Sue-Chen Hsueh

The advancement of explainable recommendations aims to improve the quality of textual explanations for recommendations. Traditional methods primarily used Recurrent Neural Networks (RNNs) or their variants to generate personalized explanations. However, recent research has focused on leveraging Transformer architectures to enhance explanations by extracting user reviews and incorporating features from interacted items. Nevertheless, previous studies have failed to fully exploit the relationship between reviews and user ratings to generate more personalized explanations. In this paper, we propose a novel model named EPER (Enhanced Personalization for Explainable Recommendation), which considers reviews, user ratings, feature words, and item titles to generate high-quality personalized explanations. The EPER model employs a masking mechanism to prevent interference between rating prediction and explanation generation. Moreover, we propose an innovative feature-handling method to manage missing interaction features in existing models. Experimental results on public datasets demonstrate that EPER generally outperforms other well-known methods, including NETE, PETER+, and MMCT. Compared with MMCT, EPER improves explanation quality (ROUGE metric) by 3.27%, personalization (FMR metric) by 6.82%, and rating prediction (MSE metric) by 1.2% for the Amazon Clothing dataset. Overall, the EPER model provides personalized recommendation explanations that match or exceed the best existing methods, demonstrating its potential for practical applications.

Text classification is an important research field in text mining and natural language processing, gaining momentum with the growth of social networks. Despite the accuracy advancements made by deep learning models, existing graph neural network-based methods often overlook the implicit class information within texts. To address this gap, we propose a graph neural network model named LaGCN to improve classification accuracy. LaGCN utilizes the latent class information in texts, treating it as explicit class labels. It refines the graph convolution process by adding label-aware nodes to capture document–word, word–word, and word–class correlations for text classification. Comparing LaGCN with leading-edge models like HDGCN and BERT, our experiments on Ohsumed, Movie Review, 20 Newsgroups, and R8 datasets demonstrate its superiority. LaGCN outperformed existing methods, showing average accuracy improvements of 19.47%, 10%, 4.67%, and 0.4%, respectively. This advancement underscores the importance of integrating class information into graph neural networks, setting a new benchmark for text classification tasks.

This study introduces session-aware recommendation models, leveraging GRU (Gated Recurrent Unit) and attention mechanisms for advanced latent interaction data integration. A primary advancement is enhancing latent context, a critical factor for boosting recommendation accuracy. We address the existing models’ rigidity by dynamically blending short-term (most recent) and long-term (historical) preferences, moving beyond static period definitions. Our approaches, pre-combination (LCII-Pre) and post-combination (LCII-Post), with fixed (Fix) and flexible learning (LP) weight configurations, are thoroughly evaluated. We conducted extensive experiments to assess our models’ performance on public datasets such as Amazon and MovieLens 1M. Notably, on the MovieLens 1M dataset, LCII-Pre_Fix achieved a 1.85% and 2.54% higher Recall@20 than II-RNN and BERT4Rec_+st+TSA, respectively. On the Steam dataset, LCII-Post_LP outperformed these models by 18.66% and 5.5%. Furthermore, on the Amazon dataset, LCII showed a 2.59% and 1.89% improvement in Recall@20 over II-RNN and CAII. These results affirm the significant enhancement our models bring to session-aware recommendation systems, showcasing their potential for both academic and practical applications in the field.

This study introduces session-aware recommendation models, leveraging GRU (Gated Recurrent Unit) and attention mechanisms for advanced latent interaction data integration. A primary ad-vancement is enhancing latent context, a critical factor for boosting recommendation accuracy. We address existing models' rigidity by dynamically blending short-term (most recent) and long-term (historical) preferences, moving beyond static period definitions. Our approaches, pre-combination (LCII-Pre) and post-combination (LCII-Post) with fixed (Fix) and flexible learning (LP) weight configurations, are thoroughly evaluated. We conducted extensive experiments to as-sess our models' performance on public datasets such as Amazon and MovieLens 1M. Notably, on the MovieLens 1M dataset, LCII-PreFix achieved a 1.85% and 2.54% higher Recall@20 than II-RNN and BERT4Rec+st+TSA, respectively. On the Steam dataset, LCII-PostLP outperformed these models by 18.66% and 5.5%. Furthermore, on the Amazon dataset, LCII showed a 2.59% and 1.89% improvement in Recall@20 over II-RNN and CAII. These results affirm the significant enhancement our models bring to session-aware recommendation systems, showcasing their po-tential for both academic and practical applications in the field.

Many modern applications such as sensor networks produce probabilistic data. These data are collected into an uncertain database. To interpret uncertainty and to mine frequent patterns in an uncertain database, all possible certain databases are considered, which generates an exponential number of combinations and makes the mining problem highly complicated. In practice, mining is interactive, which makes the discovery of frequent itemsets in an uncertain database even more challenging. The objective of interactive mining is to shorten the time that is required to obtain the desired patterns in the iterated lengthy mining process. The time-consuming mining process in an uncertain database is exacerbated by repeated processing if the mining is performed from scratch. Therefore, we propose an interactive mining algorithm called iDIP to solve this problem. The iDIP algorithm adopts an approximation mechanism to mine the patterns and prunes candidates by using the existing patterns. Comprehensive experiments using both real and synthetic datasets show that iDIP outperforms the well-known re-mining-based MB algorithm for 4.3 times faster on average. In addition, iDIP has good linear scalability.