• KIISE Transactions on Computing Practices
• /
• v.21 no.10
• /
• pp.670-679
• /
• 2015

With a rapid development of internet, many users these days refer to various recommender sites when buying items, movies, music and more. However, there are malicious users (Sybil) who raise or lower item ratings intentionally in these recommender sites. And as a result, a recommender system (RS) may recommend incomplete or inaccurate results to normal users. We suggest a recommender algorithm to separate ratings generated by users into normal ratings and outlier ratings, and to minimize the effects of malicious users. Specifically, our algorithm first ensures a stable RS against three kinds of attack models (Random attack, Average attack, and Bandwagon attack) which are the main recent security issues in RS. To prove the performance of the method of suggestion, we conducted performance analysis on real world data that we crawled. The performance analysis demonstrated that the suggested method performs well regardless of Sybil size and type when compared to existing algorithms.

• Annual Conference on Human and Language Technology
• /
• 2011.10a
• /
• pp.37-41
• /
• 2011

본 논문에서는 최근 많은 관심을 받고 있는 Opinion Mining으로서 사용자들의 자연어 형태의 영화평 문장을 분석하여 자동으로 평점을 예측하는 시스템을 제안한다. 제안 시스템은 영화평 분석에 적합한 어휘 자질, 감정 자질, 가치 자질 및 기타 자질들을 추출하고, 10점 척도의 영화평의 평점을 10개의 범주로 가정하여, 커널모델인 다중 범주 Support Vector Machine (SVM) 모델을 이용하여 높은 성능으로 영화평의 평점을 범주 분류한다.

• Journal of Internet Computing and Services
• /
• v.18 no.3
• /
• pp.19-28
• /
• 2017

Collaborative filtering (CF) is a system that interprets the relationship between a user and a product and recommends the product to a specific user. The CF model is advantageous in that it can recommend products to users with only rating data without any additional information such as contents. However, there are many cases where a user does not give a rating even after consuming the product as well as consuming only a small portion of the total product. This means that the number of ratings observed is very small and the user rating matrix is very sparse. The sparsity of this rating data poses a problem in raising CF performance. In this paper, we concentrate on raising the performance of latent factor model (especially SVD). We propose a new model that includes product similarity information and co occurrence information in SVD. The similarity and concurrence information obtained from the rating data increased the expressiveness of the latent space in terms of latent factors. Thus, Recall increased by 16% and Precision and NDCG increased by 8% and 7%, respectively. The proposed method of the paper will show better performance than the existing method when combined with other recommender systems in the future.

• Journal of Software Engineering Society
• /
• v.27 no.1
• /
• pp.8-14
• /
• 2018

As cloud computing has emerged as a hot trend in the IT industry, various types of cloud services have emerged. In addition, cloud service broker (CSB) technology has emerged to alleviate the complexity of the process of selecting the desired service that user wants among the various cloud services. One of the key features of the CSB is to recommend the best cloud services to users. In general, CSB can use a method to evaluate a service by receiving feedback about a service from users in order to recommend a cloud service. However, since each user has different criteria for giving a rating, there is a problem that reliability of service evaluation can be low when the rating is only used. In this paper, a method is proposed to supplement evaluation of rating based service by applying machine learning based sentiment analysis to cloud service user's review. In addition, the CSB prototype is implemented based on proposed method. Further, the results of comparing the performance of various learning algorithms is proposed that can be used for sentiment analysis through experiments using actual cloud service review as learning data. The proposed service evaluation method complements the disadvantages of the existing rating-based service evaluation and can reflect the service quality in terms of user experience.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.05a
• /
• pp.459-460
• /
• 2023

