面向在线讨论的时间序列建模实验*

摘要

The automatic evaluation and monitor of posts in online discussions are extremely challenging research topics. This paper proposed a calculation method for information quantity of posts in online discussions from the perspective of information theory. Meanwhile, the time series modeling experiment for online discussions was carried out by adopting the auto-regressive integrated moving average (ARIMA) model, which obtained the optimized ARIMA(5, 2, 5) model. The experimental results showed that the ARIMA (5, 2, 5) model could accurately describe the future change tendency of information quantity of posts in online discussions and give the fluctuation range. In addition, the prediction accuracy of this model was superior to that of benchmark models. This paper explored the time series modeling method for online discussions, expecting to reduce the time cost in monitoring the quality of online discussions and provide a new perspective for the study of quality evaluation methods for online discussions.%在线讨论中帖子的自动评价与监测是一个极具挑战的研究主题.文章从信息论的视角,提出了在线讨论中帖子的信息量计算方法,并采用自回归单整移动平均(Auto-Regressive Integrated Moving Average,ARIMA)模型,开展了面向在线讨论的时间序列建模实验,得到最优化的ARIMA(5, 2, 5)模型.实验结果表明,ARIMA(5, 2, 5)模型能够正确描述在线讨论中帖子信息量的未来变化趋势并给出波动范围,且其预测的准确性优于基准模型.文章探索了面向在线讨论的时间序列建模方法,以期降低监控在线讨论质量的时间成本,并为研究在线讨论质量评价方法提供新的视角.