Vectorization

The General Vectorization Process

%% General Vectorization Process %%
graph TD;
    A["Unstructured Data 
 (Text, Image, Audio)"] -- "Input" --> B{"Vectorization Model 
 (e.g., BERT, CNN)"};
    B -- "Processes & Encodes" --> C["Numerical Vector
Representation 
 [0.12, -0.45, 0.89, ..., -0.21]"];
    C -- "Used for AI Tasks" --> D(Search, Recommendation,
Classification);

Text Vectorization Example (Contextual Embeddings)

graph TD;
    A("Input Sentence: 
 'AI is transforming 
 the world'") --> B{"Transformer Model 
 (eg BERT)"};

    %% --- Define the Left Branch (Inside a Subgraph) --- %%
    subgraph "How it works"
        B_sub1("Tokenization") --> B_sub2("Attention Mechanism") --> B_sub3("Encoder Layers");
    end

    %% --- Define the Right Branch --- %%
    C("Context-Aware 
 Vector Representation") --> D("[0.76, 0.33, -0.15, ..., 0.92]");

    %% --- Create the fork from the central model to each branch --- %%
    B --> B_sub1;
    B --> C;

Image Vectorization via a Convolutional Neural Network (CNN)

%% Image Vectorization via CNN %%
graph LR;
    A["Input Image 
 (Matrix of Pixels)"] --> B["Convolutional Layers 
 (Detect edges, 
 shapes, textures)"];
    B --> C["Pooling Layers 
 (Downsample & 
 summarize features)"];
    C --> D["Flatten Layer 
 (Converts 2D feature 
 maps to 1D)"];
    D --> E["Output Feature Vector 
 [10.4, 2.1, -5.8, ..., 7.7]"];

Semantic Similarity in a Vector Space

%% --- Diagram 1: Similarity Measurement --- %%
graph LR;
    subgraph "Similarity Measurement"
        A["Distance(King, Queen) 
 is small"];
        B["Distance(King, Man) 
 is small"];
        C["Distance(King, Woman) 
 is large"];
    end

%% --- Diagram 2: Vector Operations & Relationships --- %%

    subgraph "Vector Operations & Relationships"
        op1["vector('King') - vector('Man')"] --> op2["+vector('Woman')"] --> op3["≈ vector('Queen')"];
    end

%% --- Diagram 3: High-Dimensional Vector Space --- %%
    subgraph "High-Dimensional 
 Vector Space"
        Man["vector('Man')"]
        King["vector('King')"]
        Woman["vector('Woman')"]
        Queen["vector('Queen')"]

        Man -- "Royal Relationship" --> King;
        Man -- "Gender Relationship" --> Woman;
        Woman -- "Royal Relationship" --> Queen;
        King -- "Gender Relationship" --> Queen;
    end