STVW-MVC算法在电竞比赛分析中的应用  

Application of STVW-MVC Algorithm in the Analysis of E-sports

作　　者：王欣 徐蕾艳 吴菲 WANG Xin;XU Lei-yan;WU Fei(School of Digital Business,Nanjing Vocational College of Information Technology,Nanjing 210023,China;School of Mathematics and Infomation Science,Nanjing Normal University of Special Education,Nanjing 210038,China)

机构地区：[1]南京信息职业技术学院数字商务学院,南京210023 [2]南京特殊教育师范学院数学与信息科学学院,南京210038

出　　处：《印刷与数字媒体技术研究》2025年第1期169-178,196,共11页Printing and Digital Media Technology Study

基　　金：国家自然科学基金(No.12101319);江苏省产学研合作项目(No.BY2022940);南京信息职业技术学院高层次人才科研启动项目(No.YB20220602)。

摘　　要：近年来,电子竞技类型的比赛逐渐出现在大众的视野中,电竞比赛中对于各战队和选手战术和隐藏模式的分析是重要的一环。以往对于比赛的分析大多通过专家人工分析,存在主观因素强,难以分析战队和选手的隐藏模式等问题。本研究提出了一种时序感知的视图权重融合多视图聚类算法STVW-MVC,增加了视图重要性权重,改进了原算法中对于多视图信息一致性的处理,针对比赛中的各数据进行更准确的多视图聚类,以挖掘参赛选手或战队的战术和隐藏模式。相比原有的MVC-MAE算法,该算法更注重时序信息,使得时间序列靠后的信息会与时间序列靠前的信息产生信息一致性,而时间序列靠前的信息无法与其之后的信息产生一致性,能够更好地挖掘出潜在特征,从而进行更有效的聚类。除此以外,本研究还引入了视图重要性权重和视图相关性阈值,提高了聚类效果。实验表明,本研究算法的聚类结果更加紧凑合理,为电子竞技比赛数据的分析提供了一种更高效、更准确的方法。In recent years,e-sports competitions have gradually appeared in the public visual field.The analysis of tactics and hidden patterns of teams and players in e-sports competitions is an important aspect.Previous analysis of competitions mostly relied on manual analysis by experts,which had strong subjective factors and difficulties in analyzing hidden patterns of teams and players.In this study,a Sequential Temporal View Weighted Multi-View Clustering(STVW-MVC)algorithm was proposed that was aware of the temporal aspect and integrates view importance weights.The handling of multi-view information consistency in the original algorithm was improved and enabled more accurate multi-view clustering of various data in competitions to uncover tactical and hidden patterns of participants or teams.Compared with the original MVC-MAE algorithm,this algorithm paid more attention to time series information,so that the information at the end of the time series would be consistent with the information at the beginning of the time series,while the information at the beginning of the time series could not be consistent with the information after it,which could better mine potential features and thus performed more effective clustering.In addition,this study also introduced view importance weights and view relevance thresholds to improve the clustering effect.Experiments showed that the clustering results of this study’s algorithm are more compact and reasonable,providing a more efficient and accurate method for the analysis of e-sports competition data.

关 键 词：深度自编码器聚类 MVC-MAE算法 多视图 信息一致性 时间序列 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]