Introduction

In 2003, during its 32nd session in Paris, UNESCO adopted the Convention on the Protection of Intangible Cultural Heritage. The UNESCO Convention of 2003 aimed to enhance awareness, particularly among the younger generations, regarding the significance of intangible cultural heritage (ICH) and its preservation by establishing a binding multilateral instrument (UNESCO, 2003). Authorities explicitly defined ICH for the first time. It is continually reshaped by society and social groups, seamlessly integrating with nature and history, providing a sense of cultural identity and the perpetuation of civilization. Such contributions to society foster tremendous respect for human creativity and cultural diversity (Guo and Li, 2015). ICH serves as a cultural symbol for each nation. People commonly view it as the legacy that ancestors pass down to their descendants through life productions and creations, encompassing traditional rituals, performing arts, various practices, and activities (Zhou, 2022). These cultural heritages hold significance due to the invaluable human cultural values and concepts they embody. They are the most compelling evidence of human civilization’s continuous inheritance and evolution. In the current social context, globalization, modernization, the digitalization of cultural transmission, and profound changes in social production and lifestyle pose escalating threats to the safeguarding and continuity of ICH (Lin and Lian, 2018). Due to the aging of inheritors and the lack of market-oriented economic value, numerous ICH skills are gradually fading and facing extinction risk (Del Barrio et al., 2012). ICH is a valuable treasure shared by all of humanity, and it should be extensively promoted and disseminated throughout society. However, its current inheritance faces challenges such as a narrowing audience and a shortage of inheritors (Xue et al., 2019).

In 2009, Cantonese Opera, originating from the Cantonese-speaking provinces of Guangdong and Guangxi in south-eastern China, gained recognition and was officially inscribed on the Representative List of the Intangible Cultural Heritage of Humanity (UNESCO, 2009). A blend of strings and percussion instruments distinguishes Cantonese Opera. The artistic origins of Cantonese Opera are subject to different views among scholars. Some suggest tracing it back to the Ming Dynasty (1638–1644) (Leung, 1982), while others believe it might have emerged during the Southern Song Dynasty (1127–1279) (Chan, 1999). As society embraces new cultural influences and popular music, and with Cantonese Opera groups aging, the preservation and promotion of Cantonese Opera face significant challenges. Particularly concerning is the lack of younger participants, which has become a critical concern (Mei et al., 2018). The challenges confronting traditional culture, including Cantonese Opera, are a shared concern across ICHs. These challenges focus on rejuvenating traditional culture, often labeled as ‘traditional,’ ‘old,’ and ‘ancient,’ to engage the younger generation—an area central to our research.

Leveraging social media platforms, particularly those preferred by the younger generation, is a potential solution to address the challenges posed by aging ICH and the lack of young audiences. This approach can contribute to the preservation and continuity of ICH and promote social engagement for older adults through their active participation in social media. Data indicates that China is one of the world’s fastest-aging countries with the largest aging population (He et al., 2020). The issue of aging extends beyond any specific country or region; it marks an inevitable phase in human society’s progression. The State of World Population Report 2023 reveals a rapid global aging trend, with 12.3% of the world’s population aged 60 and above projected to increase to nearly 22% by 2050 (UNFPA, 2023). In today’s digital age, social media is a significant social intermediary, facilitating social participation, interaction, and enjoyment (Yujie et al., 2022). Social media has become an integral part of people’s lives, offering a substantial platform for cultural communication (Lai et al., 2022). Recently, new-format online video websites within social media, including TikTok, YouTube, and Bilibili, have gained immense popularity as information sources for the public (Dong et al., 2022). Bilibili, distinct from other online video platforms, initially centered around animation, comics, and games (ACG) content and later evolved into a comprehensive video platform. This evolution has appealed to the younger audience (Wikipedia, 2023). The 2022 Bilibili Research Report: Bilibili Community Ecology and Commercialization Analysis, published by ZhongTai Securities, reveals that users under the age of 24 represent 75.4% of the platform’s user base, with an average age of 23.5 years old (Vzkoo, 2022).

Consequently, this study juxtaposes the “old” ICH and the “young” platform of Bilibili. The research explores strategies for enhancing the preservation and safeguarding of traditional ICH within the cultural community of Bilibili, where young people gather. Specifically, the study will investigate ways to tailor Cantonese Opera communication content to align with the preferences of young individuals in the community. Simultaneously, the research will endeavor to bridge the cultural gap between young and elderly enthusiasts of Cantonese Opera, facilitating increased social participation of older adults through new social media platforms while providing additional social care for this age group.

To achieve our research objective of understanding the acceptance of Cantonese Opera videos among young people in Bilibili, we collected 1916 Cantonese Opera-related videos using Python crawler technology. We extracted 17 variables from each video, including quantitative and qualitative data. The study utilizes the elaboration likelihood model (ELM) as the theoretical framework to explore the influence of central and peripheral routes on the transmission effect of these videos. The Method and Theoretical Framework sections elaborate on detailed methodologies and theoretical underpinnings.

In general, this study will build an influence model of the propagation effect of videos related to Cantonese Opera on Bilibili, grounded in ELM theory. We will subject the central and peripheral paths of the model to empirical testing using qualitative and quantitative research methods, including the analytic hierarchy process (AHP), linear regression analysis, and multi-factors ANOVA. Theoretically, this study develops an influence model on the transmission effect of ICH in new video websites like Bilibili, drawing from the ELM principles. Developing this model provides a fresh perspective for research on disseminating ICH content through video in communication studies. Practically, this study derives specific and empirical conclusions from analyzing the influence model of Cantonese Opera cultural video transmission. We will transform these findings into macro strategies and micro tactics to guide video content producers of ICH in creating videos with improved communication effects, particularly content that resonates with young audiences.

