how they represent entities and multiple relationships—such as capitals, historical affiliations, movies, or recipes—to enable machines like virtual assistants to retrieve and connect contextual information.
- Semantic Enrichment via Knowledge Graphs - The speaker explains how natural language processing creates semantically enriched knowledge graphs from unstructured text and highlights varied commercial uses—including video recommendations, insurance fraud detection, and retail product pairing—illustrated with a simple human‑coffee‑sleep graph.
Full Transcript
# Understanding Knowledge Graphs and Their Uses **Source:** [https://www.youtube.com/watch?v=y7sXDpffzQQ](https://www.youtube.com/watch?v=y7sXDpffzQQ) **Duration:** 00:05:34 ## Summary - Knowledge graphs power virtual assistants by storing semantic relationships—e.g., “Ottawa” linked to “Canada” via a “capital” edge—enabling quick factual answers. - They are composed of nodes (entities) and edges (relationships), allowing multiple, diverse connections between the same entities (e.g., Paris → France as “capital” and Paris → Roman Empire as “city of”). - By integrating multiple data sources (census data, online reviews, etc.) into a unified graph, statistical inference can fill gaps and reveal more accurate insights, such as estimating the true number of Chinese restaurants in New York City. - Natural language processing is used for semantic enrichment, converting unstructured text into linked nodes and edges, which supports not only question‑answering but a wide range of commercial applications. ## Sections - [00:00:00](https://www.youtube.com/watch?v=y7sXDpffzQQ&t=0s) **Explaining Knowledge Graph Basics** - The speaker introduces knowledge graphs, showing how they represent entities and multiple relationships—such as capitals, historical affiliations, movies, or recipes—to enable machines like virtual assistants to retrieve and connect contextual information. - [00:03:11](https://www.youtube.com/watch?v=y7sXDpffzQQ&t=191s) **Semantic Enrichment via Knowledge Graphs** - The speaker explains how natural language processing creates semantically enriched knowledge graphs from unstructured text and highlights varied commercial uses—including video recommendations, insurance fraud detection, and retail product pairing—illustrated with a simple human‑coffee‑sleep graph. ## Full Transcript
Perhaps you're unfamiliar
with the term of
knowledge graph,
but I suspect you
have benefited from one, maybe
even today.
Take your favorite virtual
assistant.
Did you know that when you ask
a question like what is the capital
of Canada?
The assistant is pulling the
response.
Ottawa in this case from information
in a knowledge graph
and not as graphs can be seen as a
way of representing semantic
information between two entities.
And what's really cool about them
is that modern applications allow
almost any entity you could imagine
to be described with anyone.
For example, we could have a
knowledge graph of movies and
actors, or we can describe
ingredients or recipes, as well as
the steps required to cook them.
And this means machines are able to
understand how these entities
relate to each other, along
with a shared attributes, and this
allows us to draw connections
between different things in the
world around us.
Now, a knowledge graph is made
up of nodes,
and connected by
edges.
Nodes describe any object
or person or place, and an
edge defines the relationship
between the nodes.
So, for example, a
node of Ottawa..
and
a node of Canada..
might be connected by the edge
of capital.
And the thing here is that
the pair of nodes, they can be
connected by more than one relation.
If the two are related in multiple
ways. So for example,
let's take another city.
Let's take Paris.
Now, Paris
is the capital of
France,
but it is also part
of the Roman Empire.
Or it did used to be.
So in this case, the edges are
Paris to France is capital.
And then Paris to Roman
Empire is
city of.
And we can see then how that these
nodes can be connected with multiple
edges as we expand this.
And knowledge graphs can sort of
build different data sources
and bind them together to infer
missing facts. So let's say you're
trying to predict the number of
Chinese restaurants in New York
City.
You could use one data source, let's
say census data, but that might not
tell you the whole story.
It might be out of date.
It might not classify everything
correctly and so forth.
So if we had a second data source
like, say, online reviews about
all the different restaurants and
put them all in a knowledge graph,
then we can use statistical methods
to infer that actually there are
two thousand nine hundred
restaurants serving Chinese food
in New York City, which may be a lot
different than what was reported in
the census data alone.
Now, knowledge graphs utilize
something called natural
language processing.
Or abbreviated
to NLP
to construct a view of
nodes and edges through a process
called semantic enrichment.
I can take some unstructured text,
say a white paper, and
classify that text using natural
language processing
to really sort of create datasets
which are correlated and
related to that
information.
And one that builds me is a
knowledge graph.
And beyond sort of helping with
question and answer queries, there
are a lot of other commercial
applications for knowledge graphs.
So, for example, the recommended
videos that are probably appearing
alongside this one in YouTube right
now. Well, they leverage
a knowledge graph based on queries
people are searching for and other
videos that they might enjoy.
In insurance you can use knowledge
graphs to sort of and make sure
that a given claim
for damage is actually
a true claim, or whether it's been
one that's reported by a
policyholder for fraud.
And in retail, knowledge graphs
can assist understanding the
relationship between products so
that companies can recommend
different pairings that might be of
interest to customers.
And I'll leave you here with some
wisdom that I was acutely aware
of last night and I'll
share in knowledge graph form.
It consists of three
nodes. There's human.
Then there is coffee.
And then there is sleep.
And these note connected, of
course, by edges.
The edge between human
and coffee is consumed.
The edge between human
and sleep is needs.
And what I learned last night, the
edge between coffee and sleep,
is prevents.
Avoid caffeine after five p.m.
folks.
If you have any questions, please
drop us a line below, and if
you want to see more videos like
this in the future, please like and
subscribe.
Thanks for watching.