A Chinese Tale of Three Regions: A Century of China in Thousands of Films

This research investigates the portrayal of China in films from mainland China, Taiwan, and Hong Kong from 1949 to 2018. The study's central question revolves around how the image of China has evolved across these three distinct regions, each with its unique historical and sociopolitical context, yet all sharing a common cultural root. The film industries of these regions, while largely independent, have interacted and influenced each other over time, creating a complex tapestry of cinematic representations of China. This study aims to illuminate these diverse portrayals using big data analysis and machine learning techniques, offering a novel perspective on the interplay between national identity, cultural exchange, and the global dissemination of China's image through film. The importance of this study lies in its capacity to unveil hidden cultural phenomena using large-scale data analysis, thereby overcoming the limitations of traditional case study methodologies which may lack objectivity and generalizability. By meticulously analyzing film content, topics, and sentiments, the researchers aim to shed light on the cultural sociology of Chinese cinema and how it has interacted with global audiences.

Existing scholarship on Chinese cinema often presents a linear historical narrative, focusing on a unified national identity and overlooking the diversity of cultures and experiences within China. This study challenges this approach, adopting a postcolonial lens to analyze the varied representations of China across mainland China, Taiwan, and Hong Kong. The authors review existing literature that highlights the impact of colonialism and postcolonial identity on film production in these regions. They discuss the influence of Western cultural systems and the challenge of balancing globalization with national identity in Chinese films. The study critically examines the political instrumentality of film in the PRC, the identity crisis reflected in Taiwanese cinema, and the commercial nature of Hong Kong's film industry, drawing on existing research to establish a theoretical framework for their analysis.

The research employed a quantitative methodology centered on big data analysis of film synopses from the Internet Movie Database (IMDb). The dataset encompassed 1047 films produced in mainland China, Taiwan, and Hong Kong before 2019. Three core analytical techniques were employed. First, Word2vec, a word embedding technique, was used to analyze the contextual associations of the word “China” within film synopses, revealing the most frequently associated words and their semantic connections in each region's films. The skip-gram algorithm was used to create a 128-dimensional word vector model based on the film plot corpus, using "China" as the central word and calculating the cosine distance to other words in the corpus to determine the similarity of words to "China". Secondly, Latent Dirichlet Allocation (LDA), a topic modeling technique, was utilized to identify the prevalent themes or topics in the films. The LDA model facilitated the identification of five key topics: Kung Fu, Kinship & Love, Rural & Urban, War, and Crime. This method allowed the researchers to examine the distribution of topics across the three regions and over time. The formula used for the LDA model is provided in the paper, illustrating how the conditional distribution of hidden variables (subject structure) within given observed variables (words in the document) was calculated. Thirdly, Google's sentiment analysis API was used to assess the overall sentiment expressed in the film synopses, providing a quantitative measure of the positive or negative attitudes toward China expressed in the films. This API used a scoring system from -1.0 (negative) to +1.0 (positive) to denote the sentiment expressed in the text. A spatiotemporal analysis framework was constructed to explore changes in the conception of China based on content, topics, and sentiments in films from mainland China, Taiwan, and Hong Kong. An ANOVA and LSD test were used to further compare differences between the three regions in the proportions of the five identified topics and the overall sentiment scores. Moving averages were calculated to smooth out extreme values and better understand trends over time.

The analysis revealed significant variations in the portrayal of China across the three regions. Word2vec analysis showed that the words most closely associated with "China" differed markedly. Mainland Chinese films showed strong associations with political terms, expressions related to economic development, and those focused on minority groups. Taiwanese films were more centered around everyday life, family, and cross-strait relationships, with notable mentions of Japan and war. Hong Kong films emphasized the region's unique political status and included terms related to crime, kung fu, and economic exchange with mainland China. The LDA topic analysis showed that while the "Kinship and Love" theme was prevalent across all three regions, there were significant differences in the proportions of the other topics. Mainland China exhibited a higher proportion of films on the "Kinship and Love" and "War" topics, whereas Hong Kong showed a higher proportion of films on "Kung Fu" and "Crime." Taiwan had a slightly higher proportion of films on the "War" topic than Mainland China. Furthermore, the analysis showed a temporal variation in the prevalence of certain topics, particularly "War," influenced by major historical events and political shifts in each region. The sentiment analysis revealed that overall sentiment towards China was positive across all three regions, though Hong Kong exhibited a notably lower average sentiment score compared to mainland China and Taiwan. The sentiment scores showed temporal fluctuations, closely correlated with significant political and economic events, demonstrating the strong influence of socio-political context on film sentiment. In Mainland China, high positive sentiments were observed during periods of political campaigns, followed by negative sentiments during periods of political turmoil such as the Great Leap Forward and the Cultural Revolution. A similar pattern was seen in Taiwan, with peaks in positive sentiment during periods of improved cross-strait relations and lows during times of tension. Hong Kong’s sentiment scores were more volatile, often shifting opposite to mainland China, reflecting the unique challenges of the region's identity and political status after the handover to China.

The findings of this study highlight the significant role of socio-political context in shaping the cinematic representation of China across its different regions. The variations in word associations, topic distribution, and sentiment scores reflect the complex interplay between national identity, historical memory, political ideologies, and economic circumstances. The study’s methodology contributes to cultural sociology by demonstrating the value of large-scale quantitative analysis in understanding cultural representations, complementing traditional qualitative methods. The insights gleaned into the distinct yet interconnected identities of mainland China, Taiwan, and Hong Kong provide a nuanced understanding of the multifaceted nature of Chinese culture and its evolving relationship with the global community. The study's postcolonial perspective underscores the enduring influence of historical events and colonial legacies on contemporary representations of China, reinforcing the importance of considering historical context when interpreting cultural productions.

This study provides a comprehensive analysis of the representation of China in films from mainland China, Taiwan, and Hong Kong over a century, using big data analysis and machine learning techniques. The findings reveal significant regional variations in the portrayal of China, influenced by unique historical and socio-political contexts. The study's contribution lies in its innovative methodology, its nuanced understanding of the complex relationship between the three regions, and its insights into the interplay between politics, economics, and culture in shaping national image. Future research could explore the impact of specific policy decisions or events on film production, expand the geographical scope of the study to other regions, and investigate the reception of these films by international audiences to gain a fuller understanding of how the image of China is perceived and constructed globally.

The study acknowledges certain limitations. Firstly, the focus on IMDb data may introduce selection bias, as the editors and users of this platform may not fully represent the diverse range of films produced and consumed globally. Secondly, the study did not explicitly analyze co-productions, which may represent a significant area of interaction between film industries. Finally, while the study explores correlation between film representation and socio-political events, it does not establish direct causal relationships. Further research incorporating additional datasets, qualitative analysis and deeper exploration of causal relationships are needed to address these limitations.