BeLightRec: A lightweight recommender system enhanced with BERT
Manh Mai Van, Tin T. Tran
TL;DR
BeLightRec addresses the gap where collaborative filtering misses item-level semantics by integrating a semantic similarity signal derived from item titles and descriptions via $BERT$ with a lightweight graph convolutional network. The model ingests an interaction matrix $R$ and a semantic similarity matrix $B$, both $L1$-normalized, and propagates signals over $K$ hops to yield embeddings for users and items, enabling a final score $\\widehat{y}_{ui}= e_u^{*^ op} e_i^{*}$. Training uses Bayesian Personalized Ranking to optimize ranking over observed vs unobserved interactions. Experiments on three real-world datasets show consistent gains over baselines, demonstrating the value of fusing CF with semantic text signals for improved recommendation quality.
Abstract
The trend of data mining using deep learning models on graph neural networks has proven effective in identifying object features through signal encoders and decoders, particularly in recommendation systems utilizing collaborative filtering methods. Collaborative filtering exploits similarities between users and items from historical data. However, it overlooks distinctive information, such as item names and descriptions. The semantic data of items should be further mined using models in the natural language processing field. Thus, items can be compared using text classification, similarity assessments, or identifying analogous sentence pairs. This research proposes combining two sources of item similarity signals: one from collaborative filtering and one from the semantic similarity measure between item names and descriptions. These signals are integrated into a graph convolutional neural network to optimize model weights, thereby providing accurate recommendations. Experiments are also designed to evaluate the contribution of each signal group to the recommendation results.