Biased Graph Network Improves Sequential Recommendations Using Adaptive Missing Information Imputation

Researchers have developed a sequential recommendation model designed to handle incomplete auxiliary information by adapting to sequence sparsity. Auxiliary data often improves sequential recommendation accuracy, but real-world implementations frequently encounter incomplete records. Common modeling strategies often overlook the sparsity of missing attributes and apply direct imputation across missing fields, an approach that can degrade overall recommendation performance.

To address this challenge, Xin Yang, Zuohan Huang, and Yanxin Gao introduced a model designated as BGCN-SMIRec, or Biased Graph Convolutional Network Based on Self-Adaptive Missing Information for Sequential Recommendation. The framework analyzes the missing rate of an input sequence and partitions the incoming data into two distinct subsequences based on that assessment.

The two subsequences pass through distinct processing channels within the architecture. One channel utilizes a Transformer-based imputation strategy to reconstruct missing fields, while the parallel channel implements partial convolution specifically configured to diminish the adverse effects of absent data.

Beyond subsequence imputation, the architecture incorporates Biased Graph Convolutional Networks to capture both temporal dependencies and spatial structural features simultaneously. In experimental evaluations conducted across three real-world datasets alongside their incomplete variations, the researchers reported that the proposed BGCN-SMIRec framework outperformed existing sequential recommendation baselines, demonstrating robustness against missing auxiliary attributes.

Subject of Research: Sequential recommendation models utilizing graph convolutional networks and adaptive missing information processing

Article Title: A Biased graph convolutional network based on self-adaptive missing information for sequential recommendation

Article References: Yang, X., Huang, Z., & Gao, Y. (2026). A Biased graph convolutional network based on self-adaptive missing information for sequential recommendation. Cluster Computing, 29(16), Article 940. https://doi.org/10.1007/s10586-026-06776-5

Image Credits: AI Generated

DOI: 10.1007/s10586-026-06776-5

Keywords: sequential recommendation, graph convolutional networks, missing data imputation, Transformer imputation, partial convolution, temporal dependencies, spatial structural features, auxiliary information, data sparsity, recommender systems, deep learning, machine learning, BGCN-SMIRec, information retrieval, neural networks, feature representation