Pembentukan Log Data berbasis Graf Yang Mengandung Invisible Task Berdasarkan Data Tekstual

Yuseph, Daffa Saskara (2025) Pembentukan Log Data berbasis Graf Yang Mengandung Invisible Task Berdasarkan Data Tekstual. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211249-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Penambangan proses (process mining) membutuhkan log data sebagai input utama untuk analisis dan pemodelan proses bisnis. Namun, informasi proses bisnis kerap tersedia dalam bentuk deskripsi tekstual. Penelitian yang dilakukan sebelumnya berhasil menghasilkan model proses BPMN dari data tekstual namun tidak memungkinkan pembangunan model proses secara otomatis dari log data. Metode process discovery memerlukan log data berbasis graf sebagai input dan tidak dapat memproses data tekstual secara langsung. Penelitian ini mengusulkan metodologi pembentukan log data berbasis graf dari data tekstual. Penelitian ini mengusulkan metodologi end-to-end untuk menghasilkan log data dari deskripsi tekstual proses bisnis menggunakan pendekatan Large Language Models (LLM). Metodologi terdiri dari tiga tahap utama: (1) Fact Extraction menggunakan GPT-4.1 dengan chain-of-thought prompting untuk mengekstraksi Binary Fact Types (BFT) dan Characteristics (C) dari kalimat dan memberikan label, (2) BPMN-styled JSON Generation menggunakan analisis pola struktural dan LLM generation untuk menghasilkan struktur BPMN dalam json yang menangani relasi sekuensial, paralel, kondisional, dan loop, dan (3) Event Log Generation dengan penerapan aturan transformasi struktur dan penyisipan Invisible Tasks (IT) untuk memastikan kompatibilitas dengan algoritma process mining. Evaluasi dilakukan pada 10 kasus studi, dengan 3 kasus studi memiliki proses bisnis yang membutuhkan invisible task dalam pemodelannya, dan 7 kasus studi yang tidak membutuhkan invisible task dalam pemodelannya. Tahapan 1 fact extraction dan tahapan 2 BPMN-styled json generation dievaluasi menggunakan MMRE hasil penelitian ini terhadap penelitian sebelumnya untuk 10 kasus studi. Tahapan 3 log data generation and invisible task insertion dievaluasi berdasarkan akurasi struktural hasil penelitian ini dibandingkan dengan hasil model proses menggunakan log data yang tidak disisipkan invisible task pada 3 kasus studi yang membutuhkan invisible task dalam pemodelannya, yaitu kasus studi 04, 06, dan 10. Hasil evaluasi menunjukkan tahapan 1 fact extraction menghasilkan 4,17% MMRE dan tahapan 2 BPMN-styled json generation menghasilkan 4,05% MMRE. Kemudian, hasil evaluasi juga menunjukkan tahapan 3 pada penelitian ini mendapatkan 100% kemiripan struktural, membuktikan bahwa log data yang dihasilkan oleh penelitian ini dapat dibentuk menjadi model proses graf akurat yang mengandung invisible task.
======================================================================================================================================
Process mining requires event logs as primary input for business process analysis and modeling. However, business process information is often available in textual descriptions. Previous research successfully generated BPMN process models from textual data but did not enable automatic process model construction from event logs. Process discovery methods require graph-based event logs as input and cannot directly process textual data. This research proposes a methodology for generating graph-based event logs from textual descriptions. This study presents an end-to-end methodology for generating event logs from textual business process descriptions using Large Language Models (LLM) approach. The methodology consists of three main stages: (1) Fact Extraction using GPT-4.1 with chain-of-thought prompting to extract Binary Fact Types (BFT) and Characteristics (C) from sentences and provide labels, (2) BPMN-styled JSON Generation using structural pattern analysis and LLM generation to produce BPMN structures in JSON format that handle sequential, parallel, conditional, and loop relationships, and (3) Event Log Generation with application of structural transformation rules and Invisible Tasks (IT) insertion to ensure compatibility with process mining algorithms. Evaluation was conducted on 10 case studies, with 3 case studies containing business processes requiring invisible tasks in their modeling, and 7 case studies not requiring invisible tasks in their modeling. Stage 1 fact extraction and Stage 2 BPMN-styled JSON generation were evaluated using MMRE of this research results against previous research for 10 case studies. Stage 3 log data generation and invisible task insertion were evaluated based on structural accuracy of this research results compared to process model results using event logs without invisible task insertion on 3 case studies requiring invisible tasks in their modeling, namely case studies 04, 06, and 10. Evaluation results show that Stage 1 fact extraction achieved 4.17% MMRE and Stage 2 BPMN-styled JSON generation achieved 4.05% MMRE. Furthermore, evaluation results also demonstrate that Stage 3 in this research achieved 100% structural similarity, proving that the event logs generated by this research can be formed into accurate graph process models containing invisible tasks.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Fact Extraction, Invisible Tasks, Large Language Models, Log Data, Model Proses Berbasis Graf, Process Mining. Event Log, Fact Extraction, Graph Model Process, Invisible Tasks, Large Language Models, Process Mining
Subjects:	Q Science > QA Mathematics > QA76 Computer software Q Science > QA Mathematics > QA76.9.D343 Data mining. Querying (Computer science) T Technology > T Technology (General) > T385 Visualization--Technique T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Daffa Saskara
Date Deposited:	31 Jul 2025 08:43
Last Modified:	31 Jul 2025 08:43
URI:	http://repository.its.ac.id/id/eprint/124478

Actions (login required)

View Item