DEVELOPMENT OF A SENTIMENT ANALYSIS DASHBOARD FOR TOKOPEDIA USING MACHINE LEARNING AND COMPETITIVE INTELLIGENCE APPROACH
As e-commerce competition increases, e-commerce needs a system to constantly monitor and gain feedback from the market to gain and maintain competitive advantage. This research proposes a competitive intelligence (CI) system to facilitate e-commerce with tools to monitor their competitor and to h...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Project |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/86230 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | As e-commerce competition increases, e-commerce needs a system to constantly
monitor and gain feedback from the market to gain and maintain competitive advantage.
This research proposes a competitive intelligence (CI) system to facilitate e-commerce with
tools to monitor their competitor and to help them in identifying problems. The CI system
consists of a sentiment analysis model to process the user reviews and a dashboard to display
insights of the processed data. The proposed competitive intelligence system utilizes Google
Play Store user reviews from mobile applications of major e-commerce platforms in
Indonesia as its primary data. Using Python library scraper, data such as user review,
timestamps, and star rating are collected from Tokopedia, Shopee, Bukalapak, Lazada, and
Blibli. After 50,000 user reviews are collected from each e-commerce, it is sampled and
processed to be used for training the sentiment analysis model. The competitive intelligence
system incorporates a machine learning-based sentiment analysis framework, utilizing
advanced natural language processing models such as Bidirectional Encoder
Representations from Transformers (BERT) to identify aspects (subject or features being
discussed) within reviews along with identifying its sentiment. The training of the BERT
model was optimized with Optuna, a software framework that automates the process of
searching the best training hyperparameters to ensure the best model is generated. The
sentiment analysis model performed well in aspect term extraction (ATE) and aspect polarity
classification (APC) tasks. The final generated model achieved an F1 score of 85.04% for
APC and 69.55% for ATE. The sentiment analysis result is combined with the original user
reviews before passed into the dashboard. The dashboard is built using Power BI due to its
ease of use and modular nature. The insights found from processed reviews such as sentiment
trends, average monthly star ratings, and most mentioned aspects was presented through a
dashboard. This system can automate user review processing and provide actionable insights
for e-commerce. The created dashboard can offer insights from collected user reviews while
also acting as a tool to monitor other competitors.
|
|---|