Intuitive Understanding of Dot Product & Matrix Multiplication

We use a simple house investment example to explain what a dot product means, why matrix multiplication is row × column, and what the result represents.

1. Problem

Many people struggle with three questions:

Why is matrix multiplication row × column?
What does a dot product really mean?
What does the result represent?

We will explain everything using a house investment example.

2. Represent Houses as Vectors

Suppose we have 3 houses, and each house has 4 features: location, age, area, years occupied.

X = 851203 6159010 921501 (3 houses × 4 features)

House	Location	Age	Area	Years Occupied
House 1	8	5	120	3
House 2	6	15	90	10
House 3	9	2	150	1

        Meaning:

        Each row = one house

        Each column = one feature

3. Introduce a "Perspective"

Different people evaluate houses differently. For example, an investor may care more about location and area, and dislike an old house.

w_invest = 0.8 −0.5 1.2 −0.1 (4 features × 1 perspective)

This means:

location → positive
age → negative
area → positive
years occupied → slightly negative

This vector is a scoring rule.

4. Dot Product (Core Idea)

Now compute House 1 · w_invest:

851203 · 0.8 −0.5 1.2 −0.1 = 8×0.8 + 5×(−0.5) + 120×1.2 + 3×(−0.1) = 147.6

        Meaning of dot product:

        Use one perspective to score one object.

So here, House · Investor perspective means: How valuable is this house from an investor's view?

5. Multiple Perspectives → Matrix W

In real life, we may have multiple ways to evaluate the same house, such as investment value and living comfort.

W = 0.80.6 −0.5−0.2 1.20.9 −0.10.3 (4 features × 2 perspectives)

Each column is one perspective:

Column 1 → investment value
Column 2 → living comfort

6. Matrix Multiplication: X @ W

Now we compute:

Y = 851203 6159010 921501 @ 0.80.6 −0.5−0.2 1.20.9 −0.10.3 = 147.6112.7 104.384.6 186.1140.3 (3 houses × 2 scores)

House	Investment Value	Living Comfort
House 1	147.6	112.7
House 2	104.3	84.6
House 3	186.1	140.3

7. Key Insight

Matrix multiplication = a batch of dot products.

Y[i][j] = x_i1x_i2···x_in · w_1j w_2j ⋮ w_nj = house_i · perspective_j

So:

row → house
column → perspective
value → score

8. Why "Row × Column"?

Because we are doing this: a house (row) vs a perspective (column).

Each number in the result is one dot product between one row from X and one column from W:

851203 6159010 921501 @ 0.80.6 −0.5−0.2 1.20.9 −0.10.3 → Y[2][2] = 84.6

        So row × column is not arbitrary. It naturally matches:

        object × evaluation rule.

9. Unified View

Embedding (X) → what you are
Weights (W) → how I evaluate you
Dot product → how valuable you are under this view

objects 851203 6159010 921501 @ 0.80.6 −0.5−0.2 1.20.9 −0.10.3 perspectives = 147.6112.7 104.384.6 186.1140.3 scores

Matrix multiplication = evaluate many objects under many perspectives.