This paper explores the transmission of ICH, specifically focusing on Cantonese Opera on the Bilibili platform. The paper organizes itself into seven key sections: Introduction, Literature Review, Theoretical Framework, Method, Results, Discussion, and Conclusion and Limitations. Each section systematically addresses different aspects of cultural heritage transmission, combining empirical analysis with theoretical insights to assess how digital platforms can engage younger audiences and support heritage preservation.

Literature review

Introduction to ICH

ICH constitutes a crucial element of human cultural evolution. Its protection and inheritance have become imperative not merely because it promotes the prosperous development of cultural industries but also because it is integral to human culture protection, civilization inheritance, and current national image enhancement (Song et al., 2015). Owing to the extraordinary significance of this endeavor, numerous academics have investigated it from diverse perspectives. We can broadly categorize methods for protecting and transmitting cultural heritage into traditional and modern methods. The traditional methods refer to time-honored means of cultural protection, like cultural tourism, education, and industry training. Modern methods signify their basis in innovative ideas tailored to the evolving needs of the present era. These modern approaches incorporate recent technologies, media platforms, and current trends.

Traditional and modern methods for ICH preservation

Esfehani et al. undertook an empirical study exploring tourism product planning rooted in ICH, employing qualitative ethnographic research methodologies (Esfehani and Albrecht, 2019). The research enriches understanding of tourism products in ICH destinations, especially the accessibility of ICH-based tourism products in the tourist market and the suitability of employing ICH in tourism. This study proffers a tangible strategy for ICH redevelopment: revitalizing ICH as a tourism product. A constraint of this study lies in its reliance on qualitative ethnographic research for empirical investigation. While this facilitates attention to ICH tourism and community voices, it needs a macroscopic rational appraisal of the validity of specific program outcomes. Su et al. utilized structural equation models to formulate a research framework for visitors engaging with the beliefs and customs linked to the sea goddess Mazu on Meizhou Island. The study explored the relationship between tourist engagement, awe, quality of experience, and loyalty (Su et al., 2020). Their study empirically demonstrated, through analysis of 393 questionnaire responses, that awe is an outcome variable of engagement and quality of experience and an antecedent variable of loyalty.

Furthermore, awe mediates between involvement and loyalty, as well as between experience quality and loyalty. The researchers also found a significant gender-based difference in the impact of awe on loyalty. This study adeptly leverages quantitative research methods to probe tourist activities connected to ICH and uses tourist experience analysis to generate improved strategies for promoting tourism. A drawback is that the types of ICH selected in this study, as stated by the authors, belong to the category of beliefs and customs. This ICH has certain constraints, particularly the variation in individual beliefs and the applicability of this approach to future ICH studies. Qiu et al. utilized content analysis to examine the Cognitive Image of ICH Tourism through 9074 blog posts on ICH tourism on Weibo, one of China’s foremost social media platforms (Qiu and Zhang, 2021). The research delineates the dimensions of ICH tourism’s cognitive image and the involvement of various related characters, employing matrix construction, dimension classification, and semantic network analysis. It further uncovers the regulatory influence of institutions over ICH tourism and folklore as the most recurrently developed ICH resource. While the study’s results can furnish pertinent information for policymakers, the research needs to discuss specific protection and inheritance measures or the direction of tourism transformation. Hence, further research is needed to advance the practice of protection and inheritance.

Specific studies on Cantonese Opera

Numerous studies similar to ours have delved into the field of ICH, mainly focusing on the preservation and propagation of Cantonese Opera culture. Using grounded theory research, Lo observed two Cantonese Opera singing lessons, uncovering the etiquette and core values intrinsic to these sessions (Lo, 2015). The study discerned that Cantonese Opera, a specific musical genre, can prove advantageous in education and in facilitating the participation of older adults, aiding the maintenance of friendships and physical and mental health, thereby contributing to successful aging. From the vantage point of Cantonese Opera teaching and elderly participation, this study further delves into the societal benefits of ICH, like Cantonese Opera, offering guidance and importance for enhancing its preservation and propagation. However, the beneficial conclusions derived from this study predominantly stem from the logical judgment of perceptual materials, indicating a relative lack of rational and empirical results that could provide a more substantial scientific basis for its conclusions. Luo et al. developed and piloted a standardized evaluation framework for teaching Cantonese opera singing. This framework aimed to address the challenges inhibiting the development and inheritance of Cantonese Opera within communities and schools, which stemmed from a need for standardized and practical assessment guidance (Luo and Leung, 2023). Their study proposed an assessment framework based on a holistic theoretical model derived from the revised Bloom’s taxonomy, encapsulating the Cognitive, Psychomotor, Affective, and Behavioral Domains. The findings of this study have significantly contributed to Cantonese Opera education by offering a systematic and scientific evaluation and feedback criterion. The pilot trials involving 4 Cantonese Opera teachers and 24 students yielded promising results. If we can further apply these findings in communities and schools and monitor the effects of large-scale usage, we can refine this evaluation system more effectively. Compared with the studies mentioned above on the preservation of Cantonese Opera education, Chung’s research has a stronger focus on the current state of the Cantonese Opera industry, especially in the aftermath of the epidemic. Employing the transactional stress and coping theory, Chung critically examined Cantonese Opera artists’ roles, psychological responses, and coping mechanisms during traumatic crises (Chung, 2022). Their research underscores the need for the positive evolution of the Cantonese Opera industry amid these global challenges. Achieving this evolution requires Cantonese artists to elevate their professionalism, artistry, and knowledge consistently and to innovate and adapt to continuous socio-ecological changes, ensuring Cantonese Opera’s sustainability and heritage preservation stability.

