多媒體的影響隨著科技進展與日俱增，了解多媒體所隱含的豐富資訊即是對於社會脈動主流走向與即時資訊的掌握，隨著人工智慧技術與深度學習的應用開啟了研究多媒體的大門，使得我們可以用多角度、橫跨多模態的方式綜合性描述理解與追蹤社會背後的潛在思維， 並可延伸至反映人類文化與歷史的媒材 – 電影的研究，並透過社會學、人類學 …等等的研究精神為主軸切入，使用工程的手段來更全面的了解電影所反映的人類文化，並試圖回答過去技術難以分析的問題。
另外，在視覺化的過程中。也發現許多有趣的現象符合社會偏見與常識性解釋。
本論文使用多模態鑲嵌表徵學習的技術來展現其對於多媒體資訊的掌握能力，
在架構上分成兩個子任務來驗證，第一是屬於近代研究，利用蒐集社群媒體的推文
資訊來分析 2016年美國總統大選，該模型因加入多模態特徵與特殊設計的目標函
數，在貼文屬性預測 上提升了 8%，且在視覺化鑲嵌表徵看見模型對於當年大選美
國各政黨的選舉策略差異； 第二是屬於橫跨世紀的研究，利用結構主義對電影文本
的探討來設計所需的特徵，並使用圖論捲積網路 (Graph Convolutional Network)來做時代性的分類訓練 最後 在 準確率 上 提高 6%，並且在 分類結果上看見電影黃金年代的特性與文化上的變化，最後試圖從資料的角度來探討社會偏見與潛規則 。

The impact of multimedia is increasing with the advancement of science and technology. Understanding the rich information implied by multimedia is the mastery of social pulsation and real-time information. With the application of artificial intelligence technology and deep learning, the door to research multimedia has opened, so that we can Comprehensively describe and trace the underlying thinking behind society in a multi-angle, multi-modal way, and extend to media that reflects human culture and history – film research, through sociology, anthropology, etc. The spirit of research is the spindle cut-in, using engineering methods to more fully understand the human culture reflected in the film, and trying to answer questions that were difficult to analyze in the past. In addition, in the process of visualization. Many interesting phenomena have also been found to be consistent with social prejudice and common sense interpretation.
This paper uses multi-modal embedding representation learning technology to demonstrate its ability to master multimedia information. It is divided into two sub-tasks to verify the structure. The first is modern research, which uses the tweet information collected by social media to analyze. In the 2016 US presidential election, the model increased the 8% attribute prediction by adding multimodal features and specially designed objective functions, and visualized mosaic representation to see the model's electoral strategy differences for the US political parties in the election year. The second is a study that spans the century. It uses structuralism to explore the characteristics of the film to design the required features, and uses the Graph Convolutional Network to do the classification training of the times. The rate of improvement is 6%, and the characteristics and cultural changes of the golden age of the film are seen in the classification results. Finally, we try to explore social prejudice and unspoken rules from the perspective of data.

