Digital Library[ Search Result ]
A Large Language Model-based Multi-domain Recommender System using Model Merging
http://doi.org/10.5626/JOK.2025.52.6.548
Recent research in recommender systems has increasingly focused on leveraging pre-trained large language models (LLMs) to effectively understand the natural language information associated with recommendation items. While these LLM-based recommender systems achieve high accuracy, they have a limitation in that they require training separate recommendation models for each domain. This increases the costs of storing and inferring multiple models and makes it difficult to share knowledge across domains. To address this issue, we propose an LLM-based recommendation model that effectively operates across diverse recommendation domains by applying task vector-based model merging. During the merging process, knowledge distillation is utilized from individually trained domain-specific recommendation models to learn optimal merging weights. Experimental results show that our proposed method improves recommendation accuracy by an average of 2.75% across eight domains compared to recommender models utilizing existing model merging methods, while also demonstrating strong generalization performance in previously unseen domains.
An Inference Framework for Text-Based Sequential Recommendation Model Using Nearest Neighbor Mechanism
Junyoung Kim, Hyunsoo, Jongwuk Lee
http://doi.org/10.5626/JOK.2025.52.5.435
Sequential recommendation task aims to predict the next item to interact with based on users’ interaction history. Text-based recommendation models, which represent items as text, show improved performance in cold-start problems and zero-shot recommendation tasks. However, they suffer from textual bias and the lack of collaborative knowledge. To overcome these limitations, we propose a text-based recommendation model inference framework using the nearest neighbor mechanism. The proposed method leverages text-based recommendation models as a neighbor retriever model to search neighbors with similar preferences to the user and aggregate the neighbor information with existing recommendation results to improve recommendation performance. Experiments conducted on four datasets show that the proposed method consistently outperforms existing models, with performance improvement up to 25.27% on NDCG@50. Furthermore, the proposed method effectively complements collaborative knowledge and improves model explainability by providing recommendation rationale.
Geographical Adaptive Attention Model for Points of Interest Recommendation
Muyeon Jo, Sejin Chun, Jungkyu Han
http://doi.org/10.5626/JOK.2025.52.3.217
Geographical influence, stemming from the location of Points of Interest (POIs), plays a vital role in POI recommendation. Most current studies utilize geographical information such as distance and location to define and extract POI-specific geographical influences for personalized recommendations. These approaches primarily emphasize distance-based influence, which gauges user preferences based on proximity, while often overlooking area-based influence, which reflects preferences for regions with specific POI characteristics. This paper introduces a POI recommendation model based on an attention network that integrates both distance- and area-based influences. The model adaptively assesses how previously visited POIs impact the likelihood of visiting a target POI, taking into account regional characteristics and user preferences. Experiments conducted on real-world datasets indicate that the proposed method significantly outperforms baseline models, achieving improvements of approximately 6–12% in Prec@10, 8–10% in Recall@10, and 6–7% in HR@10.
SASRec vs. BERT4Rec: Performance Analysis of Transformer-based Sequential Recommendation Models
Hye-young Kim, Mincheol Yoon, Jongwuk Lee
http://doi.org/10.5626/JOK.2024.51.4.352
Sequential recommender systems extract interests from user logs and use them to recommend items the user might like next. SASRec and BERT4Rec are widely used as representative sequential recommendation models. Existing studies have utilized these two models as baselines in various studies, but their performance is not consistent due to differences in experimental environments. This research compares and analyzes the performance of SASRec and BERT4Rec on six representative sequential recommendation datasets. The experimental result shows that the number of user-item interactions has the largest impact on BERT4Rec training, which in turn leads to the performance difference between the two models. Furthermore, this research finds that the two learning methods, which are widely utilized in sequential recommendation settings, can also have different effects depending on the popularity bias and sequence length. This shows that considering dataset characteristics is essential for improving recommendation performance.
Optimizing Computation of Tensor-Train Decomposed Embedding Layer
Seungmin Yu, Hayun Lee, Dongkun Shin
http://doi.org/10.5626/JOK.2023.50.9.729
Personalized recommendation system is ubiquitous in daily life. However, the huge amount of memory requirement to store the embedding tables used by deep learning-based recommendation system models is taking up most of the resources of industrial AI data centers. To overcome this problem, one of the solutions is to use Tensor-Train (TT) decomposition, is promising compression technique in deep neural network. In this study, we analyze unnecessary computations in Tensor-Train Gather and Reduce (TT-GnR) which is the operation of embedding layer applied with TT decomposition. To solve this problem, we define a computational unit called group to bind the item vectors into a group and propose Group Reduced TT-Gather and Reduce operation to reduce unnecessary operations by calculating with groups. Since the GRT-GnR operation is calculated in groups, computational cost varies depending on how item vectors are grouped. Experimental results showed that the GRT-GnR operation had a 41% decrease in latency compared to conventional TT-GnR operation.