Innovative methods and social media for ICH preservation

Traditional protection methods, such as those rooted in cultural tourism, education, and industries, are undoubtedly effective. However, pursuing innovative methods and strategies is paramount alongside these. Laiti et al. utilized an ethnographic approach to examine how ‘game jams’—a rapid collaborative game production format—can foster the resurgence of Indigenous self-narratives within the Sámi culture context, hence sustaining ICH through game jamming as a cultural practice (Laiti et al., 2021). Their research probes into the viability of video games as a distinctive approach to reinvigorating ICH, citing specific game jam models as case studies. While the study paves a new path for ICH revival, it poses challenges for regular ICH content creators to replicate this approach directly and unambiguously. In 2022, researchers established a scientific audio database dedicated to Cantonese Opera. It includes a classification method for Cantonese Opera singing genres based on the Cantonese Opera Genre Classification Networks (CoGCNet) model (Chen et al., 2022). The study’s results demonstrate the high classification accuracy of the proposed method, outperforming commonly used neural network models. Therefore, this method can provide technical support for identifying, researching, and excavating the ICH of Cantonese Opera. In 2023, Chen utilized virtual reality technology to digitally replicate a specific ICH—gongs and drums—and proposed enforceable strategies for subsequent preservation and protection (Chen, 2023). Integrating new technologies such as VR offers an efficient and accessible technical medium for activating ICH, profoundly impacting the expansion of the temporal and spatial dimensions of ICH preservation. Should these projects find specific applications in the future and garner substantial real-world feedback from a broad audience, the widespread application of these technical achievements becomes more plausible. Since the emergence of social media in the late 20th and early 21st centuries, numerous scholars have investigated its potential for preserving and disseminating cultural heritage. In 2013, Pietrobruno scrutinized YouTube’s role as an ICH archive, a novel challenge to the gender limitations of transmitting some traditional cultures and a potential vehicle for broader dissemination (Pietrobruno, 2013).

The authors utilize the Mevlevi Sema ritual as an exemplar in this study. It is a mystical Islamic practice encompassing the Sema dance, or what the West calls the ‘Whirling Dervish Dance,’ analyzed via actual and virtual ethnography. Indeed, the study illustrates YouTube’s role as an archive in challenging the prevailing fact that local women are predominantly excluded from participating in the ceremony, thereby fostering cultural heritage preservation. However, as the authors articulate in the article, YouTube’s potential as an archive is restricted due to occasional conflicts with state-sanctioned heritage. As YouTube can only circulate content that users have uploaded, perfecting the archive is necessary for a video hosting service to function effectively as a specific cultural heritage archive. 2018 Lim et al. present PLUGGY as a Pluggable Social Platform for Heritage Awareness and Participation (Lim et al., 2018). They posit that cultural heritage should seamlessly integrate into daily life for genuine sustainability, asserting that social platforms are suitable for this purpose. Liang et al. view social media as an epistemic medium for advancing the expression of World Heritage in China (Liang et al., 2022). These scholars hold a positive stance on the role of social media in transmitting cultural heritage. However, there remain areas for improvement. Foremost among these is harnessing the vast potential of social media for effective content transmission, ensuring maximal communicative impact. This study addresses the central inquiry of how to achieve this goal.

Summary and future directions

In summary, modern methods highlight their relevance in the contemporary context. Cultural protection and development demand diversification, especially approaches that resonate with the current era. This evolution explains the shift from traditional methods in tourism and education towards “new” strategies, such as AI, VR, digital games, and contemporary social media, as many scholars emphasize. Changing times have refocused cultural protection efforts, and the rise of new media has expanded these possibilities. New tools will likely enhance the younger generation’s involvement in preserving cultural heritage. However, this goal demands further research on the young generation. Earlier studies lacked direct solutions for the age-related challenges in traditional culture, suggesting broader societal interventions instead.

Scholars often offer macro-level suggestions regarding ICH safeguarding strategies. However, specific ICH inheritors and smaller entities require more tailored guidance, underscoring the importance of devising communication strategies to engage the youth and determining their implementation. Our research suggests that a comprehensive quantitative study on ICH transmission might offer more targeted and actionable insights, focusing on young people’s preferences in “new” social media.

Theoretical framework

Introduction to the elaboration likelihood model (ELM)

The study adopts a research framework based on the ELM (Petty and Cacioppo, 1986). This model examines the transmission effect as the dependent variable and explores the influence of the central and peripheral paths as independent variables on the acceptance of Cantonese Opera videos among young people in Bilibili. Richard E. Petty and John Cacioppo developed the ELM in 1980 (Van Lange et al., 2011). The ELM posits that people process received information through two distinct routes: the central and peripheral paths. The central path involves a rigorous and rational analysis of the relevant information, resulting in judgments that influence corresponding behaviors.

In contrast, the peripheral path signifies less energy-intensive processing of information based on superficial cues (Petty and Cacioppo, 2012). The central path is particularly effective when fine-grained processing is likely, while the peripheral path takes precedence when this probability is low (Dillard and Shen, 2013). Researchers widely apply the ELM in various domains, including advertising, marketing, consumer behavior, healthcare, and media communications (Angst and Agarwal, 2009; Bitner and Obermiller, 1985; Campbell and Chung, 2022; Flynn et al., 2011; Lee and Koo, 2016; Ott et al., 2016; Petty et al., 1983). Generally, the central path focuses on the intrinsic quality of the information, while the peripheral path directs attention toward surface attractiveness and other extrinsic features unrelated to the content essence.

