Multiview Fusion Using Transformer Model for Recommender Systems: Integrating the Utility Matrix and Textual Sources
Abstract
:1. Introduction
 Developing a novel model that combines views from both the utility matrix and textual sources, which utilizes feature extraction techniques from various information sources and a conversion algorithm to segment the feature vectors of each pair (user, item) into a sequence of token vectors, which serve as input for a classification model;
 Incorporating Transformer models into our approach for multimodal recommender systems, that serves as a strong tool for handling the challenge of integrating diverse sources of information in RSs, that can selfselect features that avoid or minimize noise or conflicting signals, and it has the potential to significantly enhance the performance of RSs in practical settings;
 Conducting experiments on the MovieLens and Amazon datasets to verify the effectiveness of our proposed model.
2. Related Works
3. Background
3.1. UserBased Recommender Systems
 $u.v$ is the dot product of vectors u and v;
 $\left\rightu\left\right$ and $\left\rightv\left\right$ are the Euclidean norms of vectors u and v, respectively.
3.2. ItemBased Recommender Systems
3.3. The Transformer Encoder Architecture
3.4. Matrix Factorization
3.5. Feature Extraction
3.5.1. Term FrequencyInverse Document Frequency (TFIDF)
 Term Frequency (TF), which is calculated by Equation (7), represents the frequency of occurrence of a term within the entire document, which can be interpreted as the likelihood of discovering a particular word in the document. It is computed as the number of occurrences of a word ${w}_{i}$ in a review ${r}_{j}$, relative to the total number of words in the review ${r}_{j}$;
 Inverse Document Frequency (IDF) is a metric that evaluates the frequency of a term across the documents in a corpus, as described by Equations (8) and (9). It emphasizes rare words that appear in only a few documents throughout the corpus and results in a high IDF score. The value of IDF is obtained by lognormalizing the ratio of the total number of documents D in the corpus to the number of documents that contain the term t.
3.5.2. Bidirectional Encoder Representations from Transformers (BERT)
4. The Proposed Model
4.1. Our Multiview Transformer Model for Recommendation
Algorithm 1 Multiview Transformer Model For Recommendation 

4.2. The Feature Extraction Algorithm
 The user features are represented by a matrix of size $(m\times f)$, where each row corresponds to a specific user and contains a vector of f features. This matrix is considered as a view of the user, meaning that it represents the user’s characteristics;
 The item features are represented by a matrix of size $(n\times f)$, where each item is represented by a vector of f features. Each row in this matrix corresponds to a specific item and is considered a view of that item.
 RatingFeature, which pertains to the user u and item i, and is represented as Rat(u) and Rat(i) respectively;
 ReviewFeature, which pertains to the user u and is represented as Rev(u);
 DescriptionFeature, which pertains to the item i and is represented as Des(i);
 CategoryFeature, which pertains to the item i and is represented as Cat(i).
4.3. The Feature Conversion Algorithm
Algorithm 2 The Feature Conversion Algorithm 

5. Experimental Results
5.1. The Dataset Summary
 The MovieLens dataset includes 77,763 ratings of 27,041 users and 8203 movies;
 The Electronic dataset includes 80,472 ratings of 1042 users and 21,200 items;
 The Video Games dataset includes 98,769 ratings of 2291 users and 24,708 items;
 The Toys and Games dataset includes 68,102 ratings of 5462 users and 3028 items.
5.2. Evaluation Metrics
5.2.1. Mean Absolute Error (MAE)
5.2.2. Root Mean Square Error (RMSE)
5.2.3. Precision
5.3. Experimental Setups for Data
 For the MovieLens dataset, there are 62,211 ratings for training, 3110 ratings for validation, and 12,442 ratings for testing;
 For the Toys and Games dataset, there are 54,482 ratings for training, 2724 ratings for validation, and 10,896 ratings for testing;
 For the Electronic dataset, there are 64,378 ratings for training, 3218 ratings for validation, and 12,876 ratings for testing;
 For the Video and Games dataset, there are 79,016 ratings for training, 3950 ratings for validation, and 15,803 ratings for testing.
5.4. Experimental Setups for Models
 Because we use BERT for representing user reviews in our dataset, the magnitude of the resulting attribute vector is 768 dimensions, which is equal to the magnitude of the output vector of the BERTbase model;
 To represent the latent attributes of users and items using the factorization matrix method, we tested different magnitudes of the latent feature vector, including 20, 50, and 100. After evaluating the performance on the validation set, we selected the magnitude of 50 as the best choice;
 When using the TFIDF measure for attribute representation, we utilized the set of all tokens in our data. Therefore, the corresponding magnitude of these representation vectors will be equal to the number of keywords.