기존의 추천시스템은 상품간 혹은 사용자 간의 유사도를 기반으로 작동한다. 하지만 이는 사용자가 유사한 상품 추천 속에 갇히게 되는 필터 버블의 문제와 추천시스템의 고질적인 문제인 데이터 희소성 문제를 피할 수 없게 된다. 따라서 본 연구에서는 사용자의 취향과 체형 정보를 반영하여 사용자의 평점을 예측하는 협업 필터링 기반 딥러닝 추천과 상품간 비유사성을 고려하여 사용자의 평점을 예측하는 내용 기반 추천을 혼합한 하이브리드 추천 모델을 구축하여 기존 추천시스템의 문제점을 해결하였다. 모델의 성능평가를 위해 인터넷 의류 쇼핑몰을 대상으로 유사한 이미지를 활용한 하이브리드 추천 모델과 NDCG 값을 비교하였고 유사도가 낮은 이미지를 활용한 모델이 더 우수한 성능을 보였다. 이는 다른 제품과는 달리 소비자가 의류를 구매할 경우 이미 구매한 상품과 유사한 상품보다는 유사하지 않은 상품을 구매할 가능성이 크다는 것을 보여준다.

• Journal of Korea Society of Industrial Information Systems
• /
• v.23 no.6
• /
• pp.9-19
• /
• 2018

As the recent developments in the game industry and people's interest in game streaming become more popular, non-professional gamers are also interested in games and buying them. However, it is difficult to judge which game is the most enjoyable among the games released in dozens every day. Although the game sales platform is equipped with the game recommendation function, it is not accurate because it is used as a means of increasing their sales and recommending users with a focus on their discount products or new products. For this reason, in this paper, we propose a game recommendation system based on the users ratings, which raises the recommendation satisfaction level of users and appropriately reflect their experience. In the system, we implement the rate prediction function using collaborative filtering and the game recommendation function using Naive Bayesian classifier to provide users with quick and accurate recommendations. As the result, the rate prediction algorithm achieved a throughput of 2.4 seconds and an average of 72.1 percent accuracy. For the game recommendation algorithm, we obtained 75.187 percent accuracy and were able to provide users with fast and accurate recommendations.

• Management & Information Systems Review
• /
• v.34 no.3
• /
• pp.161-177
• /
• 2015

Currently, the influence of the Internet portal sites that can make it quick and easy to contact the vast amount of information is increasing. Users can connect the Internet through a portal to obtain information, such as communication between Internet users, which can be used to meet a variety of purposes. People are exposed to a variety of information from other users in the search for a movie and get information. The impact on the reviews and ratings with the limited number of characters of the film allows users to form a relationship to the movie, decide whether you want to see the movie or find another movie. but, the user can not read the whole movie review. When user see the overall evaluation, the user can receive the correct information. This research conducted a study on the prediction of the rating by the use of review data. Information of reviews, is divided into two main areas: the"fact" and "opinion". "Fact" is to convey the dispassionate information and "Opinion" is, to represent the user's feelings. In this study, we built sentiment dictionary based on the assessment and evaluation of the online review and applied to evaluate other movies. In the comparative study with a simple emotion evaluation technique, we found the suggested algorithm got the more accurate results.

• The Korean Journal of Applied Statistics
• /
• v.35 no.2
• /
• pp.265-283
• /
• 2022

MovieLens data consists of recorded movie evaluations that was often used to measure the evaluation score in the recommendation system research field. In this paper, we provide additional information obtained by clustering user-specific genre preference information through movie evaluation data and movie genre data. Because the number of movie ratings per user is very low compared to the total number of movies, the missing rate in this data is very high. For this reason, there are limitations in applying the existing clustering methods. In this paper, we propose a convex clustering-based method using the pairwise fused penalty motivated by the analysis of MovieLens data. In particular, the proposed clustering method execute missing imputation, and at the same time uses movie evaluation and genre weights for each movie to cluster genre preference information possessed by each individual. We compute the proposed optimization using alternating direction method of multipliers algorithm. It is shown that the proposed clustering method is less sensitive to noise and outliers than the existing method through simulation and MovieLens data application.