Application of ELM in this study

Drawing from the ELM, this study formulated the theoretical framework for this study. Firstly, the transmission effect of Bilibili’s Cantonese Opera-related videos will serve as a core variable in our research. We aim to investigate the influence of the central and peripheral paths on video transmission effectiveness to glean critical insights for enhancing video transmission. The definition of video communication effect in our study draws on prior relevant literature that calculated Bilibili’s video communication effect (Chen et al., 2020; Cheng and Guan, 2023; Wang et al., 2022; Xiao et al., 2022; Zhang et al., 2020; Zhang and Hu, 2022). Additionally, this research employs several metrics directly available from the current Bilibili platform to define and calculate this effect. Researchers can collect Bilibili’s communication effects indicators such as Play, Like, Coin, Collect, Share, Bullet_Comment, and Comment. These constitute seven quantitative indicators out of the previously mentioned 17 variables. Unlike previous studies, this study has innovatively improved the definition of communication effects based on more stringent scientific requirements.

The difference lies in the utilization of the Play indicator. Some studies above directly summed up the seven variables representing a video’s communication effect. In this research, our interest lies in studying the contribution of the video’s content itself to the effect of communication. We aim to derive conclusions that can guide the production of related content. Current video-sharing websites like Bilibili, TikTok, and YouTube employ extensive data content push mechanisms for users. Each platform’s push mechanisms remain proprietary business secrets inaccessible to us. However, it is clear that these push mechanisms significantly influence the number of video views. Hence, this study has revised the calculation and definition of the communication effect. As a result, it will no longer use ‘Play’ as a direct representative indicator; instead, ‘Play’ will serve as a standard indicator. These changes bifurcate the communication effect into communication breadth and depth. Our study applied the natural logarithm to the results to normalize the values and mitigate potential distortions caused by the enormous value range among the seven indicators. The formal expressions are as follows:

  1. (1)

    Breadth of communication = Share/Play

  2. (2)

    Depth of communication = (Like + Coin + Collect + Bullet_Comment + Comment)/Play

  3. (3)

    Transmission_Ln = Ln(The breadth of communication + The depth of communication)

In this definition of the communication effect, our method of evaluating communication breadth utilizes the ratio of Shares to the current Plays of a video as an indicator. Similarly, this study assesses the depth of communication using the proportion of the sum of Likes, Coins, Collections, Bullet_Comments, and Comments to Plays as indicators. This definition’s advantage is its capacity to reduce the impact of other factors that can influence Play but are not directly related to the video. Theoretically, a video with a high number of Plays, implying a wide breadth of communication, should attract more Shares than videos with fewer Plays. Consequently, this ratio serves as a valuable and effective indicator. Furthermore, to bolster our assessment’s validity and scientific integrity, this research invited 31 experts to appraise the significance of the seven factors in the communication effect using the AHP and consequently procured the corresponding weight values for each indicator. The section below will elaborate on details about the AHP.

Central path analysis

Following identifying the key transmission effect variable, this study also ascertains the central and peripheral paths that investigate its impact. This study’s central path examines whether user recognition of videos influences communication effectiveness. As the recognition assessment necessitates users to engage more critically and profoundly with the video content, the recognition indicator aligns with the central path of the ELM model. This research defines recognition by incorporating Bilibili’s unique reward culture. Notably, in Bilibili’s culture, most video creators frequently include phrases such as “If you like this video, please like it, coin and collect” towards the end of their content. This sequence of actions is uniquely termed “one button three links,” or “Yi Jian San Lian” in Chinese Pinyin, within the Bilibili community. The term “Yi Jian San Lian” originates from a unique Bilibili feature, where a user can simultaneously perform three interactions—Like, Coin, and Collect—by long-pressing the like button. Over time, within the Bilibili community, audiences have come to express their recognition and appreciation for a video using this “Yi Jian San Lian” method. Lastly, drawing on the previous analysis of Play and the application of natural logarithms, this study defines and calculates the degree of recognition as follows:

  1. (4)

    Recognition_Ln = Ln [(Like + Coin + Collect)/Play]

Besides the recognition variable, this research also considers the type of video author. On Bilibili, aside from the officially self-released videos, the bulk of the content comes from three types of authors: individual users, professional users, and institutional users. Broadly, Bilibili itself fits into the category of institutional users. These three categories align with the three types of content production models prevalent on mainstream social media today: User-generated Content (UGC), Professional-User-Generated Content (PUGC), and Professionally-Generated Content (PGC). It is reasonable to presume that users may have preconceived notions about videos produced by different authors when evaluating a video’s recognition. Consequently, this study incorporates the author type into our central path analysis to examine its potential moderating role in the impact of recognition on communication effectiveness. Specifically, we will test whether it acts as a moderator.

Peripheral path analysis

In the peripheral path, this study considers eight qualitative variables—video submission type, highest video definition, tone and style of the video title, presence of subtitles or hashtags, presence of Channel tags, and video duration type—which represent external surface characteristics of a video. These variables allow users to make an initial assessment with minimal cognitive effort. Considering the impact of the fan effect, the number of followers might affect the communication effect. Thus, this research incorporates the number of an author’s fans into our peripheral path analysis as a covariate.

Research questions and hypotheses

Drawing from the ELM, we identified influencing factors for central and peripheral routes, which are the study’s focus. Subsequently, we will outline the study’s research questions. Initially, this study uses Cantonese opera to probe into new media communication strategies that address the aging issue of ICH. Specifically, we aim to encourage greater youth participation in disseminating Cantonese opera culture. Consequently, we initiated research on Bilibili, a youth-centric cultural community, examining which Cantonese opera video formats resonate most with its users. On Bilibili, where the average user age is 23.5, preferences largely mirror the younger generation’s inclinations towards traditional culture videos. Thus, our inquiry delves into the types of Cantonese opera videos that gain traction on Bilibili, determining the content that captivates and spreads among today’s young media consumers. Grounded in these defined research questions, we forward several hypotheses for empirical testing:

Hypothesis 1: The recognition of a video significantly impacts the transmission effect of the video.

Hypothesis 2: The author type plays a moderating role in the transmission effect of video.

Hypothesis 3: The type of video submission significantly impacts the transmission effect of video.

Hypothesis 4: The highest definition of the video significantly impacts the transmission effect of the video.

