Interactive community search method and device based on graph neural network

Abstract

The invention discloses an interactive community search method and device based on a graph neural network. The interactive community search method comprises the following steps: constructing a given candidate sub-graph GS according to a query node and a mark node of a user; constructing a graph neural network model M through a given candidate sub-graph GS; carrying out convergence on the graph neural network model M to obtain a graph neural network score of each node, and updating a given candidate sub-graph according to the graph neural network scores; and selecting a final target community according to the updated given candidate sub-graph and the set community size k. According to the method, a target community is positioned through a sub-graph dynamically collected in an online network, a community member relationship problem is reconstructed into a node classification problem by using a graph neural network, and a community with the size of k is introduced to describe the target community. According to the method, the similarity and the difference between the graph nodes and the annotation nodes can be captured by flexibly combining content and structural features, a community with high accuracy and accurate size is searched in an iteration and interaction mode, and the burden of a user is reduced by utilizing sorting loss.

Description

technical field [0001] The invention belongs to the field of information technology, and in particular relates to an interactive community search method and device based on a graph neural network. Background technique [0002] Community search is an important tool for network analysis. Searching a community containing a given query node in an online social network has a wide range of applications in recommendation and team organization. Its goal is to find a densely connected subgraph containing the query nodes. The discovered communities can serve as an effective candidate set for applications such as item / friend recommendation, illegal organization discovery, etc. [0003] Although this problem has been well studied, current methods still face challenges when applied to real social networks. First, almost all of these methods assume that the data has already been scraped, and they only perform analysis on the collected data. However, we cannot cleanly separate data scra...

AI Technical Summary

Problems solved by technology

Existing community search methods focus more on structure, ignoring content information, and content cannot be ignored in the research of community search, and the size of the community is uncontrollable

In addition, most existing methods analyze on the entire graph, which is not suitable for real social network scenarios, and also restricts the application of community search to a certain extent.

Images