• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.23 no.5
• /
• pp.495-507
• /
• 2019

Data sparsity is one of the main challenges for the recommender system. The recommender system contains massive data in which only a small part is the observed data and the others are missing data. Most studies assume that missing data is randomly missing from the dataset. Therefore, they only use observed data to train recommendation model, then recommend items to users. In actual case, however, missing data do not lost randomly. In our research, treat these missing data as negative examples of users' interest. Three sample methods are seamlessly integrated into SVD++ algorithm and then propose SVD++_W, SVD++_R and SVD++_KNN algorithm. Experimental results show that proposed sample methods effectively improve the precision in Top-N recommendation over the baseline algorithms. Among the three improved algorithms, SVD++_KNN has the best performance, which shows that the KNN sample method is a more effective way to extract the negative examples of the users' interest.

• Journal of KIISE
• /
• v.42 no.11
• /
• pp.1380-1385
• /
• 2015

Collaborative filtering is one of the most successfully used methods for recommender systems and has been utilized in various areas such as books and music. The key point of this method is selecting the most proper recommenders, for which various similarity measures have been studied. To improve recommendation performance, this study analyzes problems of existing recommender selection methods based on similarity and presents a method of dynamically determining recommenders based on the rate of co-rated items as well as similarity. Examination of performance with varying thresholds through experiments revealed that the proposed method yielded greatly improved results in both prediction and recommendation qualities, and that in particular, this method showed performance improvements with only a few recommenders satisfying the given thresholds.

• Smart Media Journal
• /
• v.12 no.9
• /
• pp.89-94
• /
• 2023

Recommender systems provide users with the most favorable items by analyzing explicit or implicit feedback of users on items. Recently, as the size of deep-learning-based models employed in recommender systems has increased, many studies have focused on reducing inference time while maintaining high recommendation accuracy. As one of them, a study on recommender systems with a knowledge distillation (KD) technique is actively conducted. By KD, a small-sized model (i.e., student) is trained through knowledge extracted from a large-sized model (i.e., teacher), and then the trained student is used as a recommendation model. Existing studies on KD for recommender systems have been mainly performed only for implicit feedback settings. Thus, in this paper, we try to investigate the performance and accuracy when applied to explicit feedback settings. To this end, we leveraged a total of five state-of-the-art KD methods and three real-world datasets for recommender systems.

• Smart Media Journal
• /
• v.12 no.6
• /
• pp.35-40
• /
• 2023

In this study paper, we examine pre-trained language models used in Multi-Goal Conversational Recommender Systems (MG-CRS), comparing and analyzing their performances of various pre-trained language models. Specifically, we investigates the impact of the sizes of language models on the performance of MG-CRS. The study targets three types of language models - of BERT, GPT2, and BART, and measures and compares their accuracy in two tasks of 'type prediction' and 'topic prediction' on the MG-CRS dataset, DuRecDial 2.0. Experimental results show that all models demonstrated excellent performance in the type prediction task, but there were notable provide significant performance differences in performance depending on among the models or based on their sizes in the topic prediction task. Based on these findings, the study provides directions for improving the performance of MG-CRS.

• Communications of Mathematical Education
• /
• v.26 no.3
• /
• pp.251-271
• /
• 2012

The purpose of this study is: 1) to determine error sources and the effects of each error source, 2) to investigate optimal measuring conditions from holistic and analytic scoring methods, and 3) to compare the value of reliability between Cronbach's alpha and the generalizability coefficient in self-introduction letter and teacher's recommendation letter based on the generalizability theory in identification of mathematical gifted students by observations and nominations. Data of this study were collected from the science education institute for the gifted attached to the university located within in a capital city for the 2011 academic year. Scores form two raters using holistic and analytic scoring methods in both assessment types were used. The results of this study were as follows. First, as to both assessment types, error sources for people were relatively large regardless of scoring methods. However, error sources for raters in holistic scoring methods had a more significant impact than those of analytic scoring methods. Second, to set optimal measuring conditions in the self-introduction letter and teacher's recommendation letter, if we fixed the number of raters into 2 based on holistic scoring methods, at least 5 and 10 content domains were needed, respectively. In addition, the number of items in teacher's recommendation letter should be more than 3 when we fixed the number of content domains into 4, and the number of items in self-introduction letter should be more than 8 when we fixed the number of content domains into 6 using analytic scoring methods. Third, Cronbach's alpha having only a single source of errors was higher than the generalizability coefficient regardless of assessment types and scoring methods. Hence we recommend that generalizability coefficient based on various error sources such as raters, content domains, and items should be considered to keep a satisfactory level of reliability in both assessment types.