Hypothesis 5: The tone of the video title significantly impacts the transmission effect of the video.

Hypothesis 6: The style of the video title significantly impacts the transmission effect of the video.

Hypothesis 7: Whether the video has subtitles significantly impacts the transmission effect of the video.

Hypothesis 8: Whether the video has hashtags significantly impacts the transmission effect of the video.

Hypothesis 9: Whether the video has channel tags significantly impacts the transmission effect of the video.

Hypothesis 10: The video duration type significantly impacts the transmission effect of video.

Hypothesis 11: The number of fans significantly impacts the transmission effect of video.

Figure 1 below presents the complete research framework, with corresponding data analysis methods detailed in the ensuing section.

Fig. 1: Research framework.
figure 1

This figure mainly shows the overall research ideas of this study and the corresponding analysis methods. Detailed descriptions of the abbreviations in the figure can be found in Table 1.

Methods

Research subject and Python crawler

This study centered on videos about Cantonese Opera content on the Bilibili platform. We employed Python crawler technology to collect 1916 Cantonese Opera-related videos from Bilibili to achieve our research objective. We extracted 17 variables from each video, comprising eight quantitative and nine qualitative variables. The collection process involved using ‘Cantonese Opera’ and its aliases as keywords on Bilibili to retrieve and collect information from corresponding video search results. Initially, we amassed information from 4860 videos. Due to the search system’s diffuse nature, it often disaggregates compound nouns when corresponding search results are insufficient. For instance, if the results retrieved using “Guangdong Opera” (an alias of Cantonese Opera) are inadequate, the system may resort to using “Guangdong” as a search keyword.

Consequently, three coders refined the initially collected 4860 videos, resulting in a final dataset of 1916 videos distinctly associated with Cantonese Opera content. For each of the 1916 videos, our research gathered data on 17 variables, comprising eight quantitative and nine qualitative variables. The eight quantitative variables include the number of likes, coins, collections, shares, bullet comments, comments, plays each video has, and the number of the author’s followers. The nine qualitative variables encompass the type of author, the video’s highest definition, title tone, title style, the presence of subtitles, hashtags, channel tags, video duration, and submission type. Specific descriptions of these variables and the qualitative classifications are outlined in Table 1 below.

Table 1 The description of each variable.

Coders reliability analysis

Among the 17 variables examined in the 1916 videos of this study, nine are qualitative. Based on video duration, the ‘TQ’ variable is encoded using SPSS, while three coders encode the remaining eight qualitative variables. Initially, the three coders received training in variable coding operations. Next, we randomly selected 200 videos from the pool of 1916, in which the three coders independently pre-coded the eight qualitative variables—then computed Cohen’s reliability coefficients for the three coders across the eight variables pairwise. Upon meeting the reliability standard, the three coders completed the coding of the eight qualitative variables in the remaining videos.

Given that SPSS can only compute the reliability coefficient for two coders on one variable in a single analysis, it could be more efficient for the calculation involving three coders and eight variables. As such, this study utilizes Python to create the necessary code for the analysis. To verify no discrepancies in the syntax or results of the Python-written analysis, SPSS 26 extracts two variables for validation, thereby confirming the analysis results.

Analytic hierarchy process

This study engaged 31 field experts to apply the AHP in making importance judgments and weight calculations on the seven quantitative variables, thereby bolstering the scientific rigor of the communication effects calculation. Utilizing AHP, a seven-factor hierarchical structure model was established. The 31 experts assessed the importance of two factors using the Saaty relative importance scale, constructed a judgment matrix, computed the hierarchical single order, and conducted a consistency test, which yielded the weight coefficients of the seven factors. Throughout the entire AHP process, this study employed the Yaanp software.

Variance analysis

We will test peripheral paths using multi-factor ANOVA, incorporating a covariate. We will conduct the test using SPSS 26, with eight qualitative factors—HD, VT, VTT, VTST, ST, T, CL, and TQ—serving as independent variables. The quantitative variable Fans will act as a covariate, and the transmission effect value will be the dependent variable. This research applies the natural logarithm to the result to account for potential impacts from the extensive value range among the ‘Fans’ indicators, smoothing the value. In this eight-factor variance and covariance analysis, we will test the main effects of the eight factors and a covariate and the interaction effects of the eight-factor combinations. This analysis implies that we will test the nine main effects and 247 interaction effect groups, ranging from the second to the eighth order.

Regression analysis

We will evaluate the central path using linear regression analysis in SPSS 26. In the regression analysis, variable Recognition will be the independent variable, while the transmission effect value will be the dependent variable. Furthermore, given that our assumption suggests the AT has a moderating influence on this causal relationship, and considering that AT is a qualitative variable. Consequently, beyond standard regression analysis, this research also conducts a grouped regression analysis based on AT. We examine the moderating impact of AT by observing regression coefficients across different groups.

Result

Coders reliability

This study tested the reliability of the three coders. The Cohen’s Kappa coefficients exceeded 0.7, and the p-value was less than 0.01, suggesting a high consistency in the coders’ results. Table 2 shows the specific results of the reliability test.

Table 2 The result of the coders reliability test.

Descriptive statistics

Table 3 elaborates on the data from several qualitative and quantitative variables directly incorporated into the data analysis. Among these, the values succeeding the qualitative variables are the indicators corresponding to Transmission_Ln values across various categories within each qualitative variable. Additionally, since we require variance and regression analysis later, this article initially reports the normality of the Transmission_Ln and Recognition_Ln variables here. Following the standard suggested by Kline (1998), if the absolute value of the skewness coefficient falls below three and the kurtosis coefficient is less than eight, the data can be regarded as normally distributed (Kline, 1998). Furthermore, given the large volume of data in this study and the decreasing accuracy of the skewness and kurtosis coefficient test method with a growing sample size, we also utilize the histogram with the normal curve of the two variables to assess normality. The skewness and kurtosis coefficient displayed in Table 3 and the histogram illustrated in Fig. 2 confirm that the data from these two variables adhere to a normal distribution.