• Journal of Intelligence and Information Systems
• /
• v.24 no.2
• /
• pp.85-109
• /
• 2018

Recommender system recommends the items expected to be purchased by a customer in the future according to his or her previous purchase behaviors. It has been served as a tool for realizing one-to-one personalization for an e-commerce service company. Traditional recommender systems, especially the recommender systems based on collaborative filtering (CF), which is the most popular recommendation algorithm in both academy and industry, are designed to generate the items list for recommendation by using 'overall rating' - a single criterion. However, it has critical limitations in understanding the customers' preferences in detail. Recently, to mitigate these limitations, some leading e-commerce companies have begun to get feedback from their customers in a form of 'multicritera ratings'. Multicriteria ratings enable the companies to understand their customers' preferences from the multidimensional viewpoints. Moreover, it is easy to handle and analyze the multidimensional ratings because they are quantitative. But, the recommendation using multicritera ratings also has limitation that it may omit detail information on a user's preference because it only considers three-to-five predetermined criteria in most cases. Under this background, this study proposes a novel hybrid recommendation system, which selectively uses the results from 'traditional CF' and 'CF using multicriteria ratings'. Our proposed system is based on the premise that some people have holistic preference scheme, whereas others have composite preference scheme. Thus, our system is designed to use traditional CF using overall rating for the users with holistic preference, and to use CF using multicriteria ratings for the users with composite preference. To validate the usefulness of the proposed system, we applied it to a real-world dataset regarding the recommendation for POI (point-of-interests). Providing personalized POI recommendation is getting more attentions as the popularity of the location-based services such as Yelp and Foursquare increases. The dataset was collected from university students via a Web-based online survey system. Using the survey system, we collected the overall ratings as well as the ratings for each criterion for 48 POIs that are located near K university in Seoul, South Korea. The criteria include 'food or taste', 'price' and 'service or mood'. As a result, we obtain 2,878 valid ratings from 112 users. Among 48 items, 38 items (80%) are used as training dataset, and the remaining 10 items (20%) are used as validation dataset. To examine the effectiveness of the proposed system (i.e. hybrid selective model), we compared its performance to the performances of two comparison models - the traditional CF and the CF with multicriteria ratings. The performances of recommender systems were evaluated by using two metrics - average MAE(mean absolute error) and precision-in-top-N. Precision-in-top-N represents the percentage of truly high overall ratings among those that the model predicted would be the N most relevant items for each user. The experimental system was developed using Microsoft Visual Basic for Applications (VBA). The experimental results showed that our proposed system (avg. MAE = 0.584) outperformed traditional CF (avg. MAE = 0.591) as well as multicriteria CF (avg. AVE = 0.608). We also found that multicriteria CF showed worse performance compared to traditional CF in our data set, which is contradictory to the results in the most previous studies. This result supports the premise of our study that people have two different types of preference schemes - holistic and composite. Besides MAE, the proposed system outperformed all the comparison models in precision-in-top-3, precision-in-top-5, and precision-in-top-7. The results from the paired samples t-test presented that our proposed system outperformed traditional CF with 10% statistical significance level, and multicriteria CF with 1% statistical significance level from the perspective of average MAE. The proposed system sheds light on how to understand and utilize user's preference schemes in recommender systems domain.

• Journal of Intelligence and Information Systems
• /
• v.29 no.3
• /
• pp.1-17
• /
• 2023

With the recent rapid development of ICT(Information and Communication Technology) and the popularization of digital devices, the size of the online market continues to grow. As a result, we live in a flood of information. Thus, customers are facing information overload problems that require a lot of time and money to select products. Therefore, a personalized recommender system has become an essential methodology to address such issues. Collaborative Filtering(CF) is the most widely used recommender system. Traditional recommender systems mainly utilize quantitative data such as rating values, resulting in poor recommendation accuracy. Quantitative data cannot fully reflect the user's preference. To solve such a problem, studies that reflect qualitative data, such as review contents, are being actively conducted these days. To quantify user review contents, text mining was used in this study. The general CF consists of the following three steps: user-item matrix generation, Top-N neighborhood group search, and Top-K recommendation list generation. In this study, we propose a recommendation algorithm that applies an extended similarity measure, which utilize quantified review contents in addition to user rating values. After calculating review similarity by applying TF-IDF, Word2Vec, and Doc2Vec techniques to review content, extended similarity is created by combining user rating similarity and quantified review contents. To verify this, we used user ratings and review data from the e-commerce site Amazon's "Health and Personal Care". The proposed recommendation model using extended similarity measure showed superior performance to the traditional recommendation model using only user rating value-based similarity measure. In addition, among the various text mining techniques, the similarity obtained using the TF-IDF technique showed the best performance when used in the neighbor group search and recommendation list generation step.