联系我们

论文查询结果	返回搜索

论文编号：	14704
作者编号：	1120201116
上传时间：	2024/6/6 12:04:47
中文题目：	基于多模态特征融合计算的政务短视频用户认知参与 研究
英文题目：	Research on the Cognitive Engagement of Users in Government Short Videos based on Multimodal Feature Fusion Computation
指导老师：	王芳
中文关键字：	政务短视频；多模态；认知参与；深度学习；定性比较分析
英文关键字：	Government Short Videos; Multimodality; Cognitive Engagement; Deep Learning; Qualitative Comparative Analysis
中文摘要：	在数字化新媒体时代，短视频作为一种新兴的传播方式，已成为用户获取信息的重要途径。政务短视频是政府与民众沟通的新渠道，凭借其直观、生动的传播方式和广泛的覆盖范围，为政务信息传递带来新的变革。然而，政务短视频的传播效果参差不齐，公众的认知参与程度普遍较低，部分政务短视频因缺乏关注和互动逐渐在大众视野中消失，未能达到预期效果。这不仅造成行政资源的浪费，还可能对政府形象产生不利影响。 在众多的政务信息传递方式中，政务短视频以其多模态特征——即视频内容中融合了图像、文字、声音等多种信息呈现方式——有效吸引了用户的注意力并激发他们的思考与参与互动兴趣，使用户能够更充分地了解政务信息的具体内容和实施细节，大大提升了政策信息的传播效果。然而，关于多模态特征如何影响用户认知参与的内在机制，现有研究尚缺乏系统理论进行深入分析。此外，如何有效地提取和量化这些多模态信息，以及如何将它们科学地应用于政务短视频的价值评估中，仍是亟待解决的研究问题。 围绕上述问题，本文聚焦于以下四个研究问题：（1）影响政务短视频用户认知参与的因素有哪些？相关理论要素之间有什么样的作用机制？（2）政务短视频的多模态信息有哪些？如何进行特征提取和量化？（3）如何通过多模态特征的融合计算评估政务短视频的价值？（4）影响政务短视频用户认知参与的关键因素有哪些？其组态模式如何？对这些问题的回答，在理论上可以深化多模态信息用户认知参与的理论研究，在技术上可以为多模态特征融合计算提供具体的场景检验，在实践上则可以为政务短视频制作和传播策略的优化提供依据。 本文的主要研究工作包括：第一，通过实验和访谈了解影响政务短视频用户认知参与的因素。使用扎根理论方法，构建一个全面的理论模型以揭示其内在机制。第二，分析政务短视频多模态信息的类别和特征，设计出一套行之有效的多模态特征提取方法。第三，构建价值评估计算模型，对评论文本进行价值分类和评估。为此，本文设计并实现了一个融合多模态特征、ALBERT和TextCNN深度学习模型的政务短视频多标签价值分类方法，并建立针对政务短视频价值的标准语料库，为政务短视频的价值评估提供新的思路和方法。第四，对2238条政务短视频进行多模态特征提取和分析，采用相关性分析和探索性因子分析方法验证并调整前序章节构建的理论模型，运用结构方程模型揭示多模态特征对用户认知参与的影响。第五，采用定性比较分析（QCA）方法，进一步深化政务短视频用户认知参与的影响因素分析，揭示多模态特征与用户价值评估在共同塑造用户认知参与中的多样化组态模式。 本文的研究发现如下：第一，研究将政务短视频的多模态特征分为八种类别，包括内容主题、叙事节奏、信息量、图像色彩、视频情感、视觉复杂性和音乐感染力。第二，政务短视频用户表现出从感知、认知到参与的连续过程。短视频的多模态特征不仅影响用户的初步感知和情感反应，还进一步影响他们的认知理解和后续的互动行为，如点赞、评论、转发和收藏。第三，通过设计融合多模态特征的ALBERT-TextCNN模型，从信息价值、道德价值、审美价值和情感价值四个维度对政务短视频进行价值评估，检验该模型的有效性。第四，通过结构方程模型和定性比较分析（QCA），验证了多模态特征与视频价值、用户行为之间的复杂组态关系。 本文的创新点主要体现在以下三方面： 在理论层面，构建了一个政务短视频用户认知参与的扎根理论模型，揭示了政务短视频的多模态特征、视频价值、用户个体特征以及社会环境共同影响用户认知参与的作用机制。相较于已有研究，该模型提出短视频价值评估在用户认知参与中的重要作用，探讨了个体特征和社会环境对这一过程的调节作用，为理解政务短视频用户认知参与提供了新的理论视角。 在实践层面，提出一种融合多模态特征、ALBERT和Text-CNN深度学习模型的政务短视频价值评估方法，结合图片、音乐和文本特征，通过多标签文本分类任务的性能评估，揭示了多模态信息融合在信息分类中的有效性。此外，本文对于视频价值的分类定义以及标准语料库的建立，也为政务短视频的价值评估提供了形式化的表示依据与数据支撑。 在方法层面，采用结构方程模型，剖析了政务短视频的多模态内容特征与用户行为之间的复杂动态关系，在一定程度上避免了传统单一模态或线性影响模型的局限。此外，本文将QCA方法应用于政务短视频用户认知参与的研究中，通过“组态视角”进行系统分析，为类似研究提供了方法论参考。 本文共包含图16幅，表37个，参考文献410篇。
英文摘要：	In the era of digital new media, short videos have emerged as a vital medium for users to access information. Government short videos, as a new channel for communication between the government and the public, bring new transformations to the transmission of government information with their intuitive and vivid presentation and wide coverage. However, the dissemination effectiveness of these videos is inconsistent, with public cognitive engagement generally low. Some government short videos gradually fade from public view due to a lack of attention and interaction, failing to achieve the desired impact. This not only wastes administrative resources but can also negatively affect the government's image. Among various government information dissemination methods, government short videos effectively attract users' attention and stimulate their interest and participation through their multimodal characteristics—integrating images, text, sound, and other information presentation methods. This helps users better understand the specific content and implementation details of government information, significantly enhancing the dissemination effect of policy information. However, current research lacks systematic theoretical analysis on how these multimodal features influence user cognitive engagement. Additionally, there are pressing research issues on how to effectively extract and quantify these multimodal features and apply them scientifically in evaluating the value of government short videos. To address these issues, this paper focuses on four research questions: (1) What are the multimodal information characteristics of government short videos? How can these features be extracted and quantified? (2) What factors influence user cognitive engagement with government short videos? What are the interaction mechanisms among related theoretical elements? (3) How can the value of government short videos be evaluated through the integration of multimodal features? (4) What are the key factors influencing user cognitive engagement with government short videos, and what are their configurational patterns? Answering these questions can deepen the theoretical research on multimodal information user cognitive