FINA2386 Network analysis
1. Network of Interest, Hypotheses, and Project Design
This study examines the actor-director collaboration network within Hong Kong action films from 2019 to 2024, using data manually collected from Douban’s international version. Our goal is to explore how collaboration patterns influence actor careers and illuminate the dynamics within Hong Kong's film industry.
Relevance of Network Analysis:
Hong Kong's film industry, often described as the "entertainment circle," is deeply rooted in East Asian "guanxi (connection)" culture, where social ties, loyalty, and mentor-protégé relationships heavily influence career trajectories. Prominent groups like Jackie Chan’s "Sing Ga Ban" exemplify these tight-knit networks, built on repeated collaborations and strong bonds. Using social network analysis (SNA), we can quantify and reveal key relationships and community structures driving success and continuity within this unique industry network.
Motivation:
Our motivation stems from Hong Kong's action cinema's reliance on deep social and professional connections, which shape both artistic output and career development. By studying these networks, we aim to uncover the key figures and collaborative dynamics that define success and influence in this cultural and professional space.
Hypotheses:
● H1: Directors with high degree centrality (those who collaborate with a large number of actors) play a key role in linking different actor groups within the industry.
● H2: Actors frequently working with influential directors tend to secure more roles and gain greater industry prominence.
● H3: Groups of actors often working with the same directors form. identifiable communities, reflecting consistent film styles or recurring casting patterns.
Project Design: We constructed a two-mode network where nodes represent actors and directors, and edges denote their collaborations in films. We then analyzed the projected actor-actor network to examine indirect collaborations (actors linked through shared directors). Key metrics such as centrality and community detection methods were used to identify collaboration patterns and key figures within the network.
2. Data Collection, Cleaning, and Analysis Process
Data Collection: Data was manually collected from Douban’s international version, focusing on Hong Kong action films released between 2019 and 2024. The collected data included:
● Movie Titles: Film names.
● Release Years: Year of release.
● Genre Verification: Ensured the films fit the action genre.
● Cast and Directors: Key actors and directors involved.
Data Collection Strategy:
● Selection Criteria: Films were selected based on their classification as Hong Kong action films.
● Manual Entry: Data was meticulously recorded in structured tables for consistent analysis, ensuring accuracy through manual verification.
Data Cleaning:
1. Duplicate Removal: Removed duplicates and standardized movie titles.
2. Missing Data Handling: Supplemented missing information with additional searches or excluded entries where necessary.
3. Verification: Cross-checked entries using Douban’s detailed information to ensure data accuracy and consistent genre classification.
Visualization——Network Graphs: We used visualization tools to create network graphs illustrating the actor-director network and the projected actor-actor network. These graphs provided a visual representation of the collaboration dynamics, highlighting key nodes (actors and directors) and the density of their connections.
Network Construction:
1. Two-Mode Network Construction: In order to capture the direct partnership between actors and directors and the structural characteristics of the film industry, we chose the two-mode network.
We constructed a bipartite network where nodes represented both actors and directors, and edges indicated their collaborations in Hong Kong action films. Each edge connecting an actor to a director represented their joint involvement in a particular film. This two-mode network served as the foundation for our analysis.
2. Projection to Actor-Actor Network: The two-mode network is projected to a one-mode actor-actor network, where nodes represent actors and edges connect actors who have worked with the same director. The weights of the edges represent the number of collaborative movies between the two actors, reflecting the frequency of collaboration between the two actors, and through the performance of the actor centrality one can understand the actor's status in the Hong Kong action movie and television industry.
3. Centrality Metrics:
○ Degree Centrality: Identified the most connected actors and directors based on collaboration frequency.
○ Betweenness Centrality: Highlighted actors who served as bridges between different groups.
○ Eigenvector Centrality: Measured influence by connecting to other prominent actors.
4. Community Detection: Used algorithms to identify clusters of frequently collaborating actors, reflecting potential preferences or industry trends.
3. Results, Conclusions, and Potential Problems with the Analysis
Results:
● H1 Validation: Certain directors displayed high degree centrality, connecting various actor groups and serving as key industry hubs.
● H2 Support: Actors with high centrality, often working with influential directors, demonstrated greater industry presence and opportunities.
