🎯 What You'll Learn

📋 Before You Begin

🧬 The 6 Advanced Data Types in Power BI

⚡ Intermediate ⏱ ~3 min

Part 3 introduces data types that go well beyond simple tables. Each type unlocks a different analytical dimension.

🖼️

Unstructured (Images)

Raw image files processed by AI Vision to extract text and object tags

💬

Text / Sentiment

Customer reviews scored 0–1 for positive/negative sentiment using NLP

📡

Streaming

Live sensor or API feeds that push rows into a dashboard in real-time

🕸️

Graph / Relational

Network data: nodes and edges representing people, links, or dependencies

🗂️

Hierarchy

Parent-child structures like org charts, flattened using DAX PATH()

🔮

Vector / Semantic

Text converted to numerical embeddings for semantic similarity search

🏋️ 10 Advanced Exercises

⚡ Advanced ⏱ ~25 min
1 Unstructured Data: Image Tagging with AI Vision
SOURCECreate a folder on your PC with 10 different photos (a car, a dog, a laptop, etc.).
TASKGet Data → Folder. In Power Query: click AI Insights → Vision → Tag Images. Power BI sends each image to Azure Cognitive Services.
GOALPower BI returns a table of object keywords per photo. Create a Word Cloud (from AppSource) to display the most common objects detected.
AI InsightsVisionUnstructured
2 Text Data: Sentiment Analysis on Customer Reviews
SOURCEWrite 10 fake customer reviews in an Excel sheet — mix of happy ("Great product!") and angry ("Never buying again.") ones.
TASKGet Data → Excel. In Power Query: AI Insights → Text Analytics → Score Sentiment. A new column "Sentiment Score" (0 to 1) is added.
GOALA Gauge Visual showing average sentiment score. 0 = very unhappy, 1 = very happy. Set target to 0.7 as your "happiness benchmark."
AI InsightsNLPText Data
3 Vector Data: Semantic Search with Microsoft Fabric
SOURCEUse the Microsoft Fabric trial. Upload a set of documents to a "Lakehouse."
TASKIn a Fabric Notebook, use Semantic Link to generate embeddings (vectors) for your text documents. Each document becomes a list of 1,536 numbers.
GOALIn Power BI, a Slicer lets users search by concept (e.g., "revenue") and finds documents about "sales" or "income" — even if those exact words don't appear.
Vector DataFabricSemantic Search
4 Streaming Data: Real-Time Live Dashboard
SOURCEUse a public PubNub sensor feed (simulated "Weather Station" with live values).
TASKIn Power BI Service (web): New Item → Streaming Semantic Model. Define the data schema (timestamp, temperature, humidity).
GOALA Real-Time Line Chart that moves every second as new sensor data is pushed. No manual Refresh needed — it auto-updates.
StreamingReal-TimePubNub
5 Graph Data: Network / Relationship Mapping
SOURCECreate an Excel sheet with two columns: "Person A" and "Person B" (e.g., Alice–Bob, Bob–Carol, Alice–David).
TASKIn Power BI, download the "Network Navigator" custom visual from AppSource (free). Load it onto the canvas.
GOALAn interactive social web showing nodes (people) and edges (connections). Click a node to highlight their direct connections.
Graph DataCustom VisualAppSource
6 Binary Data: Document Library from SharePoint
SOURCEConnect to a SharePoint folder containing PDFs or Word documents.
TASKIn Power Query, do NOT click "Combine Files." Instead, keep the Content (Binary) column and the Name column only.
GOALA clickable Document Library table — users click a row to open the original file from SharePoint. No data extraction needed.
Binary DataSharePoint
7 Geographical Data: KML / Shapefiles with Azure Maps
SOURCEDownload a .json or .kml file of city boundaries (e.g., London Boroughs from data.gov.uk).
TASKUse the Azure Maps visual in Power BI (enabled via Power BI Service settings).
GOALOverlay your sales data on top of custom geographic boundaries — custom boundaries that don't exist in standard Bing Maps.
KMLAzure MapsGeo Data
8 Hierarchy Data: Org Chart with DAX PATH()
SOURCEAn Excel sheet with columns: EmployeeID, EmployeeName, ManagerID (ManagerID points to another EmployeeID).
TASKCreate a new column: Path = PATH(Employees[EmployeeID], Employees[ManagerID]). This flattens the entire hierarchy into a pipe-delimited string.
GOALA Decomposition Tree visual that lets you drill down from the CEO through every management level to the newest intern.
HierarchyDAX PATH()Decomposition Tree
9 Time-Series: AI Sales Forecasting
SOURCEUse the Financials sample dataset — it has Sales data over 2 years, which is ideal for forecasting.
TASKCreate a Line Chart of Sales by Date. Click the Analytics Pane (magnifying glass icon). Find "Forecast" and expand it.
GOALTurn on Forecasting. Set forecast length to 6 months. Power BI's built-in ETS algorithm draws a shaded prediction cone on your chart.
Time SeriesAI ForecastAnalytics Pane
10 Multi-Modal: The "360 View" Product Catalog
SOURCEThree separate sources: a Table (Sales by Product), Text (Customer Reviews), and Images (Product Photos folder).
TASKIn Model View, link all three tables using a common Product ID column as the key. This is a star schema.
GOALA "Product Catalog" page where clicking one product updates: the sales chart, the product image, and the sentiment score — all at once.
Multi-ModalStar SchemaModel

