A hybrid user similarity model for collaborative filtering

Wang, Yong; Deng, Jiankang; Gao, Jerry; Zhang, Pu

doi:10.1016/j.ins.2017.08.008

Cited by 132 publications

(44 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The most relevant works to us are BCF [14] and HUSM [20], which predict user similarity by utilizing all ratings of each user and Figure 2: An illustration of our basic philosophy to evaluate user distance. The bigger the square the larger the item distance d(i, j), e.g.…”

Section: Related Workmentioning

confidence: 99%

“…The smaller d(i, j) the more similar i and j. We can derive d(i, j) from ratings on items [14,20] or content information [22], such as item tags, comments, etc. In this paper, we assume d(i, j) are given.…”

Section: The Proposed Similarity Measurementioning

confidence: 99%

“…Some traditional similarity measures, such as Cosine (COS) [4], Persons Correlation Coefficient (PCC) [4] and so on, have been widely used in CF to evaluate similarity [5,20]. However, They only consider ratings on the co-rated items [14,20], which may not represent the taste of a user properly. The reason is that the information is lost while ignoring the ratings on the non-co-rated items [14,20].…”

Section: Introductionmentioning

confidence: 99%

“…However, They only consider ratings on the co-rated items [14,20], which may not represent the taste of a user properly. The reason is that the information is lost while ignoring the ratings on the non-co-rated items [14,20]. Figure 1 shows examples of co-rated items.…”

Section: Introductionmentioning

confidence: 99%

“…Figure 1 shows examples of co-rated items. Therefore, these measures perform poorly if there are no sufficient numbers of co-rated items and thus are not suitable for real world sparse data [14,20], because the more sparse the data the less likely the corated items can exist. Moreover, these measures are not applicable when there are no co-rated items at all.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

PMD: An Optimal Transportation-Based User Distance for Recommender Systems

Meng

Dai

Xiao

et al. 2020

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Collaborative filtering, a widely-used recommendation technique, predicts a user's preference by aggregating the ratings from similar users. Traditional similarity measures utilize ratings of only co-rated items while computing similarity between a pair of users. As a result, these measures cannot fully utilize the rating information and are not suitable for real world sparse data. To solve these issues, we propose a novel user distance measure named Preference Mover's Distance (PMD) which makes full use of all ratings made by each user. Our proposed PMD can properly measure the distance between a pair of users even if they have no co-rated items. We show that this measure can be cast as an instance of the Earth Mover's Distance, a wellstudied transportation problem for which several highly efficient solvers have been developed. Experimental results show that PMD can help achieve superior recommendation accuracy than state-ofthe-art methods, especially when training data is very sparse. arXiv:1909.04239v1 [cs.IR]

show abstract

Section: Related Workmentioning

confidence: 99%

Section: The Proposed Similarity Measurementioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

PMD: An Optimal Transportation-Based User Distance for Recommender Systems

Meng

Dai

Xiao

et al. 2020

Lecture Notes in Computer Science

View full text Add to dashboard Cite

show abstract

AutoML classifier clustering procedure

Koren

Hallin

Koren

et al. 2021

Int J of Intelligent Sys

View full text Add to dashboard Cite

Recommendation systems are one of the main applications of machine learning (ML) used across different industries. This paper presents a new automated machine learning (AutoML) method of providing recommendations by processing data sets using ML algorithms, targeting, and offering cluster recommendations for new observations and as a new decision support method. The AutoML conducts a complete procedure and includes analysis and division of data into an efficient number of clusters. We apply the k‐means, using the elbow method to calculate costs per cluster, followed by analyzing the allocation of the data into the clusters, thus providing a method for prediction and for allocating new observations to the relevant clusters (knn). This study includes two experiments using the complete AutoML procedure conducted on a data set, with more than two million records and dozens of attributes. This was done to demonstrate how the AutoML method can be implemented and successfully run with a high‐capacity analysis procedure. The motivation was to analyze, examine, assign, and integrate new observations into existing clusters that have been defined. The results showed that the AutoML method provided efficient recommendations for new observations with an accuracy rate of 99.99%. Hence, the AutoML procedure can offer a full system for any organization to efficiently split existing data into clusters, assign to clusters, and predict the cluster allocation of new observations. The significant contribution of this study is a simple method that can achieve fast and high accuracy clustering for ongoing (new) classified data acquired by an organization.

show abstract

Computing User Similarity by Combining Item Ratings and Background Knowledge from Linked Open Data

2018

Web Information Systems and Applications

View full text Add to dashboard Cite

A hybrid user similarity model for collaborative filtering

Cited by 132 publications

References 21 publications

PMD: An Optimal Transportation-Based User Distance for Recommender Systems

PMD: An Optimal Transportation-Based User Distance for Recommender Systems

AutoML classifier clustering procedure

Computing User Similarity by Combining Item Ratings and Background Knowledge from Linked Open Data

Contact Info

Product

Resources

About