多特征融合下视频网站弹幕信息有用性检测研究  

Research on Usefulness Detection of Danmaku Information in Video Websites Based on Multi-Feature Fusion

作　　者：张瑞 何禄鑫 黄炜 Zhang Rui;He Luxin;Huang Wei(School of Economics and Management,Hubei University of Technology,Wuhan 430064,China)

机构地区：[1]湖北工业大学经济与管理学院,湖北武汉430064

出　　处：《现代情报》2022年第4期99-109,共11页Journal of Modern Information

基　　金：湖北省高等学校哲学社会科学研究重大项目“新时代高校突发事件网络舆情分析与引导机制研究”(项目编号:19ZD025);湖北省教育厅科学技术研究计划重点项目“大规模数据环境下基于时序模式挖掘的网络恐怖事件感知方法研究”(项目编号:D20191401)。

摘　　要：[目的/意义]随着视频平台弹幕功能被大众所熟知,弹幕信息呈现爆炸式增长,信息有用性检测逐渐凸显重要的学术和商业价值。[方法/过程]本文提出了多特征融合下视频网站弹幕信息有用性检测模型。该模型首先从弹幕信息效用、弹幕表达形式和弹幕用户特征3个角度提取与弹幕信息有用性相关的特征指标,然后利用随机森林对重要特征进行选择,建立机器学习模型逻辑回归、SVM、决策树、朴素贝叶斯、GBDT等对弹幕信息进行分类,得到各等级信息有用性的检测结果。[结果/结论]从结果中显示,集成模型(GBDT、LightGBM和XGBoost)相比于单模型算法展现了更好的优越性。最后,根据研究结果提出相应的理论和实践意义。本研究扩展了在线信息特征相关研究,也为评估和改善视频平台环境提供了决策依据。[Purpose/Significance]As the function of the video platform danmaku is well-known to the public,the information of danmaku has exploded,and information usefulness detection gradually highlights the important academic and commercial value.[Method/Process]In this paper,a multi feature fusion based video website danmaku information usefulness detection model was proposed.Firstly,the feature indexes related to the usefulness of danmaku information were extracted by the model from the three perspectives of danmaku information utility,danmaku expression and danmaku user characteristics.Then the important features were selected using random forest,and a machine learning model was established to classify danmaku information,such as logical regression,SVM,decision tree,naive bayes and GBDT,getting the test results of each level of information usefulness.[Results/Conclusion]The results show that the integrated models(GBDT,LightGBM and XGBoost)have better advantages than the single-model algorithm.Finally,according to the research results,the corresponding theoretical and practical significance is put forward.This research expands the research on the characteristics of online information,and also provides a decision-making basis for evaluating and improving the video platform environment.

关 键 词：特征融合 随机森林 机器学习 有用性检测 弹幕信息 

分 类 号：G203[文化科学—传播学]