Title: Personalised recommendation method for network digital media information based on user interests

Authors: Yaoyi Xu

Addresses: Employee of the Scientific Research Department, Nanjing University of the Arts, Nanjing, Jiangsu, China

Abstract: In the digital media era, information overload hampers effective data filtering, leading to cognitive fatigue and reduced decision-making efficiency. To address this challenge, we propose a personalised recommendation method for online digital media based on user interests. The proposed approach constructs a user interest model using term-frequency statistics and co-occurrence-based tag filtering, which is dynamically updated through a forgetting factor. An enhanced list-wise recommendation model based on deep neural networks is designed to process user and content features as vectorised inputs. The model integrates positional encoding and multi-head attention mechanisms with point-wise feedforward networks and residual connections, and predicts click-through rates while optimising ranking via a Softmax output layer. Experimental results indicate that our method achieves a stable MRR between 0.516 and 0.599 (fluctuation: 0.083) and maintains Precision above 95%, effectively mitigating the cognitive burden of information overload and delivering key content more accurately to target users.

Keywords: personalised recommendation; user interest modelling; network digital media; deep neural networks; attention mechanism.

DOI: 10.1504/IJIPT.2026.152213

International Journal of Internet Protocol Technology, 2026 Vol.19 No.1, pp.34 - 43

Received: 11 Apr 2025
Accepted: 23 May 2025
Published online: 11 Mar 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article