The Three Types of Data
Level 3 · Kinds of data
Structured, semi-structured and unstructured data, and how to tell them apart. What each one looks like, where it is kept, and why the shape matters.
Course contentsLesson 7 of 11
Level 3 · Kinds of data
Level 4 · Models and workloads
After this lesson you can
- Sort a piece of data into one of the three types
- Say where each type is usually kept
- Explain why shape decides how easy data is to count
10 mincovers 3 glossary entries
Worth reading first: Data Sources and Data Silos
How structured, semi-structured and unstructured data differ
Data comes in three shapes. Structured data fits in rows and columns, like a table of orders. Semi-structured data carries labels but no fixed table, like a JSON file. Unstructured data has no labels at all, like an email, a photo or a phone call. The shape decides how easy the data is to search and count.
Structured data has a fixed set of columns, and every row fills the same ones. A table of orders has an order number, a date, a customer and a total. Because the shape is known, a database can check each value as it is saved. It can also add up millions of rows in a second.
Semi-structured data labels each value but lets records differ. In JSON, one order might list a coupon and the next might not. A record can also hold a list inside it, such as five items in one order. Logs from apps and replies from web APIs usually come in this shape.
Unstructured data has no labels a computer can use. A support email says the parcel came late, but there is no field called late. A photo of a broken item holds the facts in pixels. To count anything in it, a person or a model has to read it first and pull the facts out.
Most data moves toward more structure over its life. A team takes a JSON log and flattens it into a table. A model reads emails and tags each one with a topic. Each step makes the data easier to count. Each step can also lose something the original said.
People often call JSON unstructured because it is not a table. But it has a label on every value. A program can read it without guessing. That makes it semi-structured. Most warehouses today can query it as it is.
The three types side by side
| Type | Looks like | Labels on values | Usually kept in |
|---|---|---|---|
| Structured | A table of orders | Yes, fixed columns | A database or warehouse |
| Semi-structured | A JSON file, an app log | Yes, but records can differ | Files, or a warehouse that reads JSON |
| Unstructured | Emails, photos, calls | No | Files in cheap storage |
Where to keep each one
- 1
Structured
Keep it in a database or a warehouse. There, each column is checked as rows come in.
- 2
Semi-structured
Land it as files first. Then flatten the fields you use most into a table.
- 3
Unstructured
Keep the files in cheap storage. Save the facts you pull out of them as a table beside them.
Worked example
One shop, three shapes
Emma's shop holds 12,000 orders in a table, 2 million app log lines in JSON and 3,000 support emails.
Counting orders by day takes one query. Counting late parcels took a week. Then a model tagged each email and saved the tags as a table.
The words for this lesson3
Check yourself
Which type of data is a JSON file from an app?
Finished this lesson?
Marking it complete also marks its 3 glossary entries learned, which updates your roadmap and your saved page. 0 are already marked.