Malaysian Chinese folk beliefs on Facebook based on LDA topic modelling

The study investigates how Malaysian Chinese folk beliefs are constructed, discussed, and disseminated on Facebook, addressing a gap in prior scholarship which has focused on historical documentation, fieldwork, and digitisation projects but not on social media discourse using NLP. Facebook, the most widely used social platform in Malaysia, provides abundant text posts that transcend regional boundaries. The research aims to identify latent content themes and characterise the nature of Malaysian Chinese folk beliefs online by applying latent Dirichlet allocation (LDA) to Facebook posts and interpreting results within a framework that integrates social construction theory and media ecology. This approach considers the interaction among constructors (practitioners, temple organisations, media organisations, and merchants), the platform, and content to understand the significance and real-life implications of these beliefs in the digital era.

Research on Malaysian Chinese folk beliefs has evolved over time. From the 1930s–1950s, scholars used historical documentation to examine deities and temple histories. In the 1960s–1970s, fieldwork became prominent, mapping temples and practices. The 1980s–1990s saw focused case studies (e.g., Dejiao, Nine Emperor Gods, I-Kuan Tao) and extensive collection of inscriptions and historical materials. In the early 21st century, studies combined historical documents with field research to explore dissemination networks and immigrant interactions. Recently, digital humanities and GIS-based projects mapped Chinese associations, temples, and cemeteries in Malaysia and Singapore, advancing interdisciplinary approaches. Despite this progression, the social media dimension of Malaysian Chinese folk beliefs and the application of NLP methods such as LDA had not been addressed, motivating the present study.

Data collection proceeded in four stages: keyword collection, data collection, data preparation, and data analysis. To operationalise 'Malaysian Chinese folk beliefs', the study used deity names as search terms. From angkongkeng.com (a directory of ~2000 Malaysian Chinese temples), researchers compiled 191 main patron deity names (Supplementary Table S1). Between January 22 and February 14, 2023 (the Spring Festival period), two researchers queried Facebook with these keywords, manually collecting up to the first 25 text posts per deity without skipping to balance representation. Some deities were highly active; others had little content. This yielded 4358 posts. Data pre-processing in Python removed duplicates, invalid entries, HTML tags, special characters, punctuation, irrelevant and redundant data, resulting in 4012 posts. For Chinese NLP, Jieba was used for word segmentation; the Baidu Chinese Stop Word List (2012) removed stop words; a custom dictionary captured domain-specific terms and synonyms (e.g., 'Baibai' for worship; 'Tian Hou' and 'Ma Zu' as the same deity). TF-IDF was used for document representation. Topic modelling employed LDA via Gensim, with Gibbs sampling for parameter estimation. The optimal number of topics was selected by minimising perplexity and maximising coherence; coherence peaked at 4 and 8 topics, perplexity was lowest at 3 and 4, leading to a final choice of 4 topics. Outputs included subject-word matrices, intertopic distance mapping (MDS on PC1/PC2), and top terms per topic (Supplementary Tables S2–S3). The four topics were interpreted and named based on their top keywords and semantic connections: Topic 1 'Practitioners Worship'; Topic 2 'Temple Activities'; Topic 3 'Deity Legends'; Topic 4 'Merchandise about Deity Statues'.

• Corpus: 4012 Facebook text posts on 191 deities (collected during the 2023 Spring Festival). - LDA revealed four topics with proportional distributions: Topic 1 Practitioners Worship (40.90% of tokens), Topic 2 Temple Activities (31.59%), Topic 3 Deity Legends (17.38%), Topic 4 Merchandise about Deity Statues (10.12%). - Intertopic relationships: Topic 1 and Topic 2 show the strongest correlation; Topic 4 is the most weakly linked to the other three. - Overlapping high-frequency terms: 'Safety and Peace' and 'Merits and Virtues' appear across all four topics; 'Pray for Demands' and 'Consecration' span three topics. Topic 1 and Topic 2 share the most keywords (15); Topic 1 and Topic 3 share 6; Topic 3 and Topic 4 share 5; Topic 2 and Topic 4 share 4; Topic 2 and Topic 3, and Topic 1 and Topic 4 each share 3. - Thematic categorisation (Table 3) shows high overall prominence of 'Objects', 'Practices and Rituals', and 'Attributes and Functions' across topics; Topic 4 distinctly emphasises 'Business and Commerce' (30%) and 'Religious Items' (23.3%). - The findings highlight a utilitarian orientation of practices (seeking prosperity, safety, vitality), rich temple-based social functions (rituals, ceremonies, fundraising, events), regional diversity of deities (notably Hokkien-linked deities such as Ma Zu, Ne Zha, Wang Ye, Xuan Tian Shang Di, Bao Sheng Da Di, Guang Ze Zun Wang), and a strong relationship to the Spring Festival and traditional calendar (e.g., 'Gui Mao', 'The First Lunar Month'). - A semantic nexus links 'Safety and Peace', 'Pray for Demands', and 'Merits and Virtues' as core concepts shaping practice cycles on social media.