Table 3 Descriptive statistics result.
Fig. 2: Histogram of Recognition_Ln and Transmission_Ln.
figure 2

This figure shows the data distribution of the two variables (Recognition_Ln and Transmission_Ln) through a histogram with a normal curve.

Furthermore, as per Cohen’s judgment (Cohen, 2008), the variance among each group is likely to be less than perfectly equal when dealing with a large ANOVA sample size. Generally, with large samples, if the variance between the groups does not exceed four times, it meets the homogeneity of variance requirements and, hence, can proceed to ANOVA. In summary, as evidenced by Table 3, the variance between the groups does not exceed four times, confirming that the dataset fulfills the homogeneity of variance prerequisites.

Analytic hierarchy process

This research invited a total of 31 experts to participate in the AHP. Upon importing the pairwise judgment results of 31 experts on seven factors into Yaanp, the software automatically generates a pairwise judgment matrix while simultaneously checking the expert judgment matrix’s data consistency and order consistency. Using variables A, B, and C as examples, we can explain the distinction between the two. If A is twice as large as B and B is twice as large as C, logically, A should be four times as large as C. Entering A as three times larger than C will cause a data consistency error. Similarly, if A is deemed smaller than C at this juncture, it would result in a sequence consistency error, as logically, given that A > B and B > C, A should invariably be more significant than C. The data consistency test uses the CR value. Generally, a CR value of less than 0.1 indicates that the data has passed the consistency test, and we base sequence consistency on logical judgment. As illustrated in the subsequent Table 4, the CR values and sequence consistency results of the 31 experts are displayed based on the computation results from Yaanp. The results show that 16 of the 31 experts failed to satisfy the consistency and sequence consistency tests concurrently. Previous scholars have posited that when more than five indicators require evaluation within the AHP, it becomes challenging to avoid issues with data consistency and sequence consistency entirely. Given that this study involves a pairwise judgment of seven factors, the potential for error is inevitably present. Nonetheless, in adherence to the rigor of scientific analysis, our researchers excluded the data from the 16 experts who failed to meet both the consistency and sequential consistency tests, retaining only the data from the remaining 15 experts for the group decision-making process within the AHP. The integration of the judgments from the 15 experts facilitated Yaanp’s calculation of the weight coefficients for the seven factors, as depicted in Table 5.

Table 4 The result of data consistency test and sequence consistency.
Table 5 The weight coefficient.

Variance analysis

In this multi-factor ANOVA, eight categorical variables serve as independent factors, with an additional quantitative variable (Fan_Ln) introduced as a covariate and Transmission_Ln set as the dependent variable. We include Fan_Ln as a covariate to control its impact, given that the number of fans may influence the video transmission effect. Hence, we integrate it into the multi-factor ANOVA as a control variable. Within this multi-factor ANOVA, including the covariate, nine main effect groups and 247 interaction effect groups ranging from the second to eighth order, summing up to 256 groups. However, given the sample set’s constraints, some groups’ interaction effects are untestable. For instance, in the fifth-order interaction effect group, one combination, HD * VTT * VTST * ST * TQ, cannot be tested due to insufficient sample data within the original set to satisfy the group test requirements. Of the 256 groups, 162 are untestable, while 94 can undergo testing. Of these 94 groups, 13 passed the significance test (P < 0.05), with the results in Table 6.

Table 6 Results of multi-factors ANOVA (main effects and interaction effects).

The multi-factor ANOVA allows us to determine the significant main and interaction effects among the eight factors influencing the transmission effect (Transmission_Ln). As depicted in the preceding figure, three primary and ten interaction effects are significant. We used Python to clean and organize the original data files. As an illustration, in the VT x TQ interaction effect group, given the presence of three VT types and two TQ types, there are six possible VT x TQ combinations: VT1 x TQ1, VT1 x TQ2, VT2 x TQ1, VT2 x TQ2, VT3 x TQ1, and VT3 x TQ2. Thus, a new independent variable emerges as a VT x TQ combination, with its corresponding transmission effect value (Transmission_Ln) as the dependent variable. Subsequently, we applied a one-way ANOVA to determine the significant difference among the combinations. We assessed the optimal combination scheme via post hoc multiple comparisons and a mean value graph. Data processing for the other interaction effect groups follows the same pattern.

In summary, through the multi-factor ANOVA, it becomes evident that significant differences exist among the eight factors’ main effect and interaction effect groups. We then subjected these groups to one-way analysis of variance and post hoc multiple comparisons. When the number of levels is less than three and the number of instances at a certain level is below two, post hoc multiple comparisons are inapplicable, and decisions can be made directly via the mean value graph. The optimal scheme within the main effect and interaction effect groups is consequently filtered and ranked based on the partial Eta square (indicating the effect size) derived from the multi-factor ANOVA. Ultimately, the optimal scheme for the eight-factor combination is HD3 x VT2 x VTT1 x VTST1 x ST2 x T1 x CL1 x TQ1. The following Table 7 exhibits the detailed results. By mapping the encoding of the optimal solution to its real-world meaning, we define the optimal configuration as follows: resolution = 1080p or higher, video type = music, video title tone = general declarative, video title style = written, subtitle presence = with, hashtag presence = with, channel label presence = with, video duration = less than 10 min.

Table 7 13 One-way ANOVA.

