Designing the recommendation algorithm for inheritance of intangible cultural heritage using wireless knowledge graphs and collaborative filtering

With the rapid progress in science and technology, wireless sensor networks are emerging as a dynamic field of innovation by offering robust potential for real-time data collection, intelligent decision-making, and seamless integration across smart environments. The online music education system can accommodate users’ different needs and behaviors in addition to realizing a variety of interactions. Because they make it possible to integrate various wireless technologies and devices seamlessly, wireless knowledge graphs can play a significant role in musical performance by giving musicians more autonomy and versatility. However, due to a large number of resources available to music performance groups and user characteristics that cannot be fully modeled, it takes time for consumers to find the necessary data quickly and easily. This paper presents a solution to facilitate users when seeking to find music performance data through the development of recommendation algorithm for music performance based on a wireless knowledge graph (WKG) and collaborative filtering. Fascinating user behavioral signals, like preferences, context and interaction histories are captured wirelessly to this goal and these signals are aligned to a WKG which encodes the semantic relations between artistic entities such as artists, genres and venues. User-relevant entity extraction module removes semantically aligned nodes from WKG as prescribed by the user and embeds them along with the user and item features into a common latent space through multi task learning. Through the use of attention-based Graph-Convolutional Networks (GCNs) that support multi-hop aggregation, the embeddings are controlled to contain subtler semantic and relational dependencies in the graph. The integration of these features allows the interaction unit to estimate user–item interaction probabilities that in turn influence the general recommendation sequence. Optimization in the system is realized through a collaborative reduction of recommendation loss, knowledge embedding loss, and regularization costs with cross-compression methods. Analysis on standard datasets indicates that the proposed model outperforms DKN, MKR, and CKE in superior AUC liftups of 20.2% and 17.5% accuracy increases on Last.FM and 12.1% AUC growth on Book-Crossing. Moreover, the model is very robust to lack of data, and systematically outperforms in terms of Precision and Recall confirming its usability for context-aware Top-K music recommendations.