誌謝
i
中文摘要
ii
ABSTRACT iii
Chapter 1 INTRODUCTION 1
Chapter 2 POLITICAL SEMENTIC SPACE LEARNING 3
CH 2.1 INTRODUCTION 3
CH 2.2 METHODOLOGY 5
2.2.1 Dataset: Twitters Political Posts ……………………………………...5
2.2.2 Content and Profile Descriptors ……………………………………...6
2.2.3 Network Architecture: Training Objective ………………………..….7
2.2.3.1 Cross-instance Distance Constraint …………………………...8
2.2.3.2 Within-instance Semantic-preserving Constraint ……………..8
CH 2.3 EXPERIMENAL RESULT AND ANALYSIS .10
2.3.1 Experimental Settings ……………………………………………….10
2.3.2 Experiment 1: Message Types Classification Result …………...…...10
2.3.3 Experiment 2: Political Party Classification Result …………………11
CH 2.4 CONCLUSIONS 14
Chapter 3 MOVIE SYMBOLIC SPACE LEARNING 15
CH 3.1 INTRODUCTION 15
CH 3.2 METHODOLOGY 17
3.2.1 Dataset: Movie Scenes ……………………………………………...17
3.2.2 Feature From Symbolic Perspective And Label Setting …………....18
3.2.3 Graph-based Supervised Learning ………………………………….19
CH 3.3 EXPERIMENAL RESULT AND ANALYSIS ........................................... 26
3.3.1 Experimental Settings …………………………………….………...26
3.3.2 Decades Classification Result …………………………….………...27
3.3.3 Visualization and Analysis ………………………………………….28
3.3.4 Prevailing Culture Trend Implication From Cinema Content ……....29
CH 3.4 CONCLUSIONS ......................................................................................... 36
Chapter 4 CONCLUSIONS ...................................................................................... 37
REFERENCE ............................................................................................................ 38

[1] Zhiwei JinEmail authorJuan CaoHan GuoYongdong ZhangYu WangJiebo Luo. Detection and Analysis of 2016 US Presidential Election Related Rumors on Twitter.
[2] Alessandro Bessi, Emilio Ferrara. Social Bots Distort the 2016 US Presidential Election Online Discussion.
[3] Erika Frankin Fowler, Travis N. Ridout, Michael M. Franz. Political Advertising in 2016: The Presidential Election as Outlier?
[4] Mamata Murthi, Anne-Catherine Guio and Jean Dreze. Mortality, Fertility, and Gender Bias in India: A District-Level Analysis.
[5] Nancy Folbre, Marjorie Abel. Women’s Work and Women’s Households: Gender Bias in the U.S. Census. [6] Megan Fountain, Social Media and its Effects in Politics: The Factors that Influence Social Media use for Political News and Social Media use Influencing Political Participation, Ph.D. thesis, The Ohio State University, 2017. [7] Joseph Kahne and Benjamin Bowyer, “The political significance of social media activity and social networks,” Political Communication, pp. 1–24, 2018. [8] Bente Kalsnes, Anders Olof Larsson, and Gunn Enli, “The social media logic of political interaction: Exploring citizens and politicians relationship on facebook and twitter,” First Monday, vol. 22, no. 2, 2017.Skinner, Burrhus Frederic. Science and human behavior. Simon and Schuster, 1953. [9] Daniel Halpern, Sebastia ́n Valenzuela, and James E Katz, “We face, i tweet: How different social media influence political participation through collective and internal efficacy,” Journal of Computer-Mediated Communication, vol. 22, no. 6, pp. 320–336, 2017. [10] Kaitlin Vonderschmitt, “The growing use of social media in political campaigns: How to use facebook, twitter and youtube to create an effective social media campaign,” 2012. [11] Ibrahim Uysal and W Bruce Croft, “User oriented tweet ranking: a filtering approach to microblogs,” in Proceedings of the 20th ACM international conference on Information and knowledge management. ACM, 2011, pp. 2261–2264. [12] John Wilkerson and Andreu Casas, “Large-scale computerized text analysis in political science: Opportunities and challenges,” Annual Review of Political
39
