* Algorithms
  - ID3 (Iterative Dichotomiser 3)
  - C4.5 (successor of ID3)
  - CART (Classification And Regression Tree)
  - CHAID (CHi-squared Automatic Interaction Detector). Performs multi-level splits when computing     classification trees.[11]
  - MARS: extends decision trees to handle numerical data better.


# Automatic Web Content Extraction by Combination of Learning and Grouping

Created: Nov 09, 2018 6:38 PM
Tags: Paper

# 1. INTRODUCTION

- The main content in a webpage is often accompanied by a lot of additional and often distracting content such ad branding banners, navigation elements, advertisements and copyright etc.
- The web pages in the World Wide Web are highly heterogeneous.
- Previous work
    - Heuristic
    - Template based Approach
        - TED

# 2. RELATED WORK

- CETR
- CETD
- VIPS

# 3. PROBLEM FORMULATION AND SOLUTION

![](Untitled-c3f13dfd-e5f1-486c-aaf6-4580e50223b5.png)

# 4. FEATURE SELECTION

$$F_x(v_i)=F'_x(v_i)\bigcup\{{\bigcup_{v_j\subseteqq Children(v_i)}F_x(v_j)}\}$$

## 4.1 Position and Area Features

- We consider the left, right, top, bottom, horizontal center and vertical center positions.

$$POS\_LEFT = 1 - |BEST\_LEFT\_LEFT|$$

## 4.2 Font Features

$$FONT\_COLOR\_POPULARITY=\sum_i\varphi_{ki} \varphi_{ri}$$

$$FONT\_SIZE=\sum_i{\rho_{ki}(z_i-z_{min}) \over (z_{max}-z_{min})}$$

## 4.3 Text, Tag and Link Features

$$TEXT\_RATIO={A_{text} \over A_{text} +A_{image} + 1}$$

$$TAG\_DENSITY={numTags \over numChars+1}$$

$$LINK\_DENSITY={numLinks \over numTags+1}$$

# 5 LEARNING

# 6 GROUPING AND REFINING

1. Grouping
2. Group Selection
3. Refining
4. EXPERIMENTAL EVALUATION
    1. Evaluation Data  Set and Metrics
    2. Comparison with the Baseline Methods
        - LR_A
        - SVM_A
        - LR
        - SVM
        - MSS
    3. Parameter Sensitivity Analysis
5. CONCLUSIONS