Gigalyze: Aspect-Based Sentiment Analysis of Gig Economy Platform Services Using Support Vector Machines

Abstract

The expansion of digital freelancing has increased the volume of user feedback associated with gig economy platforms, making manual interpretation slow and inconsistent. This paper presents Gigalyze, a web-based aspect-based sentiment analysis system for Google Play Store reviews of Fiverr, Freelancer, and Upwork. The dataset contains 30,000 English-language reviews, comprising 10,000 reviews per platform, collected for the Malaysian market between 2022 and 2025. Reviews were cleaned, spell-corrected, normalised, lemmatised, and represented using term frequency-inverse document frequency features. Non-negative matrix factorisation supported the derivation of four platform-service aspects: User Experience, App Performance and Functionality, Account and Support Issues, and Payment and Earnings. VADER compound scores supplied initial positive, neutral, and negative sentiment labels. Separate linear Support Vector Machine classifiers were developed for sentiment and aspect prediction, while the Synthetic Minority Over-sampling Technique was applied to the training data to reduce class imbalance. After balancing, sentiment accuracy reached 83.74%, neutral recall increased from 0.26 to 0.49, and macro F1-score increased from 0.68 to 0.73. Aspect accuracy reached 88.52%, with User Experience recall improving from 0.86 to 0.88. All 14 functional test cases passed. These results demonstrate the feasibility of interpretable machine-learning analytics for gig-platform reviews while also highlighting limitations stemming from automatically generated labels, overlapping aspect vocabularies, and subtle emotional expressions.