 For the Movielens dataset, the dimensions of $f{u}_{1}$, $f{i}_{1}$, and $f{i}_{2}$ are 50, 50, and 20, respectively;
 For the AmazonToys and Games dataset, the dimensions of $f{u}_{1}$, $f{i}_{1}$, $f{i}_{2}$, $f{u}_{2}$, and $f{i}_{3}$ are 50, 50, 321, 768, and 768, respectively;
 For the AmazonElectronics dataset, the dimensions of $f{u}_{1}$, $f{i}_{1}$, $f{i}_{2}$, $f{u}_{2}$, and $f{i}_{3}$ are 50, 50, 944, 768, and 768, respectively;
 For the AmazonVideo and Games dataset, the dimensions of $f{u}_{1}$, $f{i}_{1}$, $f{i}_{2}$, $f{u}_{2}$, and $f{i}_{3}$ are 50, 50, 16,978, 768, and 768, respectively.
 Experiment with data represented by 2 views, representing users and items, generated from the utility matrix;
 Experiment with data represented by all views. For MovieLens data, there are 3 views in all, of which 2 are from the utility matrix and 1 view is Genre; For Amazon data, there are 5 views in all, of which 2 are from the utility matrix and 3 are from user review, item description, and item category.
 We have conducted experiments with different configurations of the Transformer model in our proposed approach, including varying numbers of layers (2, 4, or 6) and hidden state values (20 and 50). After evaluating their performance on the validation set, we selected the optimal configuration of (Layers = 4, Hidden state = 20, Heads = 4);
 We have also designed two baseline models, namely the MLP (Feedforward Neural Network) and the Gated Recurrent Units (GRU), to demonstrate that the Transformer model performs better;
 In addition, we have tested a strong baseline model, the Singular Value Decomposition (SVD) on the utility matrix, which is known to be one of the most effective models for the recommender system (RS) problem.
 For the Amazon dataset, we incorporated category and description as attributes for the item nodes, while the user nodes were enriched with the review attribute;
 For the MovieLens dataset, the item nodes were augmented with the genre attribute;
 We established connections between the user and item nodes in the graph to model the interactions between them. To represent the strength of these interactions, we assigned ratings as the weight property of the edges. This allowed us to reflect the degree of preference that each user showed for each item, which is an important factor in our recommendation system.
5.5. Results
 Combining information from the utility matrix and other textual features using the Transformer model results in better outcomes compared to relying solely on features generated from the utility matrix;
 The Transformer model outperforms other models (specifically, we compared it with GRU and MLP) in terms of yielding better results.
6. Conclusions
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Acknowledgments
Conflicts of Interest
Abbreviations
BERT  Bidirectional Encoder Representations from Transformers 
BiLSTM  Bidirectional Long ShortTerm Memory 
CF  Collaborative Filtering 
DNN  Deep Neural Network 
GRU  Gated Recurrent Units 
KNN  K Nearest Neighbors 
LDA  Latent Dirichlet Allocation 
MLP  Multilayer Perceptron 
RS  Recommender System 
SVD  Singular Value Decomposition 
TFIDF  Term FrequencyInverse Document Frequency 
References
 Feng, C.; Liang, J.; Song, P.; Wang, Z. A fusion collaborative filtering method for sparse data in recommender systems. Inf. Sci. 2020, 521, 365–379. [Google Scholar] [CrossRef]