engagement, provide practical scenarios for multimodal feature fusion computation, and offer guidance for optimizing the production and dissemination strategies of government short videos. The main research work of this paper includes: First, understanding the factors influencing user cognitive engagement with government short videos through experiments and interviews, and constructing a comprehensive theoretical model using grounded theory to reveal the underlying mechanisms. Second, analyzing the categories and characteristics of multimodal information in government short videos and designing an effective multimodal feature extraction method. Third, developing a value evaluation computation model for categorizing and assessing comment text, and designing and implementing a multi-label value classification method for government short videos using a combination of multimodal features, ALBERT, and TextCNN deep learning models, along with establishing a standard corpus for the value of government short videos. Fourth, extracting and analyzing multimodal features from 2238 government short videos, validating and adjusting the theoretical model using correlation analysis and exploratory factor analysis, and using structural equation modeling to reveal the impact of multimodal features on user cognitive engagement. Fifth, employing Qualitative Comparative Analysis (QCA) to further deepen the analysis of factors influencing user cognitive engagement with government short videos, revealing the diverse configurational patterns of multimodal features and user value assessments in shaping user cognitive engagement. The research findings are as follows: First, the multimodal features of government short videos are classified into eight categories, including content theme, narrative rhythm, information volume, image color, video emotion, visual complexity, and musical appeal. Second, users of government short videos exhibit a continuous process from perception and cognition to participation. The multimodal features of short videos influence not only users' initial perception and emotional response but also their cognitive understanding and subsequent interactive behaviors such as liking, commenting, sharing, and saving. Third, the ALBERT-TextCNN model designed to integrate multimodal features effectively evaluates the value of government short videos from four dimensions: information value, moral value, aesthetic appreciation, and emotional experience. Fourth, using structural equation modeling and QCA, the study validates the complex configurational relationships among multimodal features, video value, and user behavior. The innovations of this paper are primarily reflected in the following three aspects: Theoretically: A grounded theory model of user cognitive engagement with government short videos is constructed, revealing the interaction mechanisms of multimodal features, video value, individual user characteristics, and social environment in influencing user cognitive engagement. Compared to existing research, this model highlights the important role of video value evaluation in user cognitive engagement and explores the moderating effects of individual characteristics and social environment, providing a new theoretical perspective for understanding user cognitive engagement with government short videos. Practically: A method for evaluating the value of government short videos by integrating multimodal features, ALBERT, and Text-CNN deep learning models is proposed, demonstrating the effectiveness of multimodal information integration in information classification through performance evaluation of multi-label text classification tasks. Additionally, the classification definition of video value and the establishment of a standard corpus provide formalized representation basis and data support for the value evaluation of government short videos. Methodologically: Using structural equation modeling, this paper analyzes the complex dynamic relationships between multimodal content features of government short videos and user behavior, addressing the limitations of traditional unimodal or linear influence models. Furthermore, the application of QCA in the study of user cognitive engagement with government short videos offers a systematic analysis from a configurational perspective, providing methodological references for similar research. This dissertation contaions: 16 pictures, 37 tables, 410 references.
查看全文：	预览  下载(下载需要进行登录)