The study elucidates how different constructors—practitioners, temple organisations, media organisations, and merchants—interact on Facebook to construct and circulate Malaysian Chinese folk beliefs. Practitioners share personal rituals, experiences, and needs; temples disseminate ceremonies, activities, and culture; media outlets report related news and features; merchants supply religious items, catalysing the 'deity economy'. Overlaps in content show practitioners and media discussing 'Practitioners Worship', while temples, practitioners, and media commonly cover 'Temple Activities'; 'Deity Legends' draw participation from temples, practitioners, and merchants; 'Merchandise about Deity Statues' is largely merchant-driven with practitioner and temple demand. The characteristics observed include utilitarianism (pragmatic engagements seeking prosperity, peace, and vitality), regional diversity and godhood specificity (e.g., Ma Zu’s expanded roles; prominence of God of Wealth; syncretic Datuk Gong), and multiple social functions of temples (rituals, fundraising, banquets, competitions, lion dances). The 'Realms' dimension reveals flows of deity statues from China, Taiwan, and Thailand to West/East Malaysia and flows of participants to favored locations (e.g., ancient temples, Penang), highlighting ties to religious tourism. Taoist influence permeates legends and customs (cultivation, manifestation, Tai Sui rituals, stove god, paper offerings, welcoming God of Wealth), extending to folk medicine and Feng Shui ideas. Commercialisation is evident in online markets for deity statues (materials like camphor wood, carving styles, sizes, pricing, customisation), spanning both West and East Malaysia. The Spring Festival intensifies temple participation (worship, incense, lighting lamps, pageants, lion dance), integrates zodiac/Tai Sui concerns, and anchors time-sensitive practices. The semantic nexus of 'Pray for Demands' → 'Merits and Virtues' → 'Safety and Peace' explains a cycle whereby practitioners undertake visible meritorious acts (donations, offerings, purchasing items) and inner devotion to seek divine intervention and stability, reinforcing community identity and philanthropy. These dynamics demonstrate social media’s role as an interactive medium that shapes meanings and practices within the Malaysian Chinese community.

The study contributes a social media–based, NLP-driven analysis of Malaysian Chinese folk beliefs, identifying four dominant thematic topics on Facebook and quantifying their prevalence. It demonstrates the central roles of practitioners, temples, media, and merchants, and characterises the online belief landscape as utilitarian, regionally diverse, socially functional, transregionally flowing, heavily influenced by Taoism, commercialised, and closely tied to the Spring Festival. The semantic linkage among 'Safety and Peace', 'Pray for Demands', and 'Merits and Virtues' offers a conceptual model of practice cycles in the digital sphere. Social media emerges as a catalyst for both continuity and innovation of folk beliefs, broadening reach and enabling new forms of interaction and commerce. Future research may extend these insights via fieldwork to validate online narratives, incorporate multimodal analysis (images/videos), and perform sentiment analysis to capture affective dimensions. The authors also anticipate continued reinforcement of the role of temples as spiritual and social hubs and sustained commercial and charitable activities associated with 'Merits and Virtues'.

Data were manually collected due to lack of access to official CrowdTangle scraping, focusing on active deities and text posts only. LDA on short texts has known limitations, especially with very brief posts. Image and short video content (e.g., TikTok) was not analysed. Suggested improvements include adding fieldwork for validation, expanding to multimedia analysis (images/videos), and incorporating sentiment analysis to assess emotional orientations. These steps could enhance understanding of dissemination strategies, cultural identity formation, and address issues around commercialisation and religious tourism.