 Geetha, G.; Safa, M.; Fancy, C.; Saranya, D. A hybrid approach using collaborative filtering and content based filtering for recommender system. J. Phys. Conf. Ser. 2018, 1000, 012101. [Google Scholar] [CrossRef]
 Hoyer, P.O. Nonnegative matrix factorization with sparseness constraints. J. Mach. Learn. Res. 2004, 5, 1457–1469. [Google Scholar]
 Wang, Z.; Chen, H.; Li, Z.; Lin, K.; Jiang, N.; Xia, F. VRConvMF: Visual recurrent convolutional matrix factorization for movie recommendation. IEEE Trans. Emerg. Top. Comput. Intell. 2021, 6, 519–529. [Google Scholar] [CrossRef]
 Xue, H.J.; Dai, X.; Zhang, J.; Huang, S.; Chen, J. Deep matrix factorization models for recommender systems. In Proceedings of the TwentySixth International Joint Conference on Artificial Intelligence (IJCAI17), Melbourne, Australia, 19–25 August 2017; Volume 17, pp. 3203–3209. [Google Scholar]
 Yang, E.; Huang, Y.; Liang, F.; Pan, W.; Ming, Z. FCMF: Federated collective matrix factorization for heterogeneous collaborative filtering. Knowl.Based Syst. 2021, 220, 106946. [Google Scholar] [CrossRef]
 Zhang, H.; Ganchev, I.; Nikolov, N.S.; Ji, Z.; O’Droma, M. FeatureMF: An item feature enriched matrix factorization model for item recommendation. IEEE Access 2021, 9, 65266–65276. [Google Scholar] [CrossRef]
 Reddy, S.; Nalluri, S.; Kunisetti, S.; Ashok, S.; Venkatesh, B. Contentbased movie recommendation system using genre correlation. In Proceedings of the Smart Intelligent Computing and Applications: Proceedings of the Second International Conference on SCI 2018; Springer: Singapore, 2019; Volume 2, pp. 391–397. [Google Scholar]
 Wahyudi, K.; Latupapua, J.; Chandra, R.; Girsang, A.S. Hotel contentbased recommendation system. J. Phys. Conf. Ser. 2020, 1485, 012017. [Google Scholar] [CrossRef]
 Singla, R.; Gupta, S.; Gupta, A.; Vishwakarma, D.K. FLEX: A content based movie recommender. In Proceedings of the 2020 International Conference for Emerging Technology (INCET), Belgaum, India, 5–7 June 2020; IEEE: Piscataway, NJ, USA, 2020; pp. 1–4. [Google Scholar]
 Ghauth, K.I.; Abdullah, N.A. Learning materials recommendation using good learners’ ratings and contentbased filtering. Educ. Technol. Res. Dev. 2010, 58, 711–727. [Google Scholar] [CrossRef]
 Ghazanfar, M.; PrugelBennett, A. An improved switching hybrid recommender system using naive bayes classifier and collaborative filtering. In Proceedings of the International MultiConference of Engineers and Computer Scientists (IMECS 2010), Hong Kong, China, 17–19 March 2010; Volume 1. [Google Scholar]
 Mulay, A.; Sutar, S.; Patel, J.; Chhabria, A.; Mumbaikar, S. Job Recommendation System Using Hybrid Filtering. ITM Web Conf. 2022, 44, 02002. [Google Scholar] [CrossRef]
 Nayak, R.; Mirajkar, A.; Rokade, J.; Wadhwa, G. Hybrid recommendation system for movies. Int. Res. J. Eng. Technol. 2018, 5, 1217–1220. [Google Scholar]
 Okaka, R.A.; Mwangi, W.; Okeyo, G. A Hybrid Approach for Personalized Recommender System Using Weighted Term Frequency Inverse Document Frequency. Int. J. Comp. Appl. Technol. Res. 2016, 5, 764–774. [Google Scholar]
 Otegbade, O. A Hybridized Recommendation System on Movie Data Using ContentBased and Collaborative Filtering. Available online: http://repository.aust.edu.ng:8080/xmlui/handle/123456789/622 (accessed on 15 May 2016).
