Model Perilaku Knowledge Sharing Di Media Sosial

Yuliazmi, Yuliazmi (2026) Model Perilaku Knowledge Sharing Di Media Sosial. Doctoral thesis, Institut Teknologi Sepuluh Nopember.

Text 7022201021-Doctoral.pdf - Accepted Version Restricted to Repository staff only Download (20MB) | Request a copy

Abstract

Berbagi pengetahuan (knowledge sharing, KS) di media sosial dapat dipandang sebagai proses berlapis. Niat pengguna untuk berbagi mendahului perwujudan niat tersebut dalam teks unggahan, yang pada gilirannya memicu respons audiens berupa engagement. Penelitian-penelitian yang ada umumnya menggarap satu lapisan saja dan jarang menjembatani ketiganya dalam satu kerangka. Disertasi ini menawarkan kerangka perilaku KS terpadu untuk menyambungkan ketiga lapisan tersebut ke dalam satu model komputasional bertahap. Tiga model empiris dibangun dari pool responden survei OSMS (Online Social Media Survey) yang sama. Model pertama mengklasifikasikan konten KS pada 1.841 unggahan Facebook dari 81 responden dengan fitur berbasis Theory of Planned Behavior (TPB). Random Forest mencapai AUC 0,78, dan lima dari enam hipotesis TPB diterima. Agregasi prediksi pada level pengguna dipakai sebagai proksi kecenderungan berbagi. Model kedua mengelompokkan 10.963 unggahan Twitter/X berbahasa Indonesia ke dalam tiga kelas (general, complaint, inquiry). LinearSVM-balanced dengan TF-IDF bigram menghasilkan Macro-F1 0,75. Kalibrasi sigmoid (Platt scaling) kemudian menekan Expected Calibration Error (ECE) dari 0,13 ke 0,03 tanpa menurunkan akurasi. Model ketiga memprediksi engagement pada 7.877 unggahan Facebook dari 184 responden dengan membandingkan empat metode regresi (Regresi Polinomial, SVR, ELM,danPSO-ELM). PSO-ELM(Particle Swarm Optimization–Extreme Learning Machine) menurunkan MSE prediksi likes sekitar 35% dibandingkan ELM standar, sedangkan Regresi Polinomial mencatat galat terendah secara keseluruhan. Ketiga model dirangkai dalam kerangka perilaku KS terpadu menjadi vektor profil tujuh dimensi per pengguna: proksi kecenderungan berbagi, distribusi tiga tipe teks, dua rata-rata metrik engagement, dan rasio komentar terhadap likes. Dari kombinasi komponen profil tersebut, diusulkan empat tipologi pengguna: Knowledge Broadcaster, Community Builder, Passive Sharer, dan Reactive Sharer. Kerangka ini terbuka untuk dipakai pada analitik media sosial dan manajemen pengetahuan organisasi guna membaca pola berbagi pengetahuan pengguna.
======================================================================================================================================
Knowledge sharing (KS) on social media can be viewed as a layered process. A user’s intention to share precedes the manifestation of that intention in posted text, which in turn triggers audience response in the form of engagement. Existing studies typically engage with only one layer and seldom join the three within a single framework. This dissertation offers an integrated KS behavior framework that links the three layers into one multi-stage computational model. Three empirical models were developed from the same pool of respondents in the Online Social Media Survey (OSMS). The first model classifies KS content in 1,841 Facebook posts from 81 respondents using features grounded in the Theory of Planned Behavior (TPB). Random Forest reached an AUC of 0.78, and five of six TPB hypotheses were supported. User-level aggregation of post predictions then serves as a proxy for the user’s sharing tendency. The second model groups 10,963 Indonesian Twitter/X posts into three classes (general, complaint, inquiry). LinearSVM-balanced with TF-IDF bigram features attained a Macro F1 of 0.75. Sigmoid (Platt) calibration subsequently reduced the Expected Calibration Error (ECE) from 0.13 to 0.03 without sacrificing accuracy. The third model predicts engagement on 7,877 Facebook posts from 184 respondents by comparing four regression methods (Polynomial Regression, SVR, ELM, and PSO-ELM). The PSO-ELM (Particle Swarm Optimization based Extreme Learning Machine) lowered the MSE for likes prediction by about 35% relative to standard ELM, whereas Polynomial Regression recorded the lowest overall error. The three models are linked through an integrated KS behavior framework that produces a seven-dimensional user profile vector: a sharing-tendency proxy, the distribution over three text types, two average engagement metrics, and a comment-to-likes ratio. From combinations of these components, four user typologies are proposed: Knowledge Broadcaster, Community Builder, Passive Sharer, and Reactive Sharer. The framework is open to use in social media analytics and organizational knowledge management for reading users’ knowledge-sharing patterns.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	knowledge sharing, theory of planned behavior, klasifikasi teks, kalibrasi probabilitas, prediksi engagement, profil pengguna, knowledge sharing, theory of planned behavior, text classification, probability calibration, engagement prediction, user profiling
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20001-(S3) PhD Thesis
Depositing User:	Yuliazmi Yuliazmi
Date Deposited:	07 Aug 2026 03:43
Last Modified:	07 Aug 2026 03:43
URI:	http://repository.its.ac.id/id/eprint/144209

Actions (login required)

View Item