Intelligent Decision-Making for Advertising Recommendation via Data Mining and Graph Neural Networks

Abstract

This paper addresses the challenges of data sparsity, complex interactions, difficulty in fully characterizing latent preferences, and lack of structural awareness in the decision-making process of advertising recommendation tasks. It proposes an intelligent decision-making method for advertising recommendation that combines hard data mining and graph neural networks. First, this method extracts multi-source information, including user, ad, and contextual information, from the original ad interaction logs. Then, it uses a hard data mining mechanism to identify high-value behavioral patterns and key relationship clues to reduce noise interference and enhance the ability to express effective information. Based on this, a heterogeneous interaction graph for advertising recommendation scenarios is constructed, unifying user interests, ad attributes, and environmental information into a graph structure modeling framework. Graph propagation learning is used to mine high-order dependencies between nodes, thereby more accurately representing potential associations in complex recommendation scenarios. Simultaneously, a graph reweighting and adaptive fusion mechanism is introduced to strengthen key structural information and improve the synergy between initial features and deep graph representations, making the resulting decision representation more consistent with the actual needs of advertising recommendation tasks. Furthermore, based on a unified decision modeling strategy, the matching relationship between users and ads is effectively estimated, thereby completing ad ranking and recommendation output. Comparative experimental results show that the proposed method achieves superior performance on multiple recommendation evaluation indicators, verifying its effectiveness and practicality in intelligent decision-making tasks for advertising recommendation.