• ETRI Journal
• /
• 제35권2호
• /
• pp.311-320
• /
• 2013

In most spectral clustering approaches, the Gaussian kernel-based similarity measure is used to construct the affinity matrix. However, such a similarity measure does not work well on a dataset with a nonlinear and elongated structure. In this paper, we present a new similarity measure to deal with the nonlinearity issue. The maximum flow between data points is computed as the new similarity, which can satisfy the requirement for similarity in the clustering method. Additionally, the new similarity carries the global and local relations between data. We apply it to spectral clustering and compare the proposed similarity measure with other state-of-the-art methods on both synthetic and real-world data. The experiment results show the superiority of the new similarity: 1) The max-flow-based similarity measure can significantly improve the performance of spectral clustering; 2) It is robust and not sensitive to the parameters.

• International Journal of Computer Science & Network Security
• /
• 제22권8호
• /
• pp.175-186
• /
• 2022

This research article develops an Arabic books' recommendation system, which is based on the content similarity that assists users to search for the right book and predict the appropriate and suitable books pertaining to their literary style. In fact, the system directs its users toward books, which can meet their needs from a large dataset of Information. Further, this system makes its predictions based on a set of data that is gathered from different books and converts it to vectors by using the TF-IDF system. After that, the recommendation algorithms such as the cosine similarity, the sequence matcher similarity, and the semantic similarity aggregate data to produce an efficient and effective recommendation. This approach is advantageous in recommending previously unrated books to users with unique interests. It is found to be proven from the obtained results that the results of the cosine similarity of the full content of books, the results of the sequence matcher similarity of Arabic titles of the books, and the results of the semantic similarity of English titles of the books are the best obtained results, and extremely close to the average of the result related to the human assigned/annotated similarity. Flask web application is developed with a simple interface to show the recommended Arabic books by using cosine similarity, sequence matcher similarity, and semantic similarity algorithms with all experiments that are conducted.

• Journal of the Korean Data and Information Science Society
• /
• 제17권4호
• /
• pp.1151-1160
• /
• 2006

The similarity weight, the pearson's correlation coefficient, which is used in the recommender system has a weak point that it cannot predict all of the prediction value. The similarity weight, the vector similarity, has a weak point of the high MAE although the prediction coverage using the vector similarity is higher than that using the pearson's correlation coefficient. The purpose of this study is to suggest how to raise the prediction coverage. Also, the MAE using the suggested method in this study was compared both with the MAE using the pearson's correlation coefficient and with the MAE using the vector similarity, so was the prediction coverage. As a result, it was found that the low of the MAE in the case of using the suggested method was higher than that using the pearson's correlation coefficient. However, it was also shown that it was lower than that using the vector similarity. In terms of the prediction coverage, when the suggested method was compared with two similarity weights as I mentioned above, it was found that its prediction coverage was higher than that pearson's correlation coefficient as well as vector similarity.

• 한국컴퓨터정보학회논문지
• /
• 제23권12호
• /
• pp.219-226
• /
• 2018

Collaborative filtering has been most popular approach to recommend items in online recommender systems. However, collaborative filtering is known to suffer from data sparsity problem. As a simple way to overcome this problem in literature, Jaccard index has been adopted to combine with the existing similarity measures. We analyze performance of such combination in various data environments. We also find optimal weights of factors in the combination using a genetic algorithm to formulate a similarity measure. Furthermore, optimal weights are searched for each user independently, in order to reflect each user's different rating behavior. Performance of the resulting personalized similarity measure is examined using two datasets with different data characteristics. It presents overall superiority to previous measures in terms of recommendation and prediction qualities regardless of the characteristics of the data environment.

• International Journal of Advanced Culture Technology
• /
• 제1권2호
• /
• pp.1-6
• /
• 2013

In this paper, we evaluate performance of existing similarity measurement metric and propose a novel method using user's preferences information entropy to reduce MAE in memory-based collaborative recommender systems. The proposed method applies a similarity of individual inclination to traditional similarity measurement methods. We experiment on various similarity metrics under different conditions, which include an amount of data and significance weighting from n/10 to n/60, to verify the proposed method. As a result, we confirm the proposed method is robust and efficient from the viewpoint of a sparse data set, applying existing various similarity measurement methods and Significance Weighting.

• Journal of Information Processing Systems
• /
• 제7권3호
• /
• pp.413-424
• /
• 2011

A variety of different metrics has been introduced to measure the similarity of two given sequences. These widely used metrics are ranging from spell correctors and categorizers to new sequence mining applications. Different metrics consider different aspects of sequences, but the essence of any sequence is extracted from the ordering of its elements. In this paper, we propose a novel sequence similarity measure that is based on all ordered pairs of one sequence and where a Hasse diagram is built in the other sequence. In contrast with existing approaches, the idea behind the proposed sequence similarity metric is to extract all ordering features to capture sequence properties. We designed a clustering problem to evaluate our sequence similarity metric. Experimental results showed the superiority of our proposed sequence similarity metric in maximizing the purity of clustering compared to metrics such as d2, Smith-Waterman, Levenshtein, and Needleman-Wunsch. The limitation of those methods originates from some neglected sequence features, which are considered in our proposed sequence similarity metric.

