${{\textsc {ber}}}_{y}{\textsc {l}}$: A System for Web Block Classification

Abstract

Web blocks such as navigation menus, advertisements, headers, and footers are key components of Web pages that define not only the appearance, but also the way humans interact with different parts of the page. For machines, however, classifying and interacting with these blocks is a surprisingly hard task. Yet, Web block classification has varied applications in the fields of wrapper induction, assistance to visually impaired people, Web adaptation, Web page topic clustering, and Web search. Our system for Web block classification, ${{\textsc {ber}}}_{y}{\textsc {l}}$, performs automated classification of Web blocks through a combination of machine learning and declarative, model-driven feature extraction based on Datalog rules. ${{\textsc {ber}}}_{y}{\textsc {l}}$ uses refined feature sets for the classification of individual blocks to achieve accurate classification for all the block types we have observed so far. The high accuracy is achieved through these carefully selected features, some even tuned to the specific block type. At the same time, ${{\textsc {ber}}}_{y}{\textsc {l}}$ avoids a high cost of feature engineering through a model-driven rather than programmatic approach to extracting features. Not only does this reduce the time for feature engineering, the model-driven, declarative approach also allows for semi-automatic optimisation of the feature extraction system. We perform evaluation to validate these claims on a selected range of Web blocks.

This work was supported by the ESPRC programme grant EP/M025268/1 “VADA: Value Added Data Systems – Principles and Architecture”.