Relationship Graph Filtering Drawing on the Example of the Site Hierarchy Extraction - Boris Belyaev

Yandex School of Data Analysis Conference
Machine Learning: Prospects and Applications

https://yandexdataschool.com/conference

Relationship graph filtering is a well-known problem. Basic data of the
problem is usually presented as a graph with great number of garbage
edges. Graph filtering and clear relationships construction sets a variety
of algorithmic problems. A solution to the problem drawing on the example
of site hierarchy extraction will be presented in this talk. Trying to get
a site’s hierarchy, we face some technical challenges: different types of
subdomains, the site’s menu, metalinks, filters’ combinatorial explosion,
different types of sortings, the hierarchy’s uncertainty. Some of the
challenges are closely connected with the site’s topic, but most of them
are widely distributed among different topics and types of sites. How to
use different techniques of graph filtering to solve these problems will be
shown in the presentation. Different methods of URL templates, hierarchy
corrections, and statistical analysis of elements will be examined.