🃏 Key Term Flashcards

⚡ Advanced ⏱ ~4 min

Click the card to reveal the definition. Use the arrows to navigate.

Embedding

Click to reveal definition

A numerical representation of text (e.g., 1,536 numbers) that captures meaning. Similar concepts end up "close" together in vector space, enabling semantic search.

1 / 6

🗺️ Exercise 10: Building the 360 View (Step-by-Step)

⚡ Advanced ⏱ ~6 min

This is the capstone exercise of the entire series. Follow these steps exactly.

1

Prepare Your Three Data Sources

Create: (a) Sales.csv with ProductID, SalesAmount columns. (b) Reviews.xlsx with ProductID, ReviewText. (c) A Photos folder with images named by ProductID (e.g., "P001.jpg").

2

Load All Three into Power BI

Use Get Data → Text/CSV for Sales, Get Data → Excel for Reviews, and Get Data → Folder for Photos. You should now have three separate tables in the Fields pane.

3

Run Sentiment Analysis on Reviews

In Power Query → select the Reviews table → AI Insights → Text Analytics → Score Sentiment. This adds a "Sentiment Score" column (0–1).

4

Build the Star Schema in Model View

Switch to Model View. Drag ProductID from Sales to ProductID in Reviews. Drag ProductID from Sales to the Name column in Photos. You now have a 3-table star.

5

Build the Report Page

Add a Product Slicer (from Sales[ProductID]). Add a Bar Chart (Sales by Product). Add a Card showing Sentiment Score. Add an Image visual pointing to the Photo URL column.

6

Test Cross-Filtering

Click a product in the Slicer. All three visuals should update simultaneously — the sales bar highlights, the sentiment score changes, and the photo switches. That is your 360 View.

🧠 Practice Quiz

⚡ Advanced ⏱ ~5 min

Q1. The AI Insights → Score Sentiment feature returns a number between 0 and 1. What does a score of 0.05 mean?

Q2. Which DAX function is used to flatten a parent-child org chart hierarchy into a pipe-delimited path string?

Q3. In Exercise 9, where exactly do you find the Forecasting option to add a prediction line to a Line Chart?

Q4. In the 360 View (Exercise 10), what is the common key that links Sales, Reviews, and Photos tables?

Q5. A Streaming Semantic Model in Power BI Service is best for which scenario?

0/5

🏁 Key Takeaways — Part 3

🎓

Series Complete!

You've completed all three parts of Power BI Beginner to Advanced. You now know how to get data from any source, clean it, model it, visualise it, and even apply AI to it.

Next step: Build a real dashboard from data you care about. There is no better teacher than a real project.