Science, vol. 20, pp. 529–544, 2017. [13] Matthew J Denny and Arthur Spirling, “Text preprocessing for unsupervised learning: why it matters, when it misleads, and what to do about it,” Political Analysis, vol. 26, no. 2, pp. 168–189, 2018. [14] Benjamin Klein, Guy Lev, Gil Sadeh, and Lior Wolf, “Fisher vectors derived from hybrid gaussian-laplacian mixture models for image annotation,” arXiv preprint arXiv:1411.7399, 2014. [15] Zhuang Ma, Yichao Lu, and Dean Foster, “Finding linear structure in large datasets with scalable canonical correlation analysis,” in International Conference on Machine Learning, 2015, pp. 169–178. [16] Yunchao Gong, Qifa Ke, Michael Isard, and Svetlana Lazebnik, “A multi-view embedding space for modeling internet images, tags, and their semantics,” Inter- national journal of computer vision, vol. 106, no. 2, pp. 210–233, 2014. [17] Tao Chen, Xiangnan He, and Min-Yen Kan, “Context-aware image tweet modelling and recommendation,” in Proceedings of the 2016 ACM on Multimedia Conference. ACM, 2016, pp. 1018–1027. [18] Chao Wu, Jia Jia, Wenwu Zhu, Xu Chen, Bowen Yang, and Yaoxue Zhang, “Affective contextual mobile recommender system,” in Proceedings of the 2016 ACM on Multimedia Conference. ACM, 2016, pp. 1375–1384. [19] Yunchao Gong, Liwei Wang, Micah Hodosh, Julia Hockenmaier, and Svetlana Lazebnik, “Improving image-sentence embeddings using large weakly annotated photo collections,” in European Conference on Computer Vision. Springer, 2014, pp. 529–545. [20] David Laniado and Peter Mika, “Making sense of twitter,” in International Semantic Web Conference. Springer, 2010, pp. 470–485. [21] Xuelong Li, Bin Zhao, and Xiaoqiang Lu, “Mam-rnn: multi-level attention model based rnn for video captioning,” in Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence, 2017. [18] Junxuan Chen, Baigui Sun, Hao Li, Hongtao Lu, and Xian-Sheng Hua, “Deep ctr prediction in display advertising,” in Proceedings of the 2016 ACM on Multimedia Conference. ACM, 2016, pp. 811–820.
[19] AngieKBeeman.2007. Emotional segregation: A content analysis of institutional
40
racism in US films,1980–2001. Ethnic and Racial Studies 30,5(2007),687–712
[20] Gina Escamilla, Angie L Cradock, and Ichiro Kawachi.2000.Women and smoking in Hollywood movies: A content analysis. American Journal of Public Health 90, 3(2000),412.
[21] Frankl in Fearing.1947. Influence of the Movies on Attitudes and Behavior. The ANNALS of the American Academy of Political and Social Science 254,1(1947), 70–79.
[22] Thomas N Kipf and Max Welling.2016. Semi-supervised classification with graph convolutional networks. arXiv preprint arXiv:1609.02907 (2016).
[23] Robert Lapsley and Michael Westlake.2006. Film theory: An introduction. Manchester University Press.
[24] Jim R Macnamara etal. 2005. Media content analysis: Its uses, benefits and best practice methodology. Asia Pacific Public Relations Journal 6,1(2005),1.
[25] Victor R Martinez, Krishna Somandepalli, Karan Singla, Anil Ramakrishna, YaldaTUhls, and Shrikanth Narayanan.2019. Violence Rating Prediction from MovieScripts. (2019).
[26] Brooke S O’Neil. 2016. Approved for All Audiences: A Longitudinal Content Analysis of the Portrayal of Women in Movie Trailers. (2016).
[27] MarušaPušnik.2015. Cinema culture and audience rituals: Early mediatisation of society. Anthropological Notebooks 21,3(2015).
[28] Edgar H Schein.1991. What is culture. Newbury Park, CA: Sage Publicatios (1991), 243–253.
[29] Manu Sharma. 2014. Transnational Cinema: A Cross Culture Communication Medium. J Mass Communical Journalism4,219(2014),2.
[30] Eirini Sifaki. 2003. Projections of Popular Culture through the Study of the CinemaMarketinContemporaryGreece.In1st LSE PhD Symposium on Modern Greece Current Social Sciences Research on Greece, Hellenic Observatory, London School of Economics and Political Science, on June,Vol.21.
[31] Sameet Sreenivasan. 2013. Quantitative analysis of the evolution of novelty in cinema through crowdsourced keywords. Scientific reports 3(2013),2758.
41
[32] Andrew Vassiliou. 2006. Analysing film content: A text-based approach. Ph.D. Dissertation.Citeseer.