如图所示，我们展示了四个数据集上不同方法及变体预测的全局及部分效果。仔细观察发现：综合考虑时序和文本...

可以改成下面这种更连贯、论文表达更自然的版本：

:::writing{variant="document" id="48291"} 如图所示，我们展示了四个数据集中不同方法及其变体在全局范围和局部区间上的预测效果。通过细致观察可以发现，在综合建模时序动态特征与文本上下文语义信息后，iAdapter 在若干波动较为显著的时间区间内能够更准确地捕捉序列变化趋势，表现出更强的时序理解与预测能力。 :::

对应的 AAAI 风格英文表述如下：

:::writing{variant="document" id="73926"} As illustrated in the figure, we present the global and local prediction results of different methods and their variants across four datasets. A closer inspection shows that, by jointly considering temporal dynamics and contextual semantics from textual information, iAdapