Recommender System for Retail Domain: An Insight on

Techniques and Evaluations

Lit-Jie Chew Su-Cheng Haw Samini Subramaniam
Faculty of Computing Informatics Faculty of Computing Informatics Faculty of Computing Informatics
Multimedia University Multimedia University Multimedia University
Persiaran Multimedia Persiaran Multimedia Persiaran Multimedia
63100 Cyberjaya 63100 Cyberjaya 63100 Cyberjaya
[email protected] [email protected] [email protected]

ABSTRACT experience as well as improve the sales of the company [3].

Recommender system has been developed as a useful tool The RS approach can be mainly classified into content-based
especially when we reached the era of big data and in the filtering (CB) and collaborative filtering (CF). In general, a CB
meanwhile the internet has been overwhelming with lots of RS takes user preferences and history details, then provide the
choices. There is a need for people to filter the information to recommendation that matches with the preferences and history.
search for their needs and wants efficiently. E-commerce website CF RS provides the recommendation based on the previous user
such as Amazon and Netflix have been using recommender that having similar preferences and history. On the other hand,
system to build and boost their sales through the personalization more recently, another group of RS namely, hybrid RS has
recommendation. With the success in the e-commerce area, emerged. Hybrid method basically overcomes the limitations of
researchers are keen on finding a method to boost traditional one approach, by combining the good features of the other
offline retailer sales thru the recommender system. Therefore, in approaches. There are also knowledge-based RS which
this paper, we introduced the existing recommender system and recommend items based on the domain knowledge.
discuss the method of filtering of each method. Then, we provide
The first RS was developed by Goldberg, Nichols, Oki & Terry in
the overview of the recent paper in retailer and e-commerce
1992 [4]. Tapestry was an electronic messaging system, in which
domain to provide the insight and trends such as the filtering
allowed the user to filter and rate the message. RS has been
techniques and evaluation metric used. Several possible research
proven that it provide better results compare to simple filtering
direction has been discussed based on the current trends and
technique [3, 5, 6]. Real world application like Amazon [1] using
problems.
item to item collaborative filtering techniques to recommend the
CCS Concepts same types of item to the same group of users. Movielens [5]
recommends the movies based on the collaborative filtering,
• Information systems➝Information retrieval➝Retrieval tasks
which utilizes the rating to search for similar user profile. Netflix
and goals➝Recommender systems.
[6] uses matrix factorization method to adapt the changing of user
Keywords preference over time in the model.
Recommender system; Content-based filtering; Collaborative Recently, many researchers have enhanced the RS and apply in
filtering; Hybrid filtering many domain such as e-commence [7, 8], video [9], tourism [10],
books [11] and social media [12]. The enhancements include
1. INTRODUCTION optimizing the RS structure, algorithms and techniques. In this
Recommender System (RS) is a system that intelligently provides paper, we focus on discussing RS in retailer domain (which
recommender lists to user according to their interests. With includes offline shops and online shops (e-commerce)) from year
numerous amount of data flow over the internet nowadays, user 2017 to 2019. We will provide an overview of the method of RS
may find it difficult to search for the actual information that they used and evaluation techniques for the RS.
needed. With the existence of RS, user can get customized
recommendations based on their input including user rating, 2. RECOMMENDATION TECHNIQUES
preferred category, search results and so on. Some e-commence OVERVIEW
company such as Amazon [1] and eBay [2] have integrated the RS Many recommendation techniques have emerged since the
into their website to promote targeted products to the customer inception of RS. These can be grouped into CB, CF and hybrid
based on their interest. A high quality RS can improve the user filtering. There are also other categories such as graph-based,
Permission to make digital or hard copies of all or part of this work for deep-learning based, knowledge-based and so on. We will briefly
personal or classroom use is granted without fee provided that copies are describe these techniques in this section.
not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page. Copyrights 2.1 Collaborative Filtering-Based
for components of this work owned by others than ACM must be honored.
Abstracting with credit is permitted. To copy otherwise, or republish, to
Recommendation Techniques
post on servers or to redistribute to lists, requires prior specific permission In CF techniques, it finds the similar user and use the similar
and/or a fee. Request permissions from [email protected]. user’s preferences or rating pattern to produce recommendation.
ICCMS '20, February 26–28, 2020, Brisbane, QLD, Australia This approach has been widely adopted in various domains. CF
© 2020 Association for Computing Machinery. techniques can be further classified as memory-based and model-
ACM ISBN 978-1-4503-7703-4/20/02…$15.00 based. The memory-based system takes user and item relationship
DOI: https://doi.org/10.1145/3408066.3408101 into consideration to find the similar user and items. The memory-

9
based can be further classified into user-based approach and item- Table 1 shows the advantages and disadvantages of each
based approach. User-based approach chooses the similar user techniques. None of the techniques is perfect as we can notice that
based on the similarity metrics between users, while item-based each technique has their own advantages and disadvantages. Thus,
approach identifies the user items and compute the predictions choosing the correct techniques is important.
based on the similarity metrics between items. Table 1. Comparison between different techniques
On the other hand, the model-based uses the training data to train
Techniques Advantages Disadvantages
the model. The model will be applied to provide recommendation
once it is trained with proper data. The training process takes most Collaborative No domain knowledge Cold start problems.
of the time to learn the pattern. filtering-based required. Sparsity problems.
Better result when having
In general, CF suffers from cold start problem and data sparsity
huge amount of data.
problems. Cold start problems occur when the number of user in
the system is not enough to predict the similarity user profile. Content- Do not rely on user Overspecialization.
Data sparsity means that user only rate few of the items, which is Based profile. User cold start
hard to predict user preferences. Quality of system problems.
improves over a time.
2.2 Content-Based Recommendation No item cold-start.
Techniques Hybrid-Based Overcome some of the Need to have
issues cause by general knowledge of using
CB recommends items based on user profile such as user
techniques. correct techniques
preferences or user’s historical data to suggest the similar items to
Performance is better. for a particular
the specific user. This technique is useful in text-intensive areas,
domain
which keyword will be used as content. According to a survey by
Beel et al. [13], 55% of the research papers from year 1998 to Knowledge- No cold-start problem. Need of knowledge
2016 touches on the content-based recommendation techniques. based User ratings not required acquisition.
Static
Contrast to CF, CB capable to suggest new item to user. CF is
unable to suggest new item where rating is not available yet. recommendations.
Unlike require item rating as a similarity metric which normally
cannot get in new item. CB tends to give more personalized data. 3. RECOMMENDER SYSTEM IN
However, this resulted CB suffers from overspecialization RETAILER DOMAIN OR E-COMMERCE
problem, which subsequently spike off many new researchers In this section, we will discuss the papers that were published
trying to solve the problem. from year 2017 to year 2019. We will discuss the recommendation
techniques used in the system and provide analysis about these
2.3 Knowledge-Based Recommendation papers. Most of the paper focus on enhancing the current
Techniques techniques such as constructs ontology for the data, combine
The knowledge-based recommendation techniques recommend information of customer review instead of just using product
items based on the domain knowledge [14]. There are basically rating and so on.
three type of knowledge, knowledge about user, knowledge about One of the enhancing methods for CB filtering is to use the details
items and knowledge about relation between user and item [15]. of the customer review. By analyzing the user review, system is
To achieve this, a popular approach which is ontology-based able to extract and build the user preferences model [19, 20]. For
approach has been widely adopted in many research papers [11, example, topic modelling method such as Latent Dirichlet
12, 16, 17]. Ontologies are used to model the user profile, item allocation (LDA) can be used to derive features in the customer
data, and relationship between it. The hierarchical structure allows review [21]. Besides, we can also make use of other sentiment
user to analyze the data at different abstraction levels. Ontology- analysis methods such as Natural Language Toolkit (NLTK) and
based recommender approach overcomes the problems such as part-of-speech (POS) tagging [22]. The NLTK and POS helps to
cold-start, rating sparsity and overspecialization due to the fact identify, or finding entity or category in the review.
that ontology-based recommenders using domain knowledge
rather than user rating [11]. The domain ontologies are use to Having a good data structure do helps in increasing the
calculate the semantic similarity between items and user. performance of the RS. By constructing an ontology-based system,
Ontology structure has been constructed in different aspect [18], it the data is processed and organize into hierarchy level [23-26].
helps to provide a common and shared structure, terminology and Customer and products are presented in nodes and relation
semantics for the item representation. between them as edges. The relation between each nodes can be
easily derive from the structure.
2.4 Hybrid Recommendation Techniques Offline physical shop products usually do not have product rating.
Hybrid recommender techniques combine two or more techniques This make the general collaborating filtering method lost the
in generating the recommendation. It used to overcome the issue feedback input to cluster the customer and item. To overcome the
cause by the specific technique to another such as cold start and situation, user purchase history can be used as an implicit
over specification [14]. It has been proven increasing the feedback [7, 27-32]. We can derive the customer purchase history
performance of the RS. However, choosing the proper techniques and cluster the customer based on pattern or product category.
to combine could be a headache. A good combination will have Customer movement detection such as time in front of products
better accuracy of the recommendation whereas a bad and usually visited area can also act as a feedback to CF [29].
combination will cause the performance of the recommendation
drops. For online e-commerce shops, the product rating is present, but
normally sparse and having cold start problem. By adding user
browsing history and click streaming, combine with the rating, it

10
can provide much more feedback rather than just using product 4. EVALUATION METRICS
ratings [7, 8, 29, 30, 33]. Evaluation metrics is needed when we want to know how the
To improve the performance of the CF RS, choosing and fine- proposed algorithm or method performs. It can be used to
tuning the clustering method is important. The general clustering compare the performance of the proposed algorithm with the
method such as calculating the similarity of the item and user by baseline algorithm. Table 3 shows the summary of the evaluation
using matrix is not efficient and computationally expensive. techniques used in the paper. We will briefly introduce the top 4
Modification to the clustering method can be done in mapping the most used techniques.
out-of-stock items to other similar product to improve the
Table 3. Evaluation techniques and its associated publications
coverage at no accuracy cost by favoring less popular items [34].
Another optimization algorithm such as cultural algorithm, a Evaluation Techniques Publication
knowledge-based evolutionary optimization algorithm can be Accuracy [21, 23, 32]
applied [35]. Singular Value Decomposition (SVD) can be used to F1-Score [20, 31, 33, 38, 39]
reduce the dimensions of the matrix thus reducing the Precision [20, 25, 31, 33, 35, 38, 39]
computational time and data sparsity problems. Another method Recall [20, 21, 24, 25, 31, 33, 35, 38, 39]
to reduce the dimensionality is to apply principal component
analysis (PCA) [36]. It is a statistical process that transforms the
Recall, sometime knows as true positive rate, which indicate how
large dataset into smaller dataset without loosing much
many was correctly classified as positive across all the positive
information. Filtering out fake account and rating will be a good
data. The higher the recall rate, the better the algorithm (see E1).
idea in clustering the customers [37]. Accounts that usually have
high rating or low rating towards specific brands or categories,
will affect the clustering results. E1
Another method to increase the performance of the RS is to
consider the sequence and pattern of the purchase history of TP: True Positive, correctly classified positive items
customer [23, 31, 32]. With the sequence and pattern of the FN: False Negative, wrongly classified negative items
purchase history, we can gather the information or products that
normally will buy together by specific group of people. Sequence Precision indicate that how many was correctly classified as
mining algorithm can be used to find the sequence items and positive across all the classified positive data. The higher the
suggest item to customer [31, 32]. Some system has implemented precision, the better the algorithm (see E2).
a preference decay function to reflect changes in preferences over
time [7].
Model-based CF techniques based on probabilistic model can be E2
enhancing by using the improved Naive Bayes algorithm, which is FP: False Positive, wrongly classified positive items
Naive Bayes algorithm with bigram language model to improve
search query analysis [38]. Association Rule Mining, Bayesian F1-Score, also knows as F-score or F-measure, is the harmonic
Probabilistic Ranking and factorized personalized Markov chains mean of precision and recall, often used when we need the
can also be used to improve the results. balance between recall and precision when there is an uneven
distributed data (See E3).
A hybrid system that can detect the situation such as user cold
start or item cold start and apply suitable algorithm is needed in
order to achieve good result [24, 39, 40]. As we know, most E3
algorithms cannot handle multiple scenario in one algorithm.
Different algorithm and techniques need to apply based on the Accuracy indicate that how often is the classified result are correct
situation to overcome the shortcoming of each algorithm. A across all the data. The higher the accuracy, the better the
hybrid system is not only combining content-based filtering algorithm (see E4).
algorithm and CF algorithm, but also combining two CF
algorithms are consider as a hybrid system. E4
Table 2. Recommendation Techniques and its associated
TN: True Negative, correctly classified negative items
publications
Recommendation Publication 5. DISCUSSION AND
Techniques RECOMMENDATION FOR POTENTIAL
Content-Based Filtering [8, 19, 20, 21, 26]
Collaborative Filtering [27-32, 37,38]
FUTURE WORK
In this paper, we have provided some insight of the RS since year
Hybrid Filtering [7, 22-25, 33-36, 39, 40] 2017. From Table 2, we can notice that more and more researcher
are moving from content-based filtering to CF and hybrid filtering.
From Table 2, we observed that the CB has the least number of The reason may due to the CF having better performance compare
publications. It may be due to the fact that CB has limited to just using content-based filtering, while hybrid filtering method
information to process the data. Current trends show that more overcoming the issue causes by single filtering method thus
researchers are looking into enhancing the CF or apply two or having better performance compare to other recommendation
more methods in the hybrid filtering method. algorithm.
First, using semantic ways to build a customer and product profile.
Traditional way to store and retrieve the massive data take times

11
and hard to maintain. Several methods can be considered such as [2] Schafer, J.B., Konstan, J. and Riedi, J. 1999. Recommender
building an ontological-based customer and product profile. The systems in e-commerce. Proceedings of the 1st ACM
ontological profile has a hierarchy relationship which helps to conference on Electronic commerce - EC 99.
organize the data in a systematic way and easy to interpret. The [3] Bobadilla, J., Ortega, F., Hernando, A. and Alcalá, J. 2011.
relationship between customer and product can be easily Improving collaborative filtering recommender system
determined by the connected edges without the necessary to results and performance using genetic algorithms.
retrieve irrelevant data. This helps to reduce the computational Knowledge-Based Systems. 24, 8, 1310–1316.
power and time. Besides, to form the hierarchy structure, certain
data pre-processing need to be done to ensure the data inserted is [4] Goldberg, D., Nichols, D., Oki, B.M. and Terry, D. 1992.
matching the structure. This helps to organize the data in a more Using collaborative filtering to weave an information
systematic way, and the system can be more scalable. tapestry. Communications of the ACM. 35, 12, 61–70.
http://doi.org/10.1145/138859.138867
Second, to solve the shortcoming of the specific RS algorithms,
we need to implement a system that can take care of these [5] Harper, F.M. and Konstan, J.A. 2016. The MovieLens
situations. For example, cold start, data sparsity and etc. The Datasets. ACM Transactions on Interactive Intelligent
system may need several algorithm to handle different situation, Systems. 5, 4, 1–19. http://doi.org/10.1145/2827872
thus a hybrid RS is suggested. We consider that any system that [6] Gomez-Uribe, C. and Hunt, N. 2016. The Netflix
using two or more algorithm is a hybrid RS. However, combining Recommender System. ACM Transactions on Management
the correct algorithm is a challenge. Throughout the years, hybrid Information Systems 6, 4, 1-19.
system has shown the capability to solve the shortcoming of http://doi.org/10.1145/2843948
different method provided achieving good results in different
[7] Hwangbo, H., Kim, Y. and Cha, K. 2018. Recommendation
situation. For example, we may use the content-base algorithm to
system development for fashion retail e-commerce.
recommend the products when facing user cold start problem as
Electronic Commerce Research and Applications 28, 94-101.
the user do not have any rating feedback regarding the products.
http://doi.org/10.1016/j.elerap.2018.01.012
When time goes on, the system will gather enough feedback and
we may start using the collaborative algorithm. To deal with the [8] Y. Xia, G. D. Fabbrizio, S. Vaibhav, and A. Datta. 2017. A
data sparsity problem, SVD and PCA can be used when Content-based Recommender System for E-commerce Offers
constructing the matrix used for clustering. We can fine-tune the and Coupons. eCOM@SIGIR.
SVD and PCA algorithms depend on our dataset structure to make [9] Pripužić, K., Zarko, I., Podobnik, V., et al. 2013. Building an
it performs better. IPTV VoD recommender system: An experience report.
Lastly, as we are in the retailer and e-commerce domain, we have Proceedings of the 12th International Conference on
to focus on the particular terms such as no product rating for Telecommunications, 155-162.
offline retailer, seasonal products and no stock situation. Previous [10] Bahramian, Z. and Ali Abbaspour, R. 2015. An Ontology-
works has shown that the user purchase history can be used to Based Tourism Recommender System Based On Spreading
interpret as user preferences and user feedback to overcome the no Activation Model. ISPRS - International Archives of the
rating problem. Not only this, the user purchase history and item Photogrammetry, Remote Sensing and Spatial Information
browse history can be combined with the online product rating to Sciences XL-1-W5, 83-90.
produce more accurate suggestion. The system should be able to http://doi.org/10.5194/isprsarchives-xl-1-w5-83-2015
handle seasonal product. For example, product such as Christmas
tree will only be sold on specific day. Customer preference decay [11] Razia Sulthana, A. and Ramasamy, S. 2019. Ontology and
function should also be implemented to reflect changes of context based recommendation system using Neuro-Fuzzy
customer preferences from time to time. Classification. Computers & Electrical Engineering 74, 498-
510. http://doi.org/10.1016/j.compeleceng.2018.01.034
6. CONCLUSION [12] Li, Y., Lin, L. and Ho, C. 2017. A social route recommender
RS plays a big role in retailer sector. It helps to boost the income mechanism for store shopping support. Decision Support
of the company and improve user shopping experience. RS helps Systems 94, 97-108.
user to get interested item intelligently from massive of data. User http://doi.org/10.1016/j.dss.2016.11.004
can get the promotion based on the user preferences or from the
same group of people which having some contrast with each other. [13] Beel, J., Gipp, B., Langer, S. and Breitinger, C. 2015.
There are many researchers working on creating more intelligent Research-paper recommender systems: a literature survey.
and high performance RS. In this paper, we have provided several International Journal on Digital Libraries 17, 4, 305-338.
insights such as the overview of the RS and the techniques and http://doi.org/10.1007/s00799-015-0156-0
evaluation metrics used in recent retailer and e-commerce domain [14] Sharma, R. and Singh, R. 2016. Evolution of Recommender
RS. Systems from Ancient Times to Modern Era: A Survey.
Indian Journal of Science and Technology 9, 20.
7. ACKNOWLEDGMENTS http://doi.org/10.17485/ijst/2016/v9i20/88005
This work is supported by the funding of TM R&D from the
Telekom Malaysia, Malaysia. [15] Tarus, J., Niu, Z. and Mustafa, G. 2017. Knowledge-based
recommendation: a review of ontology-based recommender
8. REFERENCES systems for e-learning. Artificial Intelligence Review 50, 1,
[1] Linden, G., Smith, B. and York, J. 2003. Amazon.com 21-48.
recommendations: item-to-item collaborative filtering. IEEE [16] Colombo-Mendoza, L., Valencia-García, R., Rodríguez-
Internet Computing. 7, 1, 76–80. González, A., Alor-Hernández, G. and Samper-Zapater, J.
2015. RecomMetz: A context-aware knowledge-based

12
 mobile recommender system for movie showtimes. Expert [29] Mettouris, C., Achilleos, A., Kapitsaki, G., and
Systems with Applications 42, 3, 1202-1222. Papadopoulos, G.A. 2018. The UbiCARS Model-Driven
http://doi.org/10.1016/j.eswa.2014.09.016 Framework: Automating Development of Recommender
[17] Obeid, C., Lahoud, I., Khoury, H.E., and Champin, P.-A. Systems for Commerce. Lecture Notes in Computer Science
2018. Ontology-based Recommender System in Higher Ambient Intelligence, 37–53.
Education. Companion of the The Web Conference 2018 on http://doi.org/10.1007/978-3-030-03062-9_3
The Web Conference 2018 - WWW 18. [30] Poulopoulos, D. and Kyriazis, D. 2017. Collaborative
http://doi.org/10.1145/3184558.3191533 Filtering for Producing Recommendations in the Retail
[18] Yanes, N., Sassi, S.B., and Ghezala, H.H.B. 2017. Ontology- Sector. Information Systems Lecture Notes in Business
based recommender system for COTS components. Journal Information Processing: 662–669.
of Systems and Software 132, 283–297. [31] Jia, R., Li, R., Yu, M. and Wang, S. 2017. E-commerce
http://doi.org/10.1016/j.jss.2017.07.031 purchase prediction approach by user behavior data. 2017
[19] Sheikh, A.A., Arif, T. and Malik, M.B. 2018. Framework for International Conference on Computer, Information and
Opinion Based Product Recommender System. National Telecommunication Systems (CITS).
Conference on Recent Advances in Computer Science and IT [32] Saini, S., Saumya, S., and Singh, J.P. 2017. Sequential
(NCRACIT) International Journal of Scientific Research in Purchase Recommendation System for E-Commerce Sites.
Computer Science, Engineering and Information Technology Computer Information Systems and Industrial Management
4, 1: 17-21 Lecture Notes in Computer Science, 366–375.
