机器学习在影视大数据分析中的研究及应用

发布时间：2019-03-26 21:27

【摘要】：影视产业作为我国国民经济体系中新的突破口,广受影视市场主导人员、电台运营商、各大视频网站运营机构以及一些科研学者的关注。面对大数据时代的到来,影视行业的数据存储、处理和分析等多个方面也面临着巨大的挑战,传统的数据存储模式、数据处理方法和数据分析技术将无法满足拥有海量数据的应用需求。随着数理统计理论及人工智能等诸多领域的不断发展,基于机器学习的理论体系逐渐构建起来,人们试图应用机器学习方法去对海量数据进行处理分析,以期从中提取出有用的知识和信息。因此,研究如何运用机器学习方法从海量影视大数据中挖掘出数据背后隐藏的特征和波动趋势,是具有重大的现实指导意义的。本文主要是利用机器学习方法来对影视大数据进行处理与分析,同时结合智能影视大数据分析系统对海量的电视剧收视相关数据先后进行预处理、特征降维、图表分析与收视预测,其增加了数据处理的效率和收视预测的准确性。因此,通过机器学习方法来解决影视大数据场景中的问题具有重要的意义,其给予了研究人员有效的应用思路,也为影视企业赢取最终市场并获得更高收视率创造了可能。本文的主要工作如下:[1]基于K-Means聚类算法对高维影视数据进行预处理。其针对筛选出的电视剧样本数据进行属性选择、数据聚集和数据规范化,最后利用K-Means算法对数据进行补全操作。[2]基于因子分析法对高维影视数据进行降维处理。其针对高冗余、高维度的电视剧特征数据,运用因子分析法来获得低维的冗余性小的影响因子作为降维后的特征向量。[3]基于SVM算法和AdaBoost-BP算法对电视剧收视水平和收视率进行分类与预测。其使用降维后的电视剧特征数据,运用SVM算法和AdaBoost-BP算法来建立收视预测模型。之后对相关数据进行预测分析。最后对比分析预测效果,总结出更具有效性的预测算法。[4]基于智能影视大数据分析系统对收视进行分析与展示。其针对处理后的电视剧收视相关数据,多层次多角度地进行图表关联分析与直观展示,并把文中提出的预测模型运用到影视大数据收视预测中,验证了其有效性。
[Abstract]:As a new breakthrough in China's national economic system, the film and television industry is widely concerned by the leading personnel in the film and television market, radio operators, major video website operators and some scientific researchers. In the face of the arrival of big data's era, the film and television industry's data storage, processing and analysis are also facing enormous challenges, traditional data storage mode, Data processing methods and data analysis techniques will not meet the needs of applications with huge amounts of data. With the development of mathematical statistics theory and artificial intelligence and many other fields, the theoretical system based on machine learning is gradually constructed, and people try to use machine learning method to process and analyze massive data. In order to extract useful knowledge and information from it. Therefore, it is of great practical significance to study how to use the machine learning method to dig out the hidden features and fluctuating trends behind the data from the massive film and television big data. This article mainly uses the machine learning method to process and analyze the film and television big data, at the same time combines the intelligent film and television big data analysis system to pre-process the massive TV series ratings related data successively, and reduces the feature dimension. Chart analysis and ratings prediction increase the efficiency of data processing and the accuracy of ratings prediction. Therefore, it is of great significance to solve the problems in the film and television big data scene by means of machine learning, which gives researchers effective application ideas and creates the possibility for film and television enterprises to win the final market and obtain higher ratings. The main work of this paper is as follows: [1] pre-processing high-dimensional video data based on K-Means clustering algorithm. According to the selected TV series sample data for attribute selection, data aggregation and data normalization, finally using the K-Means algorithm to complete the data. [2] based on factor analysis for high-dimensional film and television data dimensionality reduction. For highly redundant, high-dimensional TV feature data, The factor analysis method is used to obtain the lower dimension redundancy factor as the feature vector after dimension reduction. [3] based on SVM algorithm and AdaBoost-BP algorithm, the ratings and ratings of TV series are classified and predicted. It uses the reduced dimension TV series feature data, uses the SVM algorithm and the AdaBoost-BP algorithm to establish the ratings prediction model. Then the related data are predicted and analyzed. Finally, the prediction results are compared and the more effective prediction algorithms are summarized. [4] based on the intelligent film and television big data analysis system, the analysis and display of the ratings are carried out. According to the related data of TV series after processing, multi-level and multi-angle graph correlation analysis and visual display are carried out, and the prediction model proposed in this paper is applied to the movie and television big data ratings prediction to verify its effectiveness.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP181;TP311.13

【参考文献】