Regression analysis

Before conducting the regression analysis, we initially affirm the linear relationship between the variables Recognition_Ln and Transmission_Ln through a scatterplot. Figure 3a demonstrates that, aside from a few exceptions, most Recognition_Ln values exhibit a strong linear relationship with the Transmission_Ln values. Subsequently, we examine the correlation between these two variables using the Pearson correlation coefficient. Table 8 reveals a strong correlation (0.903, P < 0.01) between the Recognition and Transmission variables, as illustrated by the Pearson correlation coefficient.

Fig. 3: Two scatterplots.
figure 3

a Recognition_Ln & Transmission_Ln scatterplot, b Regression standardized predicted values & regression standardized residual scatterplot. These two scatterplots represent the linear relationship between the two variables (Recognition_Ln and Transmission_Ln), and the regression residuals of regression analysis comply with the assumption of variance homogeneity.

Table 8 Correlation test.

As it is essential to evaluate the moderating effect of the categorical variable AT, the regression analysis includes an overall and group-specific regression founded on author type. Table 9 indicates that all regression models meet the set standard (P < 0.001), with Durbin Watson coefficients approximating two, signifying the independence of residuals and the absence of autocorrelation amongst independent variables. In the comprehensive regression, the adjusted R-square value is 0.815, and the regression coefficient is 0.855. The regression models for the three distinct author types exhibit standard fitting within the remaining group-specific regressions, with adjusted R-square values of 0.813, 0.976, and 0.560 and corresponding regression coefficients of 0.860, 0.890, and 0.651, respectively. This observation suggests that the author type, as a variable, exhibits a moderating effect on the influence relationship, altering the causal relationship between Recognition_Ln and Transmission_Ln variables across different author types. The following Table 9 displays the specific data. Figure 3b’s regression standardized predicted values and residual scatterplot demonstrates that most residuals in the regression align near the zero-residual line, with a few exceptions. This pattern suggests that the regression residuals comply with the assumption of variance homogeneity.

Table 9 The result of regression analysis.

Discussion

The results of the data analysis provide a clear basis for determining the validity of the 11 hypotheses mentioned above. The regression analysis results validate Hypotheses 1 and 2, indicating a significant proportionality between the degree of recognition and the effect of transmission, with the author type serving as a moderating variable in this influential relationship. The multi-factor ANOVA reveals that Hypotheses 4, 5, 6, 10, and 11 are invalid, except for Hypotheses 7, 8, and 9. In other words, considering these eight qualitative variables’ main effects and interaction effects and treating the Fans variable as a covariate, only ST, T, and CL amongst these eight qualitative variables exert a direct and significant influence on the transmission effect. Despite the lack of direct impact from the remaining five out of eight qualitative variables, the interaction effect test reveals that these variables are not entirely unaffected but require interaction with other variables to influence the communication effect.

These empirical findings elucidate the factors influencing video transmission related to ICH, specifically Cantonese Opera, on the Bilibili platform, thereby providing empirical evidence to enhance transmission effects. There is a clear trend that video is one of the most prevalent media through which people access and receive information today. Indeed, video consumption has become an integral part of modern life. Consequently, this medium is a crucial opportunity for preserving and propagating ICH in the current era of flourishing video-sharing platforms like YouTube, TikTok, and Bilibili. Our research’s significance lies in determining how to disseminate videos related to ICH widely.

Empirical testing based on the ELM theory reveals that the central path is predominant among the two influencing the communication effect. The adjusted R-square from the previous regression analysis and the partial Eta square from the multi-factor analysis of variance support this conclusion. It is evident that the degree of recognition profoundly influences the effect of communication. Although some main and interaction effects bear significance in the peripheral path, these effects appear relatively minor when considering the value of the partial Eta square.

Drawing upon these empirical results, we propose that adherence to the’ content is king’ principle is the overarching strategy for enhancing video communication’s impact on ICH. In simple terms, the producer should focus more on expressing the core content for a video presenting ICH content. This approach is recommended because, from the user’s perspective—especially considering Bilibili’s current cultural climate—users are more inclined to perform ‘one button three links’ for videos they endorse, thereby facilitating the evolution of these ICH in the new media formats of the current era.

While the empirical test supports “content is king” as a theoretically viable development strategy, it is clear that the production of high-quality, re-created ICH content is a long-term endeavor. Consequently, this study’s peripheral path proposes several specific and immediately implementable tactics for content creators. It becomes evident that the peripheral path primarily involves external measures in ICH videos such as video definition, subtitle inclusion, hashtag usage, channel selection for submission, title tone and style, video length, and the submission type of Cantonese opera videos. The peripheral path test provides theoretical guidance for these aspects. The multi-factor ANOVA, followed by One-way ANOVA and post hoc multiple testing, allows us to derive a theoretically optimal eight-factor combination model. This model recommends producing videos with 1080p or higher definition, positioning the type as music, using general declarative sentences for video title tone, adopting a written style for video titles, using hashtags, submitting to specific channels, adding subtitles, and ensuring a video length of less than ten minutes. It is crucial to note that this optimal combination derives from the analysis of Cantonese opera videos on Bilibili, implying that it may not be directly applicable when the specific ICH form or video platform differs, necessitating a similar analysis for precise guidance.

Employing the ELM theory, this study empirically establishes the central and peripheral paths’ impact on the communication effect of Cantonese Opera videos on Bilibili, thereby procuring macroscopic long-term development strategies and microscopic instant tactics. As previously discussed in the introduction and literature review, the preservation and transmission of ICH require enhancement, particularly concerning ICH aging, as noted by preceding researchers. The platform under scrutiny, Bilibili, is a cultural community primarily comprising young individuals with an average age of approximately 23 years. Consequently, studying this platform’s user preferences approximately equates to studying young individuals as the emerging primary target for ICH. If traditional ICH forms like Cantonese Opera can evolve on Bilibili, the issue of the aging communication group can be addressed to some extent.

