Linear Step and Baseline Bias

In Topic 1, we calculated the raw dot product ( $w \cdot x$ ) by multiplying script genre features by audience influence weights, yielding raw scores of $+9.5$ for 'Die Hard in Space' and $-1.5$ for 'The Notebook 2'.

In that setup, audience influence on individual genres was what determined the movie's score:

\text{Raw Score} = w \cdot x

However, whether a movie becomes a hit rarely depends on audience preferences alone.

What if the entire movie industry is in a box-office slump, and theater attendance is down across the board? In a cold market, it's harder for any movie to become a hit—regardless of its script or genres. Conversely, during the holidays when theater attendance is high in general, a positive baseline makes it easier for any movie to become a hit.

That background climate affects the outcome, yet it has nothing to do with any individual script. It is an industry-wide baseline condition.

That starting offset is simply an independent baseline influence, known as the bias ( $b$ ).

The Linear Step: Adding Baseline Bias

To account for baseline conditions, every artificial neuron adds an independent scalar parameter called the bias ( $b$ ) to its accumulated dot product:

z = \sum_{i=1}^n w_i x_i + b = w \cdot x + b

The resulting value ( $z$ ) is called the linear score. In neural networks, calculating this score is known as the linear step.

Notice the two distinct components of the linear score:

Feature Evidence ( $w \cdot x$ ): The raw score contributed by the input features and weights.
Baseline Bias ( $b$ ): An independent offset that shifts the score up or down regardless of the inputs.

Depending on the environment, the baseline bias can take on three states:

Negative Bias ( $b < 0$ ): Acts as a hurdle. It lowers the linear score, demanding stronger feature evidence to cross zero.
Positive Bias ( $b > 0$ ): Acts as a head start. It raises the linear score, making it easier to cross zero.
Zero Bias ( $b = 0$ ): A neutral baseline where the outcome relies purely on the raw dot product ( $z = w \cdot x$ ).

Applying Baseline Bias to Our Movie Predictor

Now, let's put concrete numbers on our movie studio scenario.

Recall our audience influence weights ( $w$ ) across the fixed genre order [Action, Romance, Comedy, Sci-Fi]:

w = \begin{bmatrix} 5.0 \\ -2.0 \\ 1.0 \\ 4.0 \end{bmatrix} \begin{matrix} \text{(Action Weight)} \\ \text{(Romance Weight)} \\ \text{(Comedy Weight)} \\ \text{(Sci-Fi Weight)} \end{matrix}

Imagine the movie industry enters a severe box-office slump where theater attendance drops across the board. To model this harsh market climate, our studio predictor introduces a negative baseline hurdle:

b = -2.0

Let's evaluate how this market slump hurdle affects our two scripts:

🎬 'Die Hard in Space': Linear Score

Feature	Feature Value (x)	Weight (w)	Product (w_i · x_i)	Dot Product (w · x)	Industry Bias (b)	Linear Score (z = w · x + b)
Action	1.0	+5.0	+5.0	+9.5	−2.0	+7.5
Romance	0.0	−2.0	0.0
Comedy	0.5	+1.0	+0.5
Sci-Fi	1.0	+4.0	+4.0

🎬 'The Notebook 2': Linear Score

Feature	Feature Value (x)	Weight (w)	Product (w_i · x_i)	Dot Product (w · x)	Industry Bias (b)	Linear Score (z = w · x + b)
Action	0.0	+5.0	0.0	−1.5	−2.0	−3.5
Romance	1.0	−2.0	−2.0
Comedy	0.5	+1.0	+0.5
Sci-Fi	0.0	+4.0	0.0

The industry slump hurdle penalizes both films equally by $-2.0$ . 'Die Hard in Space' maintains a strong positive score of $+7.5$ , while 'The Notebook 2' is pushed deeper into negative territory at $-3.5$ .