| dc.description.abstract | The more data, the less work we must do since machines are capable of dealing with those
complicated and heavy loads of data. However, when it comes to businesses, the more seems to be
the less since customers are not fond of overwhelming options. Therefore, multiple attempts on
recommendation systems have been delivered, yet just a few utilized and realistic ones be
genuinely applied. There are two noticeable techniques in the field to be mentioned. The first
candidate is Topic Modelling, particularly the Linear Dirichlet Allocation Model (LDA), which is
one of the “warriors” in Natural Language Processing [1]. As a recommender engine, it usually
takes customer reviews as input, outputs transparent classifications/ or opaque groups of topics that
customers belong to, and then recommend products of users in similar group. Nevertheless, this
method is not explicitly built for recommendation but more on grouping users with same
preferences. The other method is the application of neural networks as a tool for capturing users’ s
preferences. A particular figure of this is Time interval awareness Self-attention- based sequential
recommendation (TiSASRec) [2]. The main idea of TiSASRec is to make use of time intervals
between user interactions in accordance with their sequential item frames for recommendation.
However, the items chosen for ranking are unqualified. As illustrated, each aforementioned model
is built individually and lack of completion. This research is expected to incorporate full advantages
of each model to build a complete recommendation mechanism. | en_US |