 Walek, B.; Fajmon, P. A hybrid recommender system for an online store using a fuzzy expert system. Expert Syst. Appl. 2023, 212, 118565. [Google Scholar] [CrossRef]
 Liu, K.; Li, Y.; Xu, N.; Natarajan, P. Learn to combine modalities in multimodal deep learning. arXiv 2018, arXiv:1805.11730. [Google Scholar]
 Conceiç ao, F.L.; Pádua, F.L.; Lacerda, A.; Machado, A.C.; Dalip, D.H. Multimodal data fusion framework based on autoencoders for topN recommender systems. Appl. Intell. 2019, 49, 3267–3282. [Google Scholar] [CrossRef]
 Ren, X.; Yang, W.; Jiang, X.; Jin, G.; Yu, Y. A deep learning framework for multimodal course recommendation based on LSTM+ attention. Sustainability 2022, 14, 2907. [Google Scholar] [CrossRef]
 Wang, J.; Mao, H.; Li, H. FMFN: Finegrained multimodal fusion networks for fake news detection. Appl. Sci. 2022, 12, 1093. [Google Scholar] [CrossRef]
 Choudhury, S.S.; Mohanty, S.N.; Jagadev, A.K. Multimodal trust based recommender system with machine learning approaches for movie recommendation. Int. J. Inf. Technol. 2021, 13, 475–482. [Google Scholar] [CrossRef]
 He, X.; Liao, L.; Zhang, H.; Nie, L.; Hu, X.; Chua, T.S. Neural collaborative filtering. In Proceedings of the 26th International Conference on World Wide Web, Perth, Australia, 3–7 April 2017; pp. 173–182. [Google Scholar]
 NikzadKhasmakhia, N.; Balafara, M.; FeiziDerakhshia, M.R.; Motamedb, C. BERTERS: Multimodal Representation Learning for Expert Recommendation System with Transformer. arXiv 2020, arXiv:2007.07229. [Google Scholar]
 Yang, B.; Mei, T.; Hua, X.S.; Yang, L.; Yang, S.Q.; Li, M. Online video recommendation based on multimodal fusion and relevance feedback. In Proceedings of the 6th ACM International Conference on Image and Video Retrieval, Amsterdam, The Netherlands, 9–11 July 2007; pp. 73–80. [Google Scholar]
 Pliakos, K.; Kotropoulos, C. Simultaneous image tagging and geolocation prediction within hypergraph ranking framework. In Proceedings of the 2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Florence, Italy, 4–9 May 2014; pp. 6894–6898. [Google Scholar] [CrossRef]
 Xia, X.; Yin, H.; Yu, J.; Wang, Q.; Cui, L.; Zhang, X. Selfsupervised hypergraph convolutional networks for sessionbased recommendation. In Proceedings of the AAAI Conference on Artificial Intelligence, Virtual, 2–9 February 2021; Volume 35, pp. 4503–4511. [Google Scholar]
 Fan, W.; Ma, Y.; Li, Q.; He, Y.; Zhao, E.; Tang, J.; Yin, D. Graph neural networks for social recommendation. In Proceedings of the World Wide Web Conference, San Francisco, CA, USA, 13–17 May 2019; pp. 417–426. [Google Scholar]
 Rendle, S.; Freudenthaler, C.; SchmidtThieme, L. Factorizing personalized markov chains for nextbasket recommendation. In Proceedings of the 19th International Conference on World Wide Web, Raleigh, NC, USA, 26–30 April 2010; pp. 811–820. [Google Scholar]
 Rendle, S.; Krichene, W.; Zhang, L.; Anderson, J. Neural collaborative filtering vs. matrix factorization revisited. In Proceedings of the 14th ACM Conference on Recommender Systems, Virtual Event, Brazil, 22–26 September 2020; pp. 240–248. [Google Scholar]
 Sun, F.; Liu, J.; Wu, J.; Pei, C.; Lin, X.; Ou, W.; Jiang, P. BERT4Rec: Sequential recommendation with bidirectional encoder representations from transformer. In Proceedings of the 28th ACM International Conference on Information and Knowledge Management, Beijing, China, 3–7 November 2019; pp. 1441–1450. [Google Scholar]
 Qiu, G.; Yu, X.; Jiang, L.; Ma, B. Textaware recommendation model based on multiattention neural networks. In Proceedings of the Knowledge Science, Engineering and Management: 14th International Conference, KSEM 2021, Tokyo, Japan, 14–16 August 2021, Proceedings, Part I 14; Springer: Cham, Switzerland, 2021; pp. 590–603. [Google Scholar]
 Chen, J.; Zhang, H.; He, X.; Nie, L.; Liu, W.; Chua, T.S. Attentive Collaborative Filtering: Multimedia Recommendation with Item and ComponentLevel Attention. In Proceedings of the 40th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR’17, Tokyo, Japan, 7–11 August 2017; Association for Computing Machinery: New York, NY, USA, 2017; pp. 335–344. [Google Scholar] [CrossRef]
 Chen, L.; Cao, J.; Wang, Y.; Liang, W.; Zhu, G. Multiview Graph Attention Network for Travel Recommendation. Expert Syst. Appl. 2022, 191, 116234. [Google Scholar] [CrossRef]
 Tai, C.Y.; Wu, M.R.; Chu, Y.W.; Chu, S.Y.; Ku, L.W. Mvin: Learning multiview items for recommendation. In Proceedings of the 43rd international ACM SIGIR Conference on Research and Development in Information Retrieval, Virtual, 25–30 July 2020; pp. 99–108. [Google Scholar]
 Ghasemi, N.; Momtazi, S. Neural text similarity of user reviews for improving collaborative filtering recommender systems. Electron. Commer. Res. Appl. 2021, 45, 101019. [Google Scholar] [CrossRef]
 Terzi, M.; Rowe, M.; Ferrario, M.A.; Whittle, J. Textbased userknn: Measuring user similarity based on text reviews. In Proceedings of the International Conference on User Modeling, Adaptation, and Personalization; Springer: Cham, Switzerland, 2014; pp. 195–206. [Google Scholar]
 Wang, H.M.; Yu, G. Personalized recommendation system Kneighbor algorithm optimization. In Proceedings of the International Conference on Information Technologies in Education and Learning (ICITEL 2015), Hangzhou, China, 19–20 December 2015. [Google Scholar]
 Cui, B.B. Design and implementation of movie recommendation system based on Knn collaborative filtering algorithm. ITM Web Conf. 2017, 12, 04008. [Google Scholar] [CrossRef]
 Kamali, P.; Sudha, P.; Akshaya, S. Outfit Recommender System using KNN Algorithm. Int. J. Eng. Res. Technol. 2018, 6. Available online: https://www.ijert.org/research/outfitrecommendersystemusingknnalgorithmIJERTCONV6IS07133.pdf (accessed on 30 January 2023).