Beyond the practical insights on Cantonese opera video communication, our study offers significant theoretical contributions. Using the ELM theory, we have developed and empirically tested an influence model tailored for Cantonese opera video communication, equivalent to applying the classic ELM theoretical model in a specific context. Empirical data also prove that the different persuasion methods proposed in the ELM model depend on the possibility of elaborate processing of the disseminated information. Regarding Cantonese opera video dissemination, superficial video features like subtitles or clarity necessitate minimal user engagement, making the peripheral route effective yet weak. Conversely, in-depth video elements trigger careful user engagement, strengthening the central route’s impact.

Contrasting with prior ELM video communication studies, our work introduces innovative metrics for video communication effectiveness. As mentioned earlier in this article, there are currently about seven communication effect indicators available for researchers to refer to on the front end of the Bilibili platform, and most studies are based on these seven indicators to define the communication effect of a video. Chen et al. provided a calculation method for the communication effect in their research: Cn = ln [0.5Pn + 0.3 (Ln + Mn + Sn) + 0.2 (Dn + R n)], where C is the information communication effect, P is the number of views, M is the number of coins, S is the number of collections, L is the number of likes, D is the number of comments, R is the number of comments, and the number in front of each indicator is the weight coefficient of each indicator (Chen et al., 2020). Our study diverges from Chen et al.‘s in two ways: omitting play volume due to its vulnerability to platform recommendation systems and integrating comment volume. Recommendation algorithms notably sway the playback counts of new media platforms. In other words, this part of the information is highly disruptive in studying video content and user preferences. Therefore, we do not use it as a direct indicator of communication effect evaluation. However, we use the ratio of the remaining indicators to the play volume as a quantitative indicator of the communication effect. The advantage of this processing is that it avoids some interference information that may exist in the playback volume.

Moreover, this calculation method is more conducive to calculating high-quality communication because the data of playback volume cannot explain the user’s actual viewing. For example, if a user quickly clicks and exits the video, almost no information is spread, but the playback volume will still increase. Although they provide varying weight coefficients, Cheng et al. suggest a similar methodology (Cheng and Guan, 2023). Similar to the study of Chen et al. they both provide weight coefficients for different indicators. One relies on the weight assignments of previous studies, and the other depends more on subjective judgment. To aim for a more rigorous approach, we consulted 31 experts and derived indicator weightings via the AHP. These derived weightings further enrich the academic discourse in this domain.

Additionally, juxtaposing “old” ICH with “young” platforms reveals a promising cultural preservation and engagement avenue. Young platforms like Bilibili, with their interactive and user-centric features, provide an ideal environment for ICH, such as Cantonese Opera, to reach a new, younger audience. This integration revitalizes traditional art forms and ensures their continuity in the digital age. The engagement of younger generations with ICH through platforms like Bilibili can address the challenge of aging audiences, providing a sustainable model for transmitting cultural heritage.

The findings suggest that young audiences on platforms like Bilibili are highly receptive to high-quality content that resonates with their preferences and behaviors. This receptiveness implies that traditional ICH can thrive if presented in ways that appeal to these digital natives. The success of Cantonese opera videos on Bilibili highlights the potential of digital platforms to serve as bridges between generations, fostering a renewed appreciation for traditional arts.

Conclusion and limitations

This study comprehensively analyzes the factors influencing the transmission of ICH videos, specifically Cantonese opera, on the Bilibili platform. By employing the ELM, we identified the predominant role of the central path in influencing communication effects, emphasizing the “content is king” principle. Our findings highlight that high-quality, content-focused videos are more likely to resonate with users and enhance the transmission effect.

Through quantitative methods such as the AHP, multi-factor ANOVA, and regression analysis, we determined that video recognition greatly influences the communication effect, with author type as a moderating variable. The multi-factor ANOVA results further identified an optimal eight-factor combination for video dissemination: 1080p or higher video definition, categorizing the video type as music, using general declarative sentences for video tone, employing a written style for video titles, attaching hashtags, submitting to specific channels, incorporating subtitles, and maintaining a video duration of less than 10 min. These insights offer macro-strategic direction and micro-level tactical execution for Cantonese Opera video creators.

Moreover, this study underscores the potential of young platforms like Bilibili to revitalize traditional art forms and ensure their continuity in the digital age. Engaging younger generations with ICH through digital platforms can address the challenge of aging audiences, providing a sustainable model for transmitting cultural heritage.

However, this study is not without its limitations:

  1. (1)

    Quantitative disparities: Our descriptive statistics reveal significant disparities in video quantities across different categories. For example, in the author type (AT) category, there are 1780 videos from ordinary users, but only 45 and 91 from professional and institutional users, respectively. These disparities impact the empirical data analysis results. However, this distribution reflects the prevalent content production modes on video-sharing platforms, where User-Generated Content (UGC) dominates over Professional User-Generated Content (PUGC) and Professional Generated Content (PGC).

  2. (2)

    Dataset limitations: Despite collecting 1916 Cantonese Opera videos using Python crawlers, the dataset may be incomplete due to the niche nature of ICH Cantonese Opera on a predominantly young platform. Consequently, some interaction effect groups need more data for robust testing. To address this limitation, we need to continue collecting and analyzing data as new videos are produced.

  3. (3)

    Cross-sectional data: The study utilizes cross-sectional data collected on March 15, 2023, reflecting a specific point in time. Future research should incorporate time-series data to provide a more dynamic understanding of the factors influencing ICH video transmission.

By acknowledging these limitations, we highlight areas for future research to refine further and improve the strategies for enhancing the transmission and preservation of ICH, ensuring that our communication strategy evolves in response to new data and technological advancements. In summary, this study primarily focuses on Cantonese Opera, explicitly aiming to enhance the inheritance and preservation of ICH in this new era. It specifically addresses how to optimally expand the breadth and depth of heritage in a time when video consumption has become a ubiquitous part of people’s daily lives.