Enhancing the dissemination of Cantonese Opera among youth via Bilibili: a study on intangible cultural heritage transmission

The paper addresses the preservation and transmission challenges of intangible cultural heritage (ICH), emphasizing Cantonese Opera (Yueju), which UNESCO inscribed in 2009. Globalization, modernization, digitization, and aging inheritor populations threaten ICH continuity. Cantonese Opera faces audience aging and declining youth participation. Social media—especially youth-dominated platforms like Bilibili—offers promise for revitalization. Bilibili’s user base skews young (average ~23.5 years), making it a suitable site to understand youth preferences for traditional culture content. The study’s objective is to identify factors influencing the acceptance and transmission of Cantonese Opera videos among Bilibili’s young users and to derive strategies that both engage youth and facilitate older enthusiasts’ participation via digital platforms. Guided by the elaboration likelihood model (ELM), the study examines central (content recognition) and peripheral (video surface features) routes affecting transmission effects, using a dataset of 1916 videos and 17 variables, with empirical testing through AHP, linear regression, and multi-factor ANOVA. The research proposes to build an influence model for ICH video dissemination and translate findings into practical macro strategies and micro tactics for content creators.

The review distinguishes traditional ICH safeguarding approaches (tourism, education, training) and modern methods (digital technologies, social media). Studies have explored ICH in tourism (e.g., product planning and experience-driven loyalty), evaluation frameworks for Cantonese Opera teaching, and the industry’s resilience post-crisis. Innovative approaches include game-based revival of indigenous narratives, AI-driven genre classification for Cantonese Opera, and VR replication of ICH practices, highlighting technical pathways for preservation. Social media research positions platforms like YouTube and purpose-built systems (PLUGGY) as archives and participation tools for heritage, though platform constraints and alignment with official heritage policies pose challenges. The review concludes that diversifying protection methods to align with contemporary media landscapes is essential, with a need for youth-focused, quantitative, and actionable strategies that guide specific inheritors and content creators on how to engage younger audiences effectively on new media platforms.

Design and framework: The study adopts the elaboration likelihood model (ELM) to analyze how central and peripheral routes affect the transmission effect of Cantonese Opera videos on Bilibili. Data collection: Using Python crawler technology, researchers initially retrieved 4860 videos with keywords related to Cantonese Opera. After manual refinement by three trained coders, 1916 videos definitively about Cantonese Opera were retained. For each video, 17 variables were extracted: eight quantitative (Play, Share, Like, Coin, Collect, Bullet_Comment, Comment, Fans) and nine qualitative (AT author type; HD highest definition; VT submission type; VTT title tone; VTST title style; ST subtitles; T hashtags; CL channel tags; TQ duration type). Operationalization of transmission effect: To mitigate confounds from opaque recommendation algorithms, Play is used as a standardizer rather than a direct effect indicator. Communication breadth = Share/Play; Communication depth = (Like + Coin + Collect + Bullet_Comment + Comment)/Play; Transmission_Ln = ln(breadth + depth). Recognition (central-path cue) leverages Bilibili’s “one button three links” (likes, coins, collections): Recognition_Ln = ln[(Like + Coin + Collect)/Play]. Variables and routes: Central path—Recognition_Ln as predictor; moderation by Author Type (AT: ordinary, professional, institutional). Peripheral path—eight qualitative factors (HD, VT, VTT, VTST, ST, T, CL, TQ) as independent variables; Fans_Ln included as covariate due to potential fanbase influence. Coder reliability: Three coders were trained for eight qualitative variables. A 200‑video subset was independently coded and Cohen’s Kappa computed pairwise; all variables exceeded 0.7 (p<0.01), indicating high reliability. Python scripts were used for efficiency, validated via SPSS for two variables. Analytic hierarchy process (AHP): Thirty‑one experts performed pairwise importance judgments on the seven quantitative indicators relevant to communication effect. Consistency (CR<0.1) and sequence consistency were assessed in Yaanp; 15 experts’ matrices met both criteria and were retained to compute final weights for Play, Like, Coin, Collect, Bullet_Comment, Share, Comment. Statistical analyses: Normality of Recognition_Ln and Transmission_Ln was verified (skewness, kurtosis, histograms). Multi-factor ANOVA with a covariate (SPSS 26) tested nine main effects (eight factors + Fans) and up to 247 interaction groups (second to eighth order), acknowledging some untestable higher-order combinations due to sparse cells. Linear regression assessed the central-path effect (Recognition_Ln → Transmission_Ln) overall and within AT subgroups to evaluate moderation. Assumptions were checked via scatterplots and residual diagnostics (Durbin–Watson).