 Li, L.; Zhang, Z.; Zhang, S. Hybrid algorithm based on content and collaborative filtering in recommendation system optimization and simulation. Sci. Program. Towards Smart World 2021, 4. [Google Scholar] [CrossRef]
 Shi, W.; Wang, L.; Qin, J. User embedding for rating prediction in SVD++based collaborative filtering. Symmetry 2020, 12, 121. [Google Scholar] [CrossRef]
 Hasan, M.; Roy, F. An item–item collaborative filtering recommender system using trust and genre to address the coldstart problem. Big Data Cogn. Comput. 2019, 3, 39. [Google Scholar] [CrossRef]
 Duan, R.; Jiang, C.; Jain, H.K. Combining reviewbased collaborative filtering and matrix factorization: A solution to rating’s sparsity problem. Decis. Support Syst. 2022, 156, 113748. [Google Scholar] [CrossRef]
 Ho, T.L.; Le, A.C. Multiview Transformation in Recommender Systems. In Proceedings of the 2021 International Conference on System Science and Engineering (ICSSE), Ho Chi Minh City, Vietnam, 26–28 August 2021; IEEE: Piscataway, NJ, USA, 2021; pp. 88–91. [Google Scholar]
 Kim, H.N.; ElSaddik, A.; Jo, G.S. Collaborative errorreflected models for coldstart recommender systems. Decis. Support Syst. 2011, 51, 519–531. [Google Scholar] [CrossRef]
 Vaswani, A.; Shazeer, N.; Parmar, N.; Uszkoreit, J.; Jones, L.; Gomez, A.N.; Kaiser, Ł; Polosukhin, I. Attention is all you need. Adv. Neural Inf. Process. Syst. 2017, 30. [Google Scholar] [CrossRef]
 Kotu, V.; Deshpande, B. Data Science: Concepts and Practice; Morgan Kaufmann: Burlington, MA, USA, 2018. [Google Scholar]
 Eklund, M. Comparing Feature Extraction Methods and Effects of PreProcessing Methods for MultiLabel Classification of Textual Data. Available online: http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva231438 (accessed on 10 February 2023).
 Devlin, J.; Chang, M.W.; Lee, K.; Toutanova, K. Bert: Pretraining of deep bidirectional transformers for language understanding. arXiv 2018, arXiv:1810.04805. [Google Scholar]
 Schneider, P.; Xhafa, F. Anomaly Detection and Complex Event Processing Over IoT Data Streams: With Application to EHealth and Patient Data Monitoring; Academic Press: Cambridge, MA, USA, 2022. [Google Scholar]
 Salam Patrous, Z.; Najafi, S. Evaluating Prediction Accuracy for Collaborative Filtering Algorithms in Recommender Systems. Ph.D. Thesis, KTH Royal Inst. of Technol., Stockholm, Sweden, 2016. Available online: https://urn.kb.se/resolve?urn=urn%3Anbn%3Ase%3Akth%3Adiva186456 (accessed on 20 January 2023).
