Analisis Temporal Topic Modeling Untuk Identifikasi Isu Teknis Pada Aplikasi Superapp Perbankan Indonesia

Salwa, Rabithah Zahiratus (2026) Analisis Temporal Topic Modeling Untuk Identifikasi Isu Teknis Pada Aplikasi Superapp Perbankan Indonesia. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221078-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Perpindahan layanan perbankan ke dalam satu aplikasi superapp menambah kompleksitas sistem sekaligus memperbesar peluang munculnya gangguan teknis, terutama pada tahun pertama setelah peluncuran. Pengguna merekam gangguan tersebut melalui ulasan di toko aplikasi, tetapi analisis ulasan pada umumnya memotret keseluruhan periode sekaligus sehingga tidak menunjukkan kapan suatu keluhan muncul, memuncak, atau mereda. Penelitian ini menelusuri pola temporal keluhan pengguna pada wondr by BNI dan BYOND by BSI selama dua belas bulan pertama pasca-peluncuran. Ulasan dengan rating 1 dan 2 dari Google Play Store digunakan sebagai proksi keluhan teknis dan diuji keabsahannya melalui validasi manual atas sampel berstrata. Topik keluhan diekstraksi menggunakan BERTopic dengan representasi makna (embedding) IndoBERT-base-p2 yang dibangun terpisah untuk setiap aplikasi. Perubahan proporsi topik antarbulan ditelusuri melalui dynamic topic modeling, konsistensi isi topik diukur melalui kemiripan kosinus pada dimensi semantik dan leksikal, sedangkan arah trennya diuji menggunakan uji Mann-Kendall termodifikasi Hamed-Rao beserta Sen's slope dan koreksi Benjamini-Hochberg. Pemodelan menghasilkan 13 topik pada wondr (Cᵥ = 0,6964) dan 14 topik pada BYOND (Cᵥ = 0,6762), dengan validitas proksi terbobot 94,17% dan 90,85%. Keluhan wondr memuncak sekali pada bulan kelima, sedangkan BYOND memperlihatkan tiga lonjakan dengan puncak utama pada bulan keempat. Lima topik meningkat secara signifikan pada masing-masing aplikasi, tetapi hanya BYOND yang memiliki topik menurun signifikan, yaitu empat topik. Perbandingan lintas aplikasi menemukan enam pasangan isu sistemik: kedua aplikasi konvergen dalam keberadaan persoalan mendasarnya, namun divergen dalam laju penanganannya, paling jelas pada friksi biometrik yang mereda pada BYOND tetapi bertahan pada wondr. Temuan ini menyediakan dasar empiris bagi pengembang untuk menyusun prioritas penanganan keluhan sepanjang siklus pasca-peluncuran.
===================================================================================================================================
The migration of banking services into a single superapp adds system complexity and widens the room for technical disruption, especially during the first year after launch. Users record these disruptions in app store reviews, yet most review analyses capture the whole observation period at once and therefore cannot show when a complaint appears, peaks, or subsides. This study traces the temporal patterns of user complaints on wondr by BNI and BYOND by BSI across the first twelve months after launch. One- and two-star reviews from the Google Play Store serve as a proxy for technical complaints, and the proxy is verified through manual validation on a stratified sample. Complaint topics are extracted using BERTopic with meaning representations (embeddings) from IndoBERT-base-p2, fitted separately for each application. Month-to-month changes in topic proportion are traced through dynamic topic modeling, the consistency of topic content is measured by cosine similarity on the semantic and lexical dimensions, and the direction of each trend is tested using the Hamed-Rao modified Mann-Kendall test together with Sen's slope and the Benjamini-Hochberg correction. The modeling yields 13 topics for wondr (Cᵥ = 0.6964) and 14 topics for BYOND (Cᵥ = 0.6762), with weighted proxy validity of 94.17% and 90.85%. Complaints on wondr peak once in the fifth month, whereas BYOND shows three surges with its main peak in the fourth month. Five topics rise significantly in each application, but only BYOND has significantly declining topics, namely four. The cross-application comparison identifies six systemic issue pairs: both applications converge in the presence of the underlying problems yet diverge in how quickly they are resolved, most clearly in biometric friction that eases on BYOND but persists on wondr. These findings offer an empirical basis for developers to prioritize complaint handling throughout the post-launch cycle.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	BERTopic, Keluhan pengguna, Pemodelan topik temporal, Superapp perbankan, Uji tren, Banking superapp, BERTopic, Temporal topic modeling, Trend testing, User complaints
Subjects:	Q Science Q Science > QA Mathematics Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QA Mathematics > QA76.9.D343 Data mining. Querying (Computer science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Rabithah Zahiratus Salwa
Date Deposited:	04 Aug 2026 05:41
Last Modified:	04 Aug 2026 05:41
URI:	http://repository.its.ac.id/id/eprint/143099

Actions (login required)

View Item