Descriptive and reliability: All coded qualitative variables achieved Cohen’s Kappa > 0.7 (p<0.01). Recognition_Ln and Transmission_Ln approximated normal distributions. AHP weights (communication indicators): Play 0.1890; Like 0.1608; Coin 0.1424; Collect 0.1342; Share 0.1501; Comment 0.1133; Bullet_Comment 0.1102. Central path (regression): Recognition_Ln strongly predicts Transmission_Ln. - Pearson correlation r = 0.903, p < 0.01. - Overall model: Adjusted R² = 0.815; regression coefficient (slope) = 0.855; p < 0.001; Durbin–Watson ≈ 1.976. - Moderation by author type (AT): • AT=1 (ordinary): Adj. R² = 0.813; coef = 0.860; p < 0.001; D–W ≈ 1.968. • AT=2 (professional): Adj. R² = 0.976; coef = 0.890; p < 0.001; D–W ≈ 1.898. • AT=3 (institutional): Adj. R² = 0.560; coef = 0.651; p < 0.001; D–W ≈ 1.877. Interpretation: AT moderates the recognition–transmission relationship; the effect is strongest for professional users and weakest for institutional users. Peripheral path (multi-factor ANOVA with Fans_Ln covariate): Significant main effects were found only for: - ST (subtitles): F=4.237, p=0.040, partial η²=0.002 (favoring no subtitles in raw main-effect contrast; see optimal combos below for interaction-qualified recommendations). - T (hashtags): F=9.935, p=0.002, partial η²=0.006 (presence beneficial). - CL (channel tags): F=5.149, p=0.023, partial η²=0.003 (submission to channels beneficial). Fans_Ln (covariate) was not a significant predictor in the omnibus model. Significant interactions (examples): VT*TQ; HD*VT*CL; HD*ST*T; HD*CL*TQ; VT*ST*T; VT*ST*TQ; VTT*ST*T; VTST*ST*T; HD*VT*CL*TQ; VT*ST*T*TQ. Effect sizes were small (partial η² ≈ 0.002–0.007). Optimal factor settings (from one‑way follow‑ups on significant effects/interactions): - ST: ST2 (without subtitles) as a main-effect optimum; however, interaction-informed optimal combinations favor including subtitles alongside other features (see below overall recommendation). - T: T1 (with hashtags). - CL: CL1 (with channel tags). - Interaction-derived optimal combinations included: • VT2×TQ1 (Music × <10 min) • HD3×VT2×CL1 (≥1080p × Music × With channel) • HD3×ST2×T1 (≥1080p × No subs × With hashtags) • HD3×CL1×TQ1 (≥1080p × With channel × <10 min) • VT2×ST2×T1; VT2×ST2×TQ1; VTT1×ST2×T1; VTST1×ST2×T1 • HD3×VT2×CL1×TQ1; VT2×ST2×T1×TQ1 Hypotheses: H1 (recognition effect) and H2 (AT moderation) supported. Among peripheral hypotheses, only H7 (subtitles), H8 (hashtags), and H9 (channel tags) showed direct main effects; H3 (submission type), H4 (highest definition), H5 (title tone), H6 (title style), H10 (duration type), and H11 (fans) were not supported as main effects, though several contributed via interactions. Synthesis recommendation (platform/context-specific): An empirically optimal eight-factor profile for better dissemination on Bilibili’s Cantonese Opera videos is: ≥1080p definition; Music category; general declarative title tone; written title style; include hashtags; submit to specific channels; include subtitles; duration <10 minutes.

Findings confirm the ELM’s applicability: the central route (user recognition based on content quality and value) exerts a dominant influence on transmission effects, while the peripheral route adds smaller, incremental gains via presentation and packaging features. The strong recognition–transmission linkage supports a “content is king” strategy for ICH video dissemination on youth-centric platforms like Bilibili. Author type moderates this central effect, with professional-user content showing the strongest recognition leverage and institutional content the weakest, suggesting audience preconceptions and credibility cues shape how recognition translates into spread. Peripheral features—subtitles, hashtags, channel tags—exhibit significant main effects, and several other features (HD, title tone/style, duration, submission type) contribute through interactions. These provide actionable, near-term tactics to complement long-term content improvements. The recommended eight-factor configuration offers a practical blueprint; however, it is derived from Cantonese Opera on Bilibili and may require recalibration for other ICH forms or platforms. The study advances measurement of video communication effects by normalizing engagement metrics to Plays (breadth and depth ratios) and deriving indicator weights via expert AHP, thereby reducing algorithmic recommendation bias associated with raw Plays and emphasizing quality-adjusted engagement. This framework helps disentangle content-driven transmission from platform-driven exposure, making it useful for future communication studies on ICH and beyond. Overall, leveraging a “young” platform to disseminate an “old” heritage form appears promising: if content resonates and is packaged optimally, younger audiences can be effectively engaged, helping bridge generational gaps and supporting sustainable ICH transmission.

The study develops and empirically validates an ELM-based influence model for Cantonese Opera video dissemination on Bilibili. Results underscore the primacy of the central route: higher user recognition leads to substantially greater transmission, moderated by author type (strongest for professional users). Peripheral features matter modestly in main effects (subtitles, hashtags, channel tags) and via interactions. A recommended eight-factor configuration for enhanced dissemination is: ≥1080p definition; Music category; general declarative title tone; written title style; include hashtags; submit to specific channels; include subtitles; and duration under 10 minutes. These insights yield macro-level guidance (prioritize content quality to drive recognition) and micro-level tactics (optimize video packaging) for ICH creators targeting youth audiences. Future research should generalize and tailor the model across different ICH types and platforms, validate time dynamics with longitudinal data, and refine weighting schemes with broader expert panels and user studies to further align measurement with real audience behavior.

(1) Category imbalance: Substantial disparities in group sizes (e.g., Author Type: ordinary n=1780 vs professional n=45 and institutional n=91) may affect estimates and interaction tests, though they reflect typical UGC-dominant platform dynamics. (2) Dataset coverage: Despite 1916 videos, the niche nature of Cantonese Opera on a youth platform limits some higher-order interaction tests and generalizability; ongoing data collection is needed. (3) Cross-sectional design: Data were collected on March 15, 2023; time-series analyses are needed to capture temporal dynamics and evolving platform/user behaviors.