A hybrid approach of topic model and matrix factorization based on two-step recommendation framework

Zhao, Xiangyu; Niu, Zhendong; Chen, Wei; Shi, Chongyang; Niu, Ke; Liu, Donglei

doi:10.1007/s10844-014-0334-3

Cited by 36 publications

(17 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Hybrid recommendation models: The recommendation problem literature suffers from no shortage of hybrid approaches, all providing unique advantages over individual algorithms. Zhao et al [49] proposed a cascading hybrid model that uses topic modeling in the first step and matrix factorization in the second to compute the expectations of users' ratings on items. Romov et al [50] won the RecSys 2015 Challenge award with an ensemble learning framework that deals with categorical features over implicit feedback.…”

Section: Related Workmentioning

confidence: 99%

EnsVAE: Ensemble Variational Autoencoders for Recommendations

Drif

Zerrad²,

Cherifi³

2020

IEEE Access

View full text Add to dashboard Cite

Recommender systems are information software that retrieves relevant items for users from massive sources of data. The variational autoencoder (VAE) has proven to be a promising approach for recommendation systems, as it can explore high-level user-item relations and extract contingencies from the input effectively. However, the previous variants of VAE have so far seen limited application to domain-specific recommendations that require additional side information. Hence, The Ensemble Variational Autoencoder framework for recommendations (EnsVAE) is proposed. This architecture specifies a procedure to transform sub-recommenders' predicted utility matrix into interest probabilities that allow the VAE to represent the variation in their aggregation. To evaluate the performance of EnsVAE, an instance-called the "Ensemblist GRU/GLOVE model"-is developed. It is based on two innovative recommender systems: 1-) a new "GloVe content-based filtering recommender" (GloVe-CBF) that exploits the strengths of embedding-based representations and stacking ensemble learning techniques to extract features from the item-based side information. 2-) a variant of neural collaborative filtering recommender, named "Gate Recurrent Unit-based Matrix Factorization recommender" (GRU-MF). It models a high level of nonlinearities and exhibits interactions between users and items in latent embeddings, reducing user biases towards items that are rated frequently by users. The developed instance speeds up the reconstruction of the utility matrix with increased accuracy. Additionally, it can switch between one of its sub-recommenders according to the context of their use. Our findings reveal that EnsVAE instances retain as much information as possible during the reconstruction of the utility matrix. Furthermore, the trained VAE's generative trait tackles the cold-start problem by accurately estimating the interest probabilities of newly-introduced users and resources. The empirical study on real-world datasets proves that EnsVAE significantly outperforms the state-of-the-art methods in terms of recommendation performances.

show abstract

Section: Related Workmentioning

confidence: 99%

EnsVAE: Ensemble Variational Autoencoders for Recommendations

Drif

Zerrad²,

Cherifi³

2020

IEEE Access

View full text Add to dashboard Cite

show abstract

“…Although, conventional recommender systems yield an enormous aggregate of appropriate and inappropriate ORIGINAL ARTICL E results to regular keyword-based searches [2] concerning with the interests of Scholars. Apparently, traditional recommender systems are punier to recommend most suitable research resources to the scholars which may cause wackness emanating from irrelevant results due to the cold-start problem [3][4][5] and ranking sparsity problem [6,7]. To elucidate this issue, several recommendation methods [8,9,[18][19][20][21][22][23][24][10][11][12][13][14][15][16][17] have recently appeared to provide Scholars with more appropriate and suitable results.…”

Section: N T R O D U C T I O Nmentioning

confidence: 99%

Context-Aware and Sequential Pattern Mining based recommendations for Research Papers: A Hybrid Approach

Amaad

Jhamat

Riaz³

et al. 2020

jictra

View full text Add to dashboard Cite

The availability of huge volumes of online research papers over scholarly communities has been increasing rapidly with the evolution of the Internet. Meanwhile, several researchers confront troubles while retrieving suitable and relevant research papers according to their research necessities due to information overload. Besides, the research necessities vary from researcher to researcher according to their contextual state and the online behavior in sequential access. Conventional recommendation approaches for instance content-based filtering (CBF) and collaborative filtering (CF) utilize content features and rankings correspondingly, in order to produce recommendations for the researchers. In spite of this, it is inevitable to incorporate scholar’s contextual information and sequential access behavior into the recommendation procedure to generate accurate and personalized recommendations for research papers. Conventional recommender systems do not incorporate such information in the recommended procedure to compute similarities of scholars and provide recommendations; thus, they are more liable to produce an irrelevant list of recommendations in a scholarly environment. Moreover, conventional recommendation approaches generate inaccurate recommendations in presence of a high level of sparsity in the rankings. In this article, we introduce a novel method for research paper recommendations that incorporates the benefits of collective filtering (CF), context-awareness, and sequential pattern mining (SPM) to propose research papers to scholars in a hybrid manner. Context-awareness in our methodology involves the scholar's contextual state, such as skill level and research goals; SPM is used to mine weblogs and reveal sequential access actions of scholars, and CF is used to measure predictions based on correlations between scholars and generate context-aware and sequential trend mining based recommendations for the targeted scholars. Experimental evaluations of our approach indicate the excellence of our approach over other baseline approaches in terms of precision, recall, F1, and mean absolute error (MAE).

show abstract

“…Therefore, a hybrid approach based on a topic model and matrix factorization [15] was introduced to predict the probability that a user will rate an item and the corresponding rating values to improve the accuracy of recommendations. Furthermore, Xing et al [16] devises a joint convolution matrix factorization model that considers various factors.…”

Section: A Factorization-based Modelmentioning

confidence: 99%

A Novel Learning Model Based on Trust Diffusion and Global Item for Recommender Systems

Liu

Ren

2019

IEEE Access

View full text Add to dashboard Cite

Recommender systems can provide users with an ordered list of various items, which greatly assists users to purchase products that they are satisfied with. However, item recommendation has been confronted with some inherent problems, such as sparse ratings and long-tail distribution, resulting in low accuracy of recommendations and insignificant marketing. In this paper, we propose a novel learning model based on trust diffusion and global item (TDGIL) to improve the accuracy of item rating prediction for recommender systems. Specifically, first, the rating information on items is mined and aggregated to the greatest extent based on trust diffusion characteristics among users. The benchmark prediction of item recommendation is updated by a user trust neighbor set and its item ratings, which are obtained by a trust diffusion algorithm. Then, the difference weights and compensation coefficients for all items are defined to learn users' potential preferences in the proposed global item model. Finally, the TDGIL learning algorithm is presented to train and learn the target networks by random gradient descent. The extensive experiments and results on two real-world datasets demonstrated that our proposed model can achieve significant improvements in the accuracy of rating prediction compared with some state-of-the-art methods.

show abstract

A hybrid approach of topic model and matrix factorization based on two-step recommendation framework

Cited by 36 publications

References 21 publications

EnsVAE: Ensemble Variational Autoencoders for Recommendations

EnsVAE: Ensemble Variational Autoencoders for Recommendations

Context-Aware and Sequential Pattern Mining based recommendations for Research Papers: A Hybrid Approach

A Novel Learning Model Based on Trust Diffusion and Global Item for Recommender Systems

Contact Info

Product

Resources

About