• 지능정보연구
• /
• 제9권1호
• /
• pp.179-193
• /
• 2003

군집화(clustering)는 주어진 객체들 중에서 유사한 것들을 몇몇의 집단으로 그룹화 하여 각 집단의 성격을 파악하는데, 실제적으로 각 객체가 유사한지 그렇지 않은지를 측정할 수 있는 도구가 필요하다. 기존의 군집화에서 객체간에 유사하다는 의미는 각 군집(cluster)안에 있는 객체들이 같은 속성 값이 많으면 많을수록 객체간에 유사성이 높아 유사도가 높은 객체끼리 군집을 이루게 된다는 것을 의미했다. 그 중에서도 범주형 속성을 갖는 군집화는 같은 속성 값이면 1, 서로 다르면 0으로 표현하여 유사성을 측정하는 방법이다. 제안된 알고리듬은 속성 값을 0과1로만 표현하는 것에 대한 문제점을 제시하고 서로 다른 속성이라도 속성간에 친밀한 관계가 있다는 개념을 도입하여 어느 정도 유사한 지를 보여준다. 같은 객체간에 같은 값을 갖는 속성이 하나로 없더라도 구해진 유사도에 의해 유사한 개체끼리는 하나의 군집이 될 수 있는 알고리듬을 만든 후 그 군집에 속해 있는 고객들의 니즈와 구매 선호도에 따라 적절한 타겟 마케팅(Target Marketing)을 할 수 있다.

• 한국지능정보시스템학회:학술대회논문집
• /
• 한국지능정보시스템학회 2004년도 추계학술대회
• /
• pp.221-229
• /
• 2004

Recently, there has been enormous growth in the amount of commercial and scientific data, such as protein sequences, retail transactions, and web-logs. Such datasets consist of sequence data that have an inherent sequential nature. However, only a few of the existing clustering algorithms consider sequentiality. This study presents a method for clustering such sequence datasets. The similarity between sequences must be decided before clustering the sequences. This study proposes a new similarity measure to compute the similarity between two sequences using a sequence element. Two clustering algorithms using the proposed similarity measure are proposed: a hierarchical clustering algorithm and a scalable clustering algorithm that uses sampling and a k-nearest neighbor method. Using a splice dataset and synthetic datasets, we show that the quality of clusters generated by our proposed clustering algorithms is better than that of clusters produced by traditional clustering algorithms.

• 컴퓨터교육학회논문지
• /
• 제19권4호
• /
• pp.59-66
• /
• 2016

협력 필터링 시스템에서 데이터 희소성 문제의 해결을 위해 공통평가항목수를 반영하는 방법이 연구되었다. 이러한 방법으로 널리 알려진 자카드 지수는 기존의 유사도 척도와 결합되어 성능을 개선할 수 있었다. 그러나, 다양한 데이터 환경에서 여러 유사도 척도들과 각각 결합했을 때의 성능 개선 효과에 대한 분석 연구는 미미하므로, 본 연구는 이에 대한 분석을 목적으로 한다. 우선 자카드 지수 자체를 유사도 척도로 사용했을때 희소한 데이터셋 상에서 전통적인 척도들보다 월등한 예측 성능을 보였고 추천 성능도 매우 우수하였다. 자카드 지수를 결합함으로써 기존 유사도 척도는 데이터 특성에 상관없이 성능이 대개 향상되었고, 특히 코사인 유사도는 희소한 데이터셋에서 가장 큰 향상을 이루었으나, 평균차이 제곱(Mean Squared Difference)의 유사도는 밀집된 데이터셋에서 오히려 저하된 예측 성능을 보였다. 따라서, 자카드 지수를 결합하여 사용하기 위해 데이터 환경 특성과 유사도 척도를 고려할 필요가 있다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제7권3호
• /
• pp.522-538
• /
• 2013

Content based image retrieval has become an increasingly important research topic for its wide application. It is highly challenging when facing to large-scale database with large variance. The retrieval systems rely on a key component, the predefined or learned similarity measures over images. We note that, the similarity measures can be potential improved if the data distribution information is exploited using a more sophisticated way. In this paper, we propose a similarity measure learning approach for image retrieval. The similarity measure, so called Fisher kernel, is derived from the probabilistic distribution of images and is the function over observed data, hidden variable and model parameters, where the hidden variables encode high level information which are powerful in discrimination and are failed to be exploited in previous methods. We further propose a discriminative learning method for the similarity measure, i.e., encouraging the learned similarity to take a large value for a pair of images with the same label and to take a small value for a pair of images with distinct labels. The learned similarity measure, fully exploiting the data distribution, is well adapted to dataset and would improve the retrieval system. We evaluate the proposed method on Corel-1000, Corel5k, Caltech101 and MIRFlickr 25,000 databases. The results show the competitive performance of the proposed method.