Graph Embedding-Based Point-Of-Interest Recommendation Considering Weather Features
Kun Woo Lee, Jongseon Kim, Yon Dohn Chung
http://doi.org/10.5626/JOK.2022.49.3.221
As the Location-Based Services (LBS) grow rapidly, the Point-Of-Interest (POI) recommendation becomes an active research area to provide users appropriate information relevant to their locations. Recently, translation-based recommendation systems using graph embedding, such as TransRec, are attracting great attention. In this paper, we discovered some drawbacks of TransRec; it is limited in expressing the complex relationship between users and POIs, and the relation embedding is fixed without considering weather features. We propose WAPTRec, a graph embedding-based POI recommendation method considering the weather, that overcomes the drawback of TransRec. WAPTRec can rep resent the same POI embedding in different ways according to users by using a category projection matrix and attention mechanism. In addition, it provides better recommendation accuracy by utilizing the users’ movement history, category of POIs and weather features. Experiments using public datasets illustrated that WAPTRec outperformed the conventional translation-based recommendation methods.
Improving Performance of Recurrent Neural Network based Recommendations by Utilizing Personal Preferences
Dong Shin Lim, Yong Jun Yang, Shin Cho
http://doi.org/10.5626/JOK.2021.48.11.1211
As the amount of content provided on the platform surged, a recommendation system became an essential element of the platform. The collaborative filtering technique is a widely used recommendation system in academia and industry, but it also has a limitation because it relies on quantitative information from consumers such as ratings and purchase history. To overcome this shortcoming, various studies have been done in a bid to improve its performance by collecting qualitative information such as review data in a model. Recently, some studies that applied recurrent neural networks showed better performance than the existing recommendation system by using time-series behavioral data only, but studies that reflect each customer"s preference in the recommendation model have not yet been made. In this paper, an improved recommendation model was presented by calculating a preference matrix based on customer log data and learning it in a recurrent neural network through an embedding vector. It was confirmed that the prediction performance was improved compared to the existing recurrent neural network recommendation model.
A Weight-based Multi-domain Recommendation System for Alleviating the Cold-Start Problem
http://doi.org/10.5626/JOK.2021.48.10.1090
A recommendation system predicts users’ ratings based on users’ past behaviors and item preferences. One of the most famous types of recommendation systems is the collaborative filtering method that predicts users’ ratings based on the rating information from users with similar preferences. In order to predict the preferences of users, we need adequate information about users’ interactive information on items. Otherwise, it is very difficult to make accurate predictions for users without adequate information. This phenomenon is called the cold-start problem. In this paper, we propose a multi-domain recommendation system that utilizes the rating information of multiple domains. We propose a method that calculates the weight of each auxiliary domain to maximize the confidence of predicted ratings from multiple auxiliary domains and verify the performance of the proposed method through extensive experiments. As a result, we demonstrate that our algorithm produces better recommendation results compared to the classical algorithms simply utilized in multiple domain settings.
Performance Comparison and Analysis Between Neural and Non-neural Autoencoder-based Recommender Systems
http://doi.org/10.5626/JOK.2020.47.11.1078
While deep neural networks have been bringing advances in many domains, recent studies have shown that the performance gain from deep neural networks is not as extensive as reported, compared to the higher computational complexity they require. This phenomenon is caused by the lack of shared experimental settings and strict analysis of proposed methods. In this paper, 1) we build experimental settings for fair comparison between the different recommenders, 2) provide empirical studies on the performance of the autoencoder-based recommender, which is one of the main families in the literature, and 3) analyze the performance of a model according to user and item popularity. With extensive experiments, we found that there was no consistent improvement between the neural and the non-neural models in every dataset and there is no evidence that the non-neural models have been improving over time. Also, the non-neural models achieved better performance on popular item accuracy, while the neural models relatively perform better on less popular items.
Data Modelling Method for Real-Time Advertising Service Based on Viewer Reaction and Intention in Online Broadcasting
Seongju Kang, Chaeeun Jeong, Kwangsue Chung
http://doi.org/10.5626/JOK.2020.47.11.1086
The interaction between the existing advertising service and the user is limited. To provide a personalized advertising service, advertisement systems should predict the user"s preference based on the user"s profile and the user-content relationship. Many recommendation schemes have been studied to predict the preferences of users. However, the existing recommendation system is difficult to guarantee real-time preference prediction as it performs a calculation of the matrix with high computational complexity. In this paper, we propose a data modeling method for real-time advertising services based on the reaction and intention of viewers. To predict the user"s preference in real-time, the user"s historical data is modeled in a tree structure. The tree structure allows us to retrieve and compare the data with logarithmic time complexity. To improve the accuracy of the recommendation, we have proposed a recommendation algorithm that considers both the user"s positive and negative evaluations. Finally, we have evaluated the performance of the proposed method through various methods.
Search
Journal of KIISE
- ISSN : 2383-630X(Print)
- ISSN : 2383-6296(Electronic)
- KCI Accredited Journal
Editorial Office
- Tel. +82-2-588-9240
- Fax. +82-2-521-1352
- E-mail. chwoo@kiise.or.kr