Vol.:(0123456789)
SN Computer Science (2021) 2:2
https://doi.org/10.1007/s42979-020-00351-4
SN Computer Science
ORIGINAL RESEARCH
A New Approach Item Rating Data Mining on the Recommendation
System
Anh Nguyen Thi Dieu
1
· Thanh Nguyen Vu
1
· Tuan Dinh Le
2
Received: 16 April 2020 / Accepted: 28 September 2020 / Published online: 7 November 2020
© Springer Nature Singapore Pte Ltd 2020
Abstract
Collaborative fltering (CF) in the recommendation system using user habits, behaviors, and item rating to recommend the
products which suit customer’s needs. Therefore, analyzing user rating data is one of the factors that improve the efciency
of the recommendation system. This paper proposes a new approach to analyze rating item and input the implicit efect of
items rating to the recommendation system based on the TrustSVD model and matrix factorization (MF) techniques. The
experimental results showed that our proposed solution achieves 18% better than the matrix factorization method and 15%
the Multi-Relational Matrix Factorization method, respectively.
Keywords Recommendation system · Collaborative fltering · Implicit efect · Matrix factorizations · Matrix user · Trust-
based recommender · Linear regression · Machine learning
Introduction
Recommendation systems have been using in many applica-
tion scenarios. For example, in e-commerce, the recommen-
dation system analysis based on user’s interests, searching
keywords, product reviews to make recommendations for
users. Currently, three methods, which have been widely
used in a recommendation system, are content fltering, col-
laborative fltering, and hybrid.
Content fltering (CF) based on purchase history, views
user information, thereby suggesting products with content
similar to buyers’ needs. Some popular techniques currently
used for content fltering are Bag of the word, TF-IDF (term
frequency-inverse document frequency), Graph, Grid, TF
(term frequency), VSM (Vector Space Model) [2].
Collaborative fltering is a technique that determines a
user’s interest in a new product based on previous products
they rate, recommending similar products with consumer
appreciation. The recommended system in this approach
identifes the similarity of the objects through adjacent
measurements. Current techniques for collaborative flter-
ing are: Pearson correlation (CORR), Cosine (COS), Adjust
Cosine (ACOS), Constrained Correlation (CCORR), Mean
square Diference (MSD), Euclidean (EUC), and SM SING
(singularities) [2, 4].
Hybrid methods are a combination of content fltering
and collaborative fltering, relying on the advantages of one
technique to overcome the disadvantages of the other. For
example, collaborative fltering has a problem with a cold
start, which is challenging to suggest for items that do not
have a rating, while the content-based approach can sim-
ply do it when the prediction for new items based on user
descriptions is available and straightforward.
The rating of the recommendation system is usually an
integer value from 1 to 5. In the same evaluation index, the
criteria for user ratings are diferent. For example, one hotel
is rated 4* by both guest A and guest B. This rating consists
of clean, service, attached utilities, location. Guest A may be
more satisfed with the service in a hotel, but guest B likes
the hotel’s utilities. For simplicity of calculation, this infor-
mation is often missed from the recommendation system.
* Thanh Nguyen Vu
nguyenvt@vhu.edu.vn
Anh Nguyen Thi Dieu
anhntdgm@gmail.com
Tuan Dinh Le
le.tuan@daihoclongan.edu.vn
1
Van Hien University, Ho Chi Minh City, Viet Nam
2
Long An University of Economics and Industry,
Long An province, Viet Nam