Yen-Liang Chen

The rapid introduction of new products on e-commerce platforms poses a significant challenge for recommender systems because newly listed items lack sufficient interaction histories and therefore cannot be effectively represented by conventional ID-based recommendation models. To address this problem, this study proposes an integrated semantic intelligence framework that combines large language models (LLMs) with graph learning to enhance new product recommendation. The proposed framework adopts a two-stage training strategy. In the first stage, an LLM extracts concise semantic anchors from product metadata, such as titles and descriptions information, to generate informative initial representations for new products. A hybrid item–item graph is then constructed by integrating behavioral transition edges with selectively injected semantic edges, enabling new products to connect with semantically relevant existing items despite the absence of historical interactions. Graph contrastive pre-training is further employed to align semantic representations with collaborative structural signals and to alleviate representation collapse. In the second stage, a dual-path preference aggregation mechanism dynamically integrates graph-based structural representations and LLM-derived semantic features. Masked mean pooling is subsequently applied to aggregate users’ historical preferences and generate personalized recommendation scores. Experiments conducted on the Amazon Reviews 2023 dataset demonstrate that the proposed framework substantially reduces isolated new-product nodes, maintains competitive recommendation accuracy for existing products, and consistently outperforms representative baselines in terms of HR, NDCG, and MRR for new-product recommendation. These findings indicate that integrating LLM-based semantic intelligence with graph learning provides an effective solution for bridging semantic product understanding and collaborative user behavior, thereby improving the visibility and recommendation quality of newly introduced products on e-commerce platforms.

The rapid introduction of new products on e-commerce platforms poses a significant challenge for recommender systems because newly listed items lack sufficient interaction histories and therefore cannot be effectively represented by conventional ID-based recommendation models. To address this problem, this study proposes an integrated semantic intelligence framework that combines large language models (LLMs) with graph learning to enhance new product recommendation. The proposed framework adopts a two-stage training strategy. In the first stage, an LLM extracts concise semantic anchors from product metadata, such as titles and descriptions information, to generate informative initial representations for new products. A hybrid item–item graph is then constructed by integrating behavioral transition edges with selectively injected semantic edges, enabling new products to connect with semantically relevant existing items despite the absence of historical interactions. Graph contrastive pre-training is further employed to align semantic representations with collaborative structural signals and to alleviate representation collapse. In the second stage, a dual-path preference aggregation mechanism dynamically integrates graph-based structural representations and LLM-derived semantic features. Masked mean pooling is subsequently applied to aggregate users’ historical preferences and generate personalized recommendation scores. Experiments conducted on the Amazon Reviews 2023 dataset demonstrate that the proposed framework substantially reduces isolated new-product nodes, maintains competitive recommendation accuracy for existing products, and consistently outperforms representative baselines in terms of HR, NDCG, and MRR for new-product recommendation. These findings indicate that integrating LLM-based semantic intelligence with graph learning provides an effective solution for bridging semantic product understanding and collaborative user behavior, thereby improving the visibility and recommendation quality of newly introduced products on e-commerce platforms.

Owing to their widespread practicability, recommendation systems play an important role in our daily life. Recently, several studies have examined recommendations based on learning from user purchase history using association rules to extract the complementary and substitution relationships between items. However, the integration of generalized rules into a customized recommendation model is a challenging task. This study proposes a novel recommendation model, A3BRec, which incorporates association rules with a transformer network to predict potentially interesting items for the next basket. We introduce a ternary-stage framework to integrate basket associations into sequential next-basket recommendations. Furthermore, extensive experiments were conducted on real-world datasets to demonstrate the performance and superiority of the proposed model over the state-of-the-art methods for various evaluation metrics. We also use a case study to show the improvement and influence of the proposed ternary integration in A3BRec on recommendation quality.

Recently, due to the surge in e-commerce, growing attention has been paid to how to recommend a customer’s next purchase based on sequential or session-based data. However, most prior studies have generally focused on what items may be interesting for users, but have neglected the consideration of when the next items are likely to be purchased. Clearly, the timing information is an essential factor for companies to adopt proper selling strategies at the “right” time. In this study, a novel recommendation system, G-TransRec, is proposed to predict customers’ next items of interest with the potential purchase time by exploiting a user temporal interaction sequence. Moreover, by integrating the graph embedding technique, we include the global user information to explore more collaborative knowledge for effective recommendations. Several experiments were conducted on two real datasets to demonstrate the performance and superiority of the proposed model compared with the state-of-the-art methods on several evaluation metrics. We also use a case study to show the practicability of the proposed G-TransRec for users to recommend what they want at what time from a massive amount of merchandise.

Link prediction is to discover the missing connection between any pair of individuals, which could help researcher get the whole picture of social networks and Link prediction is to discover the missing connection between any pair of individuals, which could help researcher get the whole picture of social networks and

Collaborative filtering (CF) technology has been widely used in recommendation systems. Usually, the latent factor model (LFM) is used as the basis for implementing CF recommendation in deep learning systems. This study differs from previous studies in two respects. First, for different target items, the user embedding vector should be dynamically adjusted according to the content of the target item. Therefore, we have added an attention mechanism to dynamically adjust the user’s embedding vector. However, people’s preferences usually change over time. Therefore, based on the above attention model, this study considers two time-decay functions to emphasize the user’s recent preferences. The first decay function considers the situation where the recent rating is more important than the long ago rating. The second time-decay function considers the situation, whereby users generally prefer movies that have been released recently rather than movies that have been released a long time ago. By combining these two time-decay functions with the attention model, we propose a time-decay adaptive latent factor model (TDADLFM) model for item score prediction. This study applies this model to a dataset integrating Movielens-10M and HetRec2011 and proves that all three new considerations can improve recommendation performance.

In e-commerce, session-based personalized recommendation is still a challenging task due to the limited information of users in a session. Therefore, it is not enough to include only the user's local information. Instead, we should consider global information, i.e., extract information from all users' sessions to explore more collaborative information. However, using all session information will waste computing resources, and there will be a lot of information in the global information that is not relevant to the current user, which will damage the correctness of the prediction. Therefore, instead of referencing all global information, we select only those sessions that are useful to users, the so-called Personalized Next Session (PNS), as global information to improve the lack of local information. This work is the first to adopt a deep network architecture study incorporating the concept of PNS to recommend the next item of the current session to the user. We have conducted experiments on several real-world datasets. Extensive experimental results show that our proposed deep network architecture outperforms state-of-the-art recommendation methods.

In the era of rapid development of social networks, user reviews of restaurant review websites have grown rapidly. In order to allow users to quickly grasp the key points of review information on review sites, this paper provides an abstractive multi-text summary method that can automatically generate template-based review summaries based on predefined topics and sentiments. In particular, for each predefined topic and each type of sentiment (positive or negative), this study uses the TextRank algorithm to find the most representative sentences to form a summary. This method allows users to quickly grasp the positive and negative opinions of each important aspect of the restaurant. The previous research on generating abstracts from reviews either did not generate abstracts based on topics, or they were based on topics generated by random models. However, the latter method cannot guarantee that the topics generated by the random model are really the topics that the user needs. For a restaurant review, some topics are indispensable. In order to ensure that abstracts can be generated for these essential topics, our method predefines the topics that must be generated, and then generates abstracts for these topics. In the evaluation, this study compared the template method with the Refresh and Gensim systems based on criteria such as informativeness, clarity, usefulness and likes. The results show that the method proposed in this paper is superior to the other two summary methods.