RSVD-based Dimensionality Reduction for RecommenderSystems

Ciesielczyk, Michal; Szwabe, Andrzej

doi:10.7763/ijmlc.2011.v1.25

Cited by 7 publications

(5 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Figure 7 presents the performance of the recommender algorithms as compared in the investigated scenarios (i.e., in scenarios S1-4). It may be concluded that, as it was already shown in [11], the RSVD method outperforms other methods (i.e., PureSVD and RRI) when the standard input data representation is used. As far as the S1 scenario is concerned, that is, the one with the standard data representation based on the actor-object coincidence matrix single relation, it may be seen that, in general, the decomposition-based methods (i.e., PureSVD and RSVD) achieve comparable recommendation quality and that, in general, these methods perform better than RRI (for various values of the training ratio).…”

Section: Methodssupporting

confidence: 56%

“…Firstly, we have used the algorithm based on reflexive random indexing [10]. Secondly, we have used two types of algorithms that are based on the singular value decomposition: a traditional implementation of the method (PureSVD), in which actor vectors are represented as combinations of object vectors without any specific parameterization, and an implementation of the randomized singular value decomposition (RSVD) [11], which is a combination of the reflexive random indexing and SVD. We have chosen so since SVD-based methods have been long considered to be the most efficient recommender engines in real world settings [12][13][14][15].…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

On Efficient Link Recommendation in Social Networks Using Actor-Fact Matrices

Ciesielczyk

Szwabe

Morzy

2015

Scientific Programming

Self Cite

View full text Add to dashboard Cite

Link recommendation is a popular research subject in the field of social network analysis and mining. Often, the main emphasis is put on the development of new recommendation algorithms, semantic enhancements to existing solutions, design of new similarity measures, and so forth. However, relatively little scientific attention has been paid to the impact that various data representation models have on the performance of recommendation algorithms. And by performance we do not mean the time or memory efficiency of algorithms, but the precision and recall of recommender systems. Our recent findings unanimously show that the choice of network representation model has an important and measurable impact on the quality of recommendations. In this paper we argue that the computation quality of link recommendation algorithms depends significantly on the social network representation and we advocate the use of actor-fact matrix as the best alternative. We verify our findings using several state-of-the-art link recommendation algorithms, such as SVD, RSVD, and RRI using both single-relation and multirelation dataset.

show abstract

Section: Methodssupporting

confidence: 56%

Section: Related Workmentioning

confidence: 99%

On Efficient Link Recommendation in Social Networks Using Actor-Fact Matrices

Ciesielczyk

Szwabe

Morzy

2015

Scientific Programming

Self Cite

View full text Add to dashboard Cite

show abstract

“…HPMPP has been compared with three widely-referenced methods, namely, pure Singular Value Decomposition (SVD) [13], Reflective Random Indexing (RRI) [4], Randomized SVD-RI (RSVD-RI) [3]. For both RRI and RSVD-RI, the same configuration of the random indexing function has been used, i.e., a configuration of the seed number s = 2 and the dimension number d = 500, which was chosen as the optimized settings for the MovieLens dataset according to the test results described in [17].…”

Section: Evaluation Methodologymentioning

confidence: 99%

“…The most time consuming operation is the full eigenvalue decomposition of matrix A which is a square matrix of the size equal to the summarized number of elements and facts n + m. It has to be stressed, that A is a much bigger matrix than the user-item matrices being decomposed in the case of using the SVD method. Thus, the computational complexity of the HPMPP method depends mostly on the complexity of eigendecomposition of matrix A, which is in the worst case equal to O((n + m) 3 ). Despite the fact that we use the optimized matrix eigendecomposition algorithm from Intel Math Kernel Library (Intel MKL), complexity of this operation is the main reason why the proposed method requires much more time than other compared methods.…”

Section: Computational Complexitymentioning

confidence: 99%

Holistic Entropy Reduction for Collaborative Filtering

Szwabe

Misiorek

Janasiewicz

et al. 2014

Foundations of Computing and Decision Sciences

Self Cite

View full text Add to dashboard Cite

We propose a collaborative filtering (CF) method that uses behavioral data provided as propositions having the RDF-compliant form of (user X, likes, item Y) triples. The method involves the application of a novel self-configuration technique for the generation of vector-space representations optimized from the information-theoretic perspective. The method, referred to as Holistic Probabilistic Modus Ponendo Ponens (HPMPP), enables reasoning about the likelihood of unknown facts. The proposed vector-space graph representation model is based on the probabilistic apparatus of quantum Information Retrieval and on the compatibility of all operators representing subjects, predicates, objects and facts. The dual graph-vector representation of the available propositional data enables the entropy-reducing transformation and supports the compositionality of mutually compatible representations. As shown in the experiments presented in the paper, the compositionality of the vector-space representations allows an HPMPP-based recommendation system to identify which of the unknown facts having the triple form (user X, likes, item Y) are the most likely to be true in a way that is both effective and, in contrast to methods proposed so far, fully automatic.

show abstract

“…Dimensionality reduction techniques like Singular Value Decomposition have been explored in the past to derive a lower rank approximation of the ratings matrix and consequently reduce processing time. Sarwar et al [10] propose that the SVD-based approach to recommendations produced results that were better than a traditional collaborative filtering algorithm in case of a reasonably dense dataset.In [11] the authors investigate the utility of techniques such as SVD, Random Indexing(RI), Reflective Random Indexing (RRI) and Randomized Singular Value Decomposition (RSVD) for aiding collaborative filtering. They conclude that a combination of RRI and SVD delivers a better recommendation diversity, though SVD results in lesser prediction errors.…”

Section: Literature Review 21 Collaborative Filteringmentioning

confidence: 99%

Improved Collaborative Filtering using Evolutionary Algorithm based Feature Extraction

Anand¹

2013

IJCA

View full text Add to dashboard Cite

The ubiquity of Collaborative Filtering systems is evident in the wide variety of domains to which they have been applied successfully. However a major challenge to such systems is the high dimensionality and sparsity of the expressed preferences. Dealing effectively with large user profiles would improve the scalability of the system whereas reducing sparsity would increase the quality of recommendations. Several approaches in this direction have focused on feature selection and feature extraction in order to reduce the data dimension and thus make the recommendation process more scalable. Some of the features extraction techniques are based on extracting content based features. However many such solutions have been handcrafted and thus not guaranteed to work optimally under all data environments. This work explores Evolutionary algorithms based feature extraction techniques where the extracted features may describe user or item profiles. The features constructed/extracted thus are compact, dense and are discriminative. Moreover they have the advantage of requiring no extra information (such as content description) and are adaptive, delivering the optimal feature extraction scheme for the particular dataset. We have performed experiments with the popular MovieLens dataset and have compared the user-based and item-based evolutionary feature extraction schemes with respect to their accuracy. The experiments establish that the evolutionary feature extraction schemes score over traditional algorithms as well as content-based feature extraction schemes. Moreover we find that the item-based evolutionary feature extraction schemes outperform their user-based counterparts under varying parameter values.

show abstract

RSVD-based Dimensionality Reduction for RecommenderSystems

Cited by 7 publications

References 14 publications

On Efficient Link Recommendation in Social Networks Using Actor-Fact Matrices

On Efficient Link Recommendation in Social Networks Using Actor-Fact Matrices

Holistic Entropy Reduction for Collaborative Filtering

Improved Collaborative Filtering using Evolutionary Algorithm based Feature Extraction

Contact Info

Product

Resources

About