[20] Osman, N.A. 2019. Contextual Sentiment Based http://doi.org/10.1007/978-3-319-59105-6_31
Recommender System to Provide Recommendation in the [33] Nilashi, M., Ibrahim, O., and Bagherifard, K. 2018. A
Electronic Products Domain. International Journal of recommender system based on collaborative filtering using
Machine Learning and Computing 9, 4 (2019), 425–431. ontology and dimensionality reduction techniques. Expert
[21] Lin, K.-P., Shen, C.-Y., Chang, T.-L., and Chang, T.-M. Systems with Applications 92, 507–520.
2017. A Consumer Review-Driven Recommender Service http://doi.org/10.1016/j.eswa.2017.09.058
for Web E-Commerce. 2017 IEEE 10th Conference on [34] Prévost, B., Janssen, J.L., Camacaro, J.R., and Bessega, C.
Service-Oriented Computing and Applications (SOCA). 2018. Deep inventory time translation to improve
http://doi.org/10.1109/soca.2017.35 recommendations for real-world retail. Proceedings of the
[22] Jing, N., Jiang, T., Du, T. and Sugumaran, V. 2017. 12th ACM Conference on Recommender Systems.
Personalized recommendation based on customer preference [35] Selvarajah, K., Kobti, Z., and Kargar, M. 2019. A Cultural
mining and sentiment assessment from a Chinese e- Algorithm for Determining Similarity Values Between Users
commerce website. Electronic Commerce Research 18, 1: in Recommender Systems. Applications of Evolutionary
159–179. Computation Lecture Notes in Computer Science: 270–283.
http://doi.org/10.1007/s10660-017-9275-6 [36] Chu, P.-M. and Lee, S.-J. 2017. A novel recommender
[23] Kouki, P., Fountalis, I., Vasiloglou, N., et al. 2019. Product system for E-commerce. 2017 10th International Congress on
collection recommendation in online retail. Proceedings of Image and Signal Processing, BioMedical Engineering and
the 13th ACM Conference on Recommender Systems. Informatics (CISP-BMEI).
http://doi.org/10.1145/3298689.3347003 http://doi.org/10.1109/cisp-bmei.2017.8302310
[24] Ding, L., Han, B., Wang, S., Li, X., and Song, B. 2017. User- [37] Ramesh, B. and Reeba, R. 2017. Secure recommendation
centered recommendation using US-ELM based on dynamic system for E-commerce website. 2017 International
graph model in E-commerce. International Journal of Conference on Circuit, Power and Computing Technologies
Machine Learning and Cybernetics (IJMLC)10, 4: 693–703. (ICCPCT).
[25] Agarwal, P., Vempati, S. and Borar, S. 2018. Personalizing [38] Gaikwad, R., Udmale, S. and Sambhe, V., 2017. E-
Similar Product Recommendations in Fashion E-commerce. commerce Recommendation System Using Improved
arXiv preprint arXiv:1806.11371. Probabilistic Model. Information and Communication
[26] Shaikh, S., Rathi, S., and Janrao, P. 2017. Recommendation Technology for Sustainable Development, pp.277-284.
System in E-Commerce Websites: A Graph Based [39] Hanke, J., Hauser, M., Dürr, A. and Thiesse, F. 2018.
Approached. 2017 IEEE 7th International Advance Redefining the Offline Retail Experience: Designing Product
Computing Conference (IACC). Recommendation Systems for Fashion Stores. ECIS.
http://doi.org/10.1109/iacc.2017.0189 [40] Wang, F., Wen, Y., Guo, T., Chen, J. and Cao, B. 2018.
[27] Luo, F., Ranzi, G., Wang, X. and Dong, Z. 2019. Social Personalized Commodity Recommendations of Retail
Information Filtering-Based Electricity Retail Plan Business Using User Feature Based Collaborative Filtering.
Recommender System for Smart Grid End Users. IEEE 2018 IEEE Intl Conf on Parallel & Distributed Processing
Transactions on Smart Grid 10, 1, 95-104. with Applications, Ubiquitous Computing &
[28] Dimyati, H. and Agasi, R. 2018. Collaborative Filtering in an Communications, Big Data & Cloud Computing, Social
Offline Setting Case Study: Indonesia Retail Business. Computing & Networking, Sustainable Computing &
Communications in Computer and Information Science Data Communications.
Mining, 223–232.

13