Explainable Transformer-based Severity Classification of Indonesian Public Service Complaints

Public service reviews mix minor interface friction with financial loss and blocked access to essential services.This study defines severity as reported service consequence and evaluates an Indonesian human triage framework.The corpus contains 1,030 complaints labelled Low, Medium, or High.The original 898 complaints formed fixed training, validation, and test partitions.A further 132 complaints, including 102 High cases, formed an independent challenge set.A calibrated support vector machine and five transformer conditions were compared across three seeds.Validation selected focal loss with IndoBERTweet.It improved class balanced test metrics and High recall over the lexical baseline but missed more than half of High complaints on the challenge set.Repeated local explanations from a reproduced checkpoint were stable for nine of ten selected cases.The results support an auditable human review queue, not autonomous deployment, and demonstrate the value and limitations of consequence based severity modelling.

Authors

Institutions

Publication Details

Journal
International journal of intelligent engineering and systems
Published
2026-11-17
DOI
https://doi.org/10.22266/ijies2026.1231.03
Primary Topic
Text and Document Classification Technologies
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Explainable Transformer-based Severity Classification of Indonesian Public Service Complaints

Deffa Rahadiyan, Délcio Silva
1 links
International journal of intelligent engineering and systems
Text and Document Classification Technologies
article

Explainable Transformer-based Severity Classification of Indonesian Public Service Complaints

Deffa Rahadiyan, Délcio Silva
article en

Abstract

Public service reviews mix minor interface friction with financial loss and blocked access to essential services.This study defines severity as reported service consequence and evaluates an Indonesian human triage framework.The corpus contains 1,030 complaints labelled Low, Medium, or High.The original 898 complaints formed fixed training, validation, and test partitions.A further 132 complaints, including 102 High cases, formed an independent challenge set.A calibrated support vector machine and five transformer conditions were compared across three seeds.Validation selected focal loss with IndoBERTweet.It improved class balanced test metrics and High recall over the lexical baseline but missed more than half of High complaints on the challenge set.Repeated local explanations from a reproduced checkpoint were stable for nine of ten selected cases.The results support an auditable human review queue, not autonomous deployment, and demonstrate the value and limitations of consequence based severity modelling.

International journal of intelligent engineering and systemsVol. 19(12)
President University (ID)
Text and Document Classification Technologies
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon