Scholarlink

Abstract <p>one of the important features of information retrieval systems is ranking. Ranking performs the function of ranking search results based on relevance to the user’s query. Methods developed in state-of-the-art research still require multiple iterations. In this paper, we proposed to use zone partitioning strategies for computing web page rank parameters in retrieval systems, which implements iterative calculation for only some randomly selected subgraphs (zone). The zone approach is based on the idea to use multiple neural networks to classify rank data in graph-based structures. The crawled web pages are fragmented into three distinct zones. The core zone is used for training graph convolutional&#xa0;network, in this zone, the labels are known. It is covered with an undiscovered zone, where classifiers label node parameters. The most interesting part is the intersection zone, which represents the set of nodes and edges that belong to more than one undiscovered zone. The experiments show that the probability of classifying the true labels in the intersection zones via aggregating the results of multiple classifiers in some cases is higher than in undiscovered zones.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Estimating Page Ranks with Inductive Capability of Graph Neural Networks and Zone Partitioning in Information Retrieval

  • 期刊论文

  • 发表日期：

    2025年4月1日

查看全文
  • Fargana Abdullayeva,
  • Suleyman Suleymanzade

摘要

Abstract

one of the important features of information retrieval systems is ranking. Ranking performs the function of ranking search results based on relevance to the user’s query. Methods developed in state-of-the-art research still require multiple iterations. In this paper, we proposed to use zone partitioning strategies for computing web page rank parameters in retrieval systems, which implements iterative calculation for only some randomly selected subgraphs (zone). The zone approach is based on the idea to use multiple neural networks to classify rank data in graph-based structures. The crawled web pages are fragmented into three distinct zones. The core zone is used for training graph convolutional network, in this zone, the labels are known. It is covered with an undiscovered zone, where classifiers label node parameters. The most interesting part is the intersection zone, which represents the set of nodes and edges that belong to more than one undiscovered zone. The experiments show that the probability of classifying the true labels in the intersection zones via aggregating the results of multiple classifiers in some cases is higher than in undiscovered zones.