Twittok TikTok Shop Sentiment Analysis

Overview

A hackathon project that surfaces aggregated sentiment signals from TikTok Shop reviews, giving sellers a dashboard view of customer perception across their product catalogue.

Technical Challenges

Sentiment Pipeline

Review text is passed through a pre-trained transformer-based sentiment classifier (positive / neutral / negative). Chose a lightweight distilled model to keep inference latency under 200 ms per review without requiring a GPU-backed server within the hackathon's deployment constraints.

Data Aggregation

Sentiment scores are bucketed per product and time-windowed (daily/weekly) so sellers can spot trend reversals rather than just point-in-time scores. Aggregation is computed on the fly with in-memory reducers given the limited dataset size in the hackathon scope.

Frontend

Built with React using a component-per-chart structure. Kept state local to each chart component to avoid prop-drilling; no global store was warranted for a single-page dashboard of this size.