• 대한지구과학교육학회:학술대회논문집
• /
• 2010.08a
• /
• pp.19-20
• /
• 2010

• 한국정보교육학회:학술대회논문집
• /
• 2010.01a
• /
• pp.313-318
• /
• 2010

한 국가의 미래는 인적 자원의 개발에 달려 있으며 정보영재는 중요한 인적자원이다. 영재교육진흥법 시행에 따라 영재 교육 체제의 정립을 위한 노력이 계속되고 있음에도 불구하고 영역별 특성에 맞는 정확한 판별 기준이 정립되어 있지 않다는 한계점이 제기되고 있다. 따라서 본 고에서는 이 한계를 극복하고자 초등정보영재 교육 대상자 선발을 위한 추천서를 31개 항목으로 나누어 분석하여 과학영재원 합격과의 상관관계를 통해 관찰 추천의 타당도에 대한 지표를 제공하고자 한다.

• Journal of Internet Computing and Services
• /
• v.11 no.6
• /
• pp.1-11
• /
• 2010

While a supply of GPS-enabled smartphone is increased, LBS which is studied and developed for special function is changed to personal solution. In this paper, we propose and implement on personalized method of individual LBS using collaborative filtering-based recommend system. Proposed personalized LBS system recommends contents which is expected to be interest for individual user, by predicting location-based contents within a user's setting radius. To evaluate performance of proposed system, we observed prediction accuracy with various experimental condition using our prototype. As a result, we confirmed that the convergence of collaborative filtering and LBS is effective for personalized LBS.

• Journal of Gifted/Talented Education
• /
• v.23 no.5
• /
• pp.671-695
• /
• 2013

This study provides an illustrative example of using the multivariate generalizability theory. Specifically, it investigates relative effects of each error source, and finds optimal measurement conditions for the number of items within each content domain that maximizes the reliability-like coefficients, such as a generalizability coefficient and an index of dependability. The method is based on teacher recommendation letters and self-introduction letters, using an analytic scoring method in the context of selection of mathematically gifted students by observation and nomination. This study analyzed data from the 2011 academic year in the science education institute for the gifted, which is attached to the university located in the Seoul metropolitan area. It should be noted that the optimal scoring structures of this study are not generalizable to other selection instruments. However, the methodology applied in this study can be utilized to find optimal measurement conditions for the number of raters, the number of content domains, and the number of items in other selection instruments self-developed by many institutions including: the education institutes for the gifted at provincial offices of education, gifted classes, and the science education institutes for the gifted attached to universities in general. In addition, the methodology will provide bases for making informed decisions in selection instruments of the gifted based on measurement traits.

• Journal of Gifted/Talented Education
• /
• v.24 no.4
• /
• pp.679-701
• /
• 2014

As the validity issue of teacher nominations for the identification of gifted students has been raised recently, this study purports to test the validity of teacher nominations for selecting scientifically gifted students. As the criterion variables, domain specific traits such as science creative problem solving skills and science attitudes and domain general characteristics such as divergent thinking skills, creative attitudes, intrinsic motivation, and leadership were analyzed. Scientifically gifted students, potentially gifted students who had never been enrolled in gifted programs but were nominated as the scientifically gifted by teachers, and general class students participated in the study. The results of ANOVA showed that there were significant differences in all variables but originality factors of the TTCT and science creative problem solving skill test between gifted/nominated students and general class students; gifted/nominated students were significantly superior in these variables to general class students. The discriminant functions analysis yielded a discriminant function that significantly discriminated between gifted/nominated and general class students. Variables loaded on the discriminant function were science creative problem solving skills except for the originality subfactor, and science efficacy. These results imply that while teachers are likely to consider adaptation-oriented academic excellency related to logical thinking skills, problem solving skills, and science performance when nominating students, they may ignore the innovation-oriented property which is indicated as the fluency and originality factors of TTCT. Also, the criteria of teacher nominations are presumed to be congruent with the selection criteria of the gifted education program which pursued academic excellency as the educational goal. This suggests that with such criteria, high performing students in the science area can be sufficiently identified by teachers with no further identification procedures or/and tests.