基于深度学习卷积神经网络的电影票房预测

发布时间：2018-05-18 01:37

本文选题：电影票房 + 实证分析　；参考：《首都经济贸易大学》2017年硕士论文

【摘要】：随着国民经济快速发展,人们的精神和文化生活也不断丰富。电影是文化生活的重要组成,它不仅丰富了人们的精神世界,而且还促进了中国影视文化产业的发展。电影是我国文化产业的中流砥柱,给国家的发展带来了巨大的经济收益,而衡量电影经济收益最重要且最直观的一个指标就是电影票房收入。中国电影市场发展迅猛,2016年电影票房总收益已达457.12亿元。因此,依据我国电影市场的特点,研究电影票房的影响因素对我国电影的投资决策具有重要的参考价值。目前已有大量电影票房预测研究的相关文献,但大部分都是在传统的统计方法和传统神经网络层面上分析。多元线性回归模型在预测精度上不如神经网络,但是回归可以清楚的解释影响因素对票房的影响程度;神经网络模型在电影票房的预测上的精确度要高于回归模型,但是各变量对电影票房的影响程度不好解释。因此,神经网络模型被称为“黑盒子”。本文的研究目标是基于中国电影市场特性,依据历史电影票房数据,创新性的提出将深度学习模型中的卷积神经网络应用在电影票房预测中。论文主要使用多元线性回归、BP神经网络和卷积神经网络(CNN)分别对电影票房样本数据建立模型,并且研究各种影响因素对电影票房的影响。实验结果表明,多元回归模型预测效果精确度较低,BP神经网络和卷积神经网络预测的效果都比回归好,且卷积神经网络模型还有待于进一步的研究学习,根据深度学习可以挖掘更多特征变量间的关系这一特性,从而可以更有效的提高预测精度。
[Abstract]:With the rapid development of national economy, people's spiritual and cultural life has been enriched. Film is an important component of cultural life. It not only enriches people's spiritual world, but also promotes the development of Chinese film and television industry. Film is the mainstay of the cultural industry in China, which brings huge economic benefits to the development of the country, and the most important and intuitive index to measure the economic benefits of film is the box office income of the film. China's film market is growing rapidly, with total box office receipts reaching 45.712 billion yuan in 2016. Therefore, according to the characteristics of the film market in China, it is of great reference value to study the influencing factors of the film box office for the investment decision of the film in our country. At present, there are a large number of related documents on box office prediction, but most of them are analyzed on the level of traditional statistical methods and traditional neural networks. The prediction accuracy of multivariate linear regression model is lower than that of neural network, but regression can clearly explain the influence degree of influencing factors on box office, and the accuracy of neural network model in predicting movie box office is higher than that of regression model. However, the impact of various variables on the film box office is difficult to explain. Therefore, the neural network model is called the "black box". Based on the characteristics of the Chinese film market and the historical box office data, this paper proposes an innovative application of the convolution neural network in the depth learning model to the prediction of movie box office. In this paper, multiple linear regression BP neural network and convolutional neural network are used to model the sample data of film box office, and the influence of various factors on the box office is studied. The experimental results show that the prediction accuracy of multivariate regression model is lower than that of BP neural network and convolutional neural network, and the convolution neural network model needs to be further studied. According to the feature that more feature variables can be mined by depth learning, the prediction accuracy can be improved more effectively.
【学位授予单位】：首都经济贸易大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：C912.3;J943

【参考文献】