● Community Structures: Community detection revealed clusters of actors consistently appearing together under certain directors, highlighting potential patterns in casting and collaboration preferences.
Conclusions: Our analysis of Hong Kong action films from 2019 to 2024, using data from Douban’s international version, demonstrated how directors shape collaboration dynamics and influence actor trajectories. Directors often serve as pivotal figures, while actors' prominence correlates with their network connections through these collaborations.
Potential Problems:
1. Manual Data Collection Limitations: Manual collection posed a risk of human error and incomplete data. Future work could involve automated scraping (if allowed).
2. Data Completeness: The Douban dataset might not cover all relevant Hong Kong action films, creating potential gaps.
3. Genre Classification: Genre boundaries can be fluid, which may lead to inclusion of films with mixed genres.
4. Role Significance: All collaborations were treated equally, though different roles (lead vs. supporting) may carry varying levels of influence.
4. Contributions of Each Group Member
● [Member A]: Responsible for collecting data from Douban’s international version and verifying data accuracy.
● [Member B]: Focused on network construction, projections, and centrality calculations.
● [Member C]: Conducted hypothesis testing, results interpretation, and created visualizations.
● [Member D]: Handled report writing, ensuring a clear structure, and contributed to identifying challenges and future directions.
Expanded Analysis Process
Network Construction:
1. Two-Mode Network Creation: We constructed a bipartite network where nodes represented both actors and directors, and edges indicated their collaborations in Hong Kong action films. Each edge connecting an actor to a director represented their joint involvement in a particular film. This two-mode network served as the foundation for our analysis.
2. Projection to Actor-Actor Network: The two-mode network was projected into a one-mode actor-actor network, where nodes represented actors, and edges connected actors who had worked with the same director. The edge weight denoted the number of films in which two actors collaborated under the same director, providing insight into the strength of these collaborative ties.
Centrality Metrics:
1. Degree Centrality: This metric was used to identify the most well-connected actors and directors in the network. High degree centrality for a director indicated their collaboration with numerous actors, suggesting their pivotal role in connecting different groups within the industry. For actors, high degree centrality reflected frequent collaborations, potentially increasing their visibility and opportunities.
2. Betweenness Centrality: We computed betweenness centrality to identify actors who acted as bridges between different clusters of actors through their shared work with various directors. High betweenness indicated that these actors might play a crucial role in connecting disparate groups within the network, suggesting greater influence or flexibility in their career trajectories.
3. Eigenvector Centrality: This metric evaluated an actor's influence based on their connections to other highly central actors. An actor with high eigenvector centrality was not only well-connected but was also connected to other influential figures in the industry, potentially enhancing their reputation and prominence.
Community Detection: Using modularity-based community detection algorithms, we identified clusters of actors who frequently collaborated under the direction of the same or similar directors. These communities provided insight into recurring collaborative patterns and suggested how certain directors tended to work with specific groups of actors, possibly due to stylistic preferences, casting choices, or industry practices.
1.
Findings from the Analysis:
1. Influential Directors: Certain directors demonstrated high degree centrality, indicating their central role in bringing together diverse groups of actors. These directors often acted as hubs, facilitating collaborations and influencing actor communities.
2. Prominent Actors: Actors who frequently collaborated with these influential directors showed high network centrality, supporting our hypothesis that strong director relationships correlate with greater career opportunities and visibility.
3. Community Patterns: The detected communities revealed that some actor groups were repeatedly cast by the same directors, suggesting the existence of preferred collaborations that may shape genre-specific subgroups within the Hong Kong film industry.
Challenges and Limitations in the Analysis:
1. Edge Weights and Role Significance: In our network, all collaborations were treated equally, which may overlook the varying significance of roles (e.g., lead roles vs. minor roles). Future analyses could introduce weighted edges to better capture these distinctions.
2. Incomplete Data: The accuracy of our findings depended heavily on IMDb data. Incomplete or missing data, particularly for less well-known films, may have affected the comprehensiveness of our analysis.
3. Genre Classification Issues: Some films listed as "action" may have blended genres, potentially introducing noise into the dataset. Future studies could involve more granular genre classifications to ensure a focused analysis.