Academic coursework · Data Mining / Project 07
Computer Science · Ton Duc Thang University
CausaSent
Turning Vietnamese product reviews into an explorable analysis dashboard
Project focus
A review-analysis interface that turns a long-running model pipeline into visible progress, charts, priorities, and searchable evidence. Its frontend connects text or CSV input with streamed results and several levels of detail.
- Dashboard UI
- Streaming feedback
- Data visualization
Watch the project
- Context
- Data Mining · Course project
- Project status
- Completed
- Tools & technologies
- Next.js · TypeScript · Tailwind CSS · Motion · Server-Sent Events · FastAPI
A flow in motion
A review becomes a clearer signal.
Follow review input through visible processing, aspect summaries, and a priority traced back to its evidence.
Three sample reviews
Ready to analyze
Aspect signals
Begin with the customer’s words.
Text or CSV input becomes a review batch. This illustration uses three deliberately simple sample comments.
Plays while in view. Pause or choose a stage to explore at your own pace. A conceptual visualization using sample data, independent of the original application.
01 / The project
The project
in context.
CausaSent analyzes Vietnamese e-commerce reviews and groups their aspects and sentiment into a seller-facing summary. The frontend turns a model pipeline into an experience people can follow: submit reviews, see analysis progress, inspect the patterns, and explore recommended actions. The course project includes the analysis implementation, evaluation artifacts, and a Next.js dashboard.
The user flow
- 01Paste reviews or upload a CSV
- 02Start analysis
- 03Follow streamed progress
- 04Explore overview and aspects
- 05Inspect suggested actions
- 06Search the underlying reviews
02 / The making
What went
into it.
A clear start to a complex process
The input screen accepts review text or a CSV, exposes API readiness, and offers sample data. Review counts and processing limits help explain what will be analyzed before the user starts.
Progress that stays visible
Server-Sent Events feed processing counts, partial summaries, and review results into the dashboard. Starting, processing, completion, and stream-error states make a long-running task visible instead of leaving the user with a static loading screen.
Several levels of detail
Overview, Actions, Aspects, and Reviews tabs separate the main reading tasks. KPI summaries, a sentiment donut, aspect radar, breakdown bars, and priority distributions provide different views of the same review batch.
Trace a summary back to its reviews
The reviews table supports text search, sentiment-related filters, and expandable rows. Action cards and aspect summaries sit alongside the detailed evidence so a user can move from a broad pattern to the source comments.
Results that remain useful
Stored result state supports revisiting the dashboard, while its print/export action makes the analysis shareable. Sample-result exploration is available separately from running a new analysis.