Yongjie Yan

Standard matrix factorization (MF) models for collaborative filtering train a single set of latent factors, which can lead to an ‘averaging effect’ that poorly serves users with niche preferences. To address this, this work explores a ‘divide-and-conquer’ framework, the Louvain community MF (LC-MF). The framework first constructs a user–user similarity graph and applies the Louvain algorithm—chosen for its scalability and ability to identify meaningful community structures—to partition users. Following this, independent MF models are trained for each community, crucially allowing each model to learn its own set of item latent factors. This design aims to enable context-dependent item representations. We also present pragmatic strategies for handling cold-start scenarios. Our experiments on three benchmark datasets (MovieLens-1 M, Amazon-Electronics, and Yelp) suggest that LC-MF can yield performance gains over standard baselines, particularly for recall-oriented metrics on large, sparse datasets. For instance, in the Yelp dataset, we observe a notable improvement in Recall@10 compared to a strong graph neural network baseline. Further analysis reveals that our approach also enhances recommendation diversity and novelty. Our analysis suggests this performance gain may stem from the model’s ability to learn context-dependent item representations, indicating a promising direction for improving recommendation personalization.

Most of the existing recommendation systems using deep learning are based on the method of RNN (Recurrent Neural Network). However, due to some inherent defects of RNN, recommendation systems based on RNN are not only very time consuming but also unable to capture the long-range dependencies between user comments. Through the sentiment analysis of user comments, we can better capture the characteristics of user interest. Information entropy can reduce the adverse impact of noise words on the construction of user interests. Information entropy is used to analyze the user information content and filter out users with low information entropy to achieve the purpose of filtering noise data. A self-attention recommendation model based on entropy regularization is proposed to analyze the emotional polarity of the data set. Specifically, to model the mixed interactions from user comments, a multi-head self-attention network is introduced. The loss function of the model is used to realize the interpretability of recommendation systems. The experiment results show that our model outperforms the baseline methods in terms of MAP (Mean Average Precision) and NDCG (Normalized Discounted Cumulative Gain) on several datasets, and it achieves good interpretability.

The recommendation system in the online medical consultation website is a system to assist patients to find appropriate doctors. Based on the analysis of the current situation of the development of an online medical community (Haodf.com) in China, this paper puts forward recommendation suggestions of finding the right hospital and doctor to promote the rapid integration of Internet technology and traditional medical services. A new recommendation model called Probabilistic Matrix Factorization integrated with Convolutional Neural Network (PMF-CNN) is proposed in the paper. Doctors’ data in Haodf.com were used to evaluate the performance of our system. The model improves the performance of medical consultation recommendations by fusing review text and doctor information based on CNN (Convolutional Neural Network). Specifically, CNN is used to learn the feature representation of the review text and the doctors’ information. Furthermore, the extended matrix factorization model is exploited to fuse the review information feature and the initial value of the doctors’ information for recommendation. As is shown in the experimental results on Haodf.com datasets, the proposed PMF-CNN achieves better recommendation performances than the other state-of-the-art recommendation algorithms. And the recommendation system in an online medical website improves the utilization efficiency of doctors and the balance of public health resources allocation.

We investigate opinion dynamics as a stochastic process in social networks. We introduce the stubborn agent in order to determine the impact of network structure on the emergence of consensus. Depending on the fraction of undirected long-range connections, we observe fascinatingly rich dynamical behavior and transitions from disordered to ordered states. In general, we find that the stubborn agent promotes the emergence of consensus due to the so-called “group effect” that facilitates coalescence between separated network components. Agents are also behaviorally constrained Shannon information entropy in networks. However, since agents want to evolve their opinion with Brownian motion, which may in turn impede full consensus, sufficiently frequent long-range links are in such situations crucial for the network to converge into an absorbing phase. Our experimental findings indicate that, for a large range of control parameters, our model yields stable and fluctuating polarized states.