Computer Science · Big Data · Topic 2 of 5

Before anything gets analysed, it has to be collected, cleaned, and shaped.

The 2026 specification names this whole chain explicitly: "generation, processing and analysing data in different formats." Papers repeatedly ask you to distinguish structured from unstructured data and explain how that difference changes what a platform can actually do with it.

AS91908 · Big Data·Builds on Topic 1's vocabulary

Photo: Omar Al-Ghosson / Unsplash

Executive summary

Goal: explain where big data comes from, define structured vs unstructured (vs semi-structured) data precisely, and describe the pipeline that turns raw data into something a system can analyse.

Section	What it covers
Where big data comes from	Five common sources, none of them "someone typing it in"
Structured vs unstructured data	The most consistently examined distinction in this topic
A simple processing pipeline	Collection → Cleaning → Storage → Processing → Analysis
Worked example	Structured vs unstructured data on X/TikTok, straight from a past paper

Where big data comes from

Big data is generated continuously by many different sources, often without a single person deliberately "entering" it:

Source	Example
User activity	Posts, likes, shares, comments, watch time on a platform
Sensors / IoT devices	Smartwatches, smart thermostats, industrial machinery sensors
Transactions	Purchases, bank transfers, ride-share trips
Location data	GPS from phones, Wi-Fi/cell tower triangulation
Web and app logs	Every page view, click, and search query

Structured vs unstructured data

This distinction is one of the most consistently examined ideas in this topic.

	Structured data	Unstructured data
Definition	Organised into a fixed, predictable format (rows/columns, key-value pairs)	Has no predefined format or model
Examples	User IDs, timestamps, like counts, follower numbers, stored in a database table	Video content, images, free-text captions, comments
How it's queried	Directly with structured query languages (e.g. SQL) — fast, precise	Requires additional processing (e.g. natural language processing, computer vision) before it can be queried meaningfully
Storage	Relational databases	Distributed file systems, object storage, NoSQL databases

Semi-structured data sits between the two — it has some organisational tags but not a fixed schema. JSON or XML data (e.g. a social media post object that contains structured fields like timestamp and like_count alongside an unstructured caption field) is the clearest example, and it's worth naming this middle category if a scenario clearly doesn't fit neatly into either extreme.

Why the difference matters for processing

Because unstructured data has no fixed shape, it can't be searched or aggregated the way a spreadsheet column can. A platform can instantly calculate "average likes per post" from structured numeric fields with a simple query. To do anything with the content of a video or the sentiment of a caption, it first needs to be run through a separate process — object recognition, speech-to-text, sentiment analysis — that converts it into structured features a system can then analyse alongside the rest.

A simple processing pipeline

Regardless of format, big data generally moves through the same broad stages before it becomes useful:

• Collection — data is captured from the source (an app, a sensor, a form)
• Cleaning & validation — duplicates, missing values, and obviously invalid entries (e.g. impossible timestamps) are removed or flagged
• Storage — structured and unstructured data are often stored differently (see Topic 4)
• Processing — unstructured data is converted into structured, analysable features
• Analysis / output — patterns, trends, or predictions are generated from the processed data

Errors propagate forward

A key exam point: errors introduced early in this pipeline propagate forward. If cleaning misses duplicate accounts, every later statistic (average engagement, trend velocity) is skewed, even though the analysis step itself was performed correctly.

Past-paper practice

(i) What is the difference between structured and unstructured data from X? (ii) How do these differences impact the methods used to process and analyse the data? (91908-cat-B-2023.pdf, Question Two (d); the equivalent TikTok version appears in 91908-cat-A-2023.pdf, Question Two (d).)

Reveal how to answer well

• For (i): give the definition and a platform-specific example of each — e.g. structured: post timestamp, retweet count, follower count; unstructured: the text of the post itself, an attached image or video
• For (ii): don't just repeat the definitions — explain the consequence. Structured data supports fast, direct statistical queries (e.g. "posts per hour"); unstructured data must first be converted (e.g. natural language processing to detect topic or sentiment) before it can be combined with structured statistics

A related question style asks you to go further:

How could a platform process and analyse data in different types and formats (such as numbers and text stored in databases) and images, videos, and social media posts? (91908-cat-A-2023.pdf / 91908-cat-B-2023.pdf, Question Two (f).)

This is asking you to describe the pipeline above in the platform's own terms: structured fields go straight into database queries; text is processed with language techniques to extract meaning; images and video are processed with computer-vision techniques to extract tags or detect content — and all of these outputs can then be combined for a fuller analysis.

Exit question

A platform logs a post's timestamp, like-count, caption text, and an attached video. Sort these into structured and unstructured, and explain what has to happen to the unstructured ones before they can be analysed alongside the rest.

Key terms

• structured data
• unstructured data
• semi-structured data
• data cleaning / validation
• data pipeline
• natural language processing (NLP)
• computer vision

1. 1
2. 2
3. 3
4. 4
5. 5

← Back to Topic 1: Characteristics of Big Data

Next → Topic 3: Interpretation, Bias & Display