 Ho, T.L.; Le, A.C.; Vu, D.H. A Hybrid Model for Recommendation Systems. In Proceedings of the Advances in Intelligent Information Hiding and Multimedia Signal Processing: Proceeding of the 16th International Conference on IIHMSP in Conjunction with the 13th International Conference on FITAT, Ho Chi Minh City, Vietnam, 5–7 November 2020; Springer: Singapore, 2021; Volume 2, pp. 316–323. [Google Scholar]
 Ni, J.; Li, J.; McAuley, J. Justifying recommendations using distantlylabeled reviews and finegrained aspects. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLPIJCNLP), Hong Kong, China, 3–7 November 2019; pp. 188–197. [Google Scholar]
Dataset  Ratings  Users  Items 

MovieLens  77,763  27,041  8203 
Amazon: Toys and Games subcategory  68,102  5462  3028 
Amazon: Video and Games subcategory  98,769  2291  24,708 
Amazon: Electronic subcategory  80,472  1042  21,200 
The Dataset  Trained Ratings  Validated Ratings  Tested Ratings 

MovieLens  62,211  3110  12,442 
AmazonToys and Games  54,482  2724  10,896 
AmazonElectronic  64,378  3218  12,876 
AmazonVideo and Games  79,016  3950  15,803 
The Dataset  The Dimensions of User Latent Features  The Dimensions of Item Latent Features  The Dimensions of Item Genre/Category Features  The Dimensions of User Review Features  The Dimensions of Item Description Features 

MovieLens  50  50  20  
AmazonToys and Games  50  50  321  768  768 
AmazonElectronic  50  50  944  768  768 
AmazonVideo and Games  50  50  16,978  768  768 
Models  MAE  RMSE  Precision 

MovieLens:  
MLP RS [three views]  1.261  1.677  51.03% 
GRU RS [three views]  0.852  1.062  65.39% 
Transformerbased RS [two views]  0.973  1.230  67.11% 
Transformerbased RS [three views]  0.964  1.216  67.16% 
SVD RS  1.562  2.293  47.75% 
Amazon: Toys and Games:  
MLP RS [five views]  0.527  0.776  86.65% 
GRU RS [five views]  0.729  0.889  86.31% 
Transformerbased RS [two views]  0.570  1.09  86.85% 
Transformerbased RS [five views]  0.445  0.743  92.07% 
SVD RS  0.844  1.120  68.06% 
Amazon: Video and Games:  
MLP RS [five views]  0.729  0.999  73.74% 
GRU RS [five views]  0.941  1.174  76.89% 
Transformerbased RS [two views]  0.842  2.062  77.09% 
Transformerbased RS [five views]  0.689  1.471  83.31% 
SVD RS  1.572  1.254  63.99% 
Amazon: Electronic:  
MLP RS [five views]  0.628  0.914  83.22% 
GRU RS [five views]  0.78  1.007  85.80% 
Transformerbased RS [two views]  0.606  1.362  85.94% 
Transformerbased RS [five views]  0.554  1.195  87.83% 
SVD RS  0.85  1.159  69.59% 
Dataset  Methods  Number of Views  MAE  RMSE  Precision 

MovieLens  Graph  Three views  1.401  1.908  60.43% 
Our proposed model: Transformer  Three views  0.964  1.216  67.16%  
Amazon: Toys and Games  Graph  Five views  1.285  1.823  60.30% 
Our proposed model: Transformer  Five views  0.445  0.743  92.07%  
Amazon: Video and Games  Graph  Five views  1.305  1.742  56.01% 
Our proposed model: Transformer  Five views  0.689  1.471  83.31%  
Amazon: Electronic  Graph  Five views  1.542  1.863  42.25% 
Our proposed model: Transformer  Five views  0.554  1.195  87.83% 
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. 
© 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Ho, T.L.; Le, A.C.; Vu, D.H. Multiview Fusion Using Transformer Model for Recommender Systems: Integrating the Utility Matrix and Textual Sources. Appl. Sci. 2023, 13, 6324. https://doi.org/10.3390/app13106324
Ho TL, Le AC, Vu DH. Multiview Fusion Using Transformer Model for Recommender Systems: Integrating the Utility Matrix and Textual Sources. Applied Sciences. 2023; 13(10):6324. https://doi.org/10.3390/app13106324
Chicago/Turabian StyleHo, ThiLinh, AnhCuong Le, and DinhHong Vu. 2023. "Multiview Fusion Using Transformer Model for Recommender Systems: Integrating the Utility Matrix and Textual Sources" Applied Sciences 13, no. 10: 6324. https://doi.org/10.3390/app13106324