Abstract:
Product recommendation systems use ratings from users to rank
products. There’s a gap between analyzing user ratings and
reviews in purchase decision making process. 97% people read
reviews to confirm product quality and trust reviews than ratings.
This research proposes a data mining and machine learning model
to rank products based on textual reviews. When considering
methodology and design of this study, a survey was conducted and
outcomes show a needfulness of using reviews in ranking. A model
to rank products considering review sentiment polarity is
proposed and implemented using Python programming language.
Well-structured unique workflow of data pre-processing,
sentiment-polarity estimation, algorithm training for high
accuracy, best algorithm selection, value prediction and calculation
for ranking products for recommendation is used. As the results,
Survey indicates that 98.8% of people read reviews though the
star rating is presented. 85.8% say they trust this kind of system
more. Among four algorithms, K-Neighbors algorithm was
proposed as best performing algorithm for value prediction for
this type of research. Products were successfully ranked based on
sentiment score. Most of existing researches are proposal
researches while this is an implementation research. Proposed
algorithm and model with high accuracy can use as base for future
researches. Illustrated Python implementation method also can be
used for future work. There are some practical implications as
Fake review generation can mislead the outcomes and reviews
from other languages, rather than English, will not be considered
for calculations and have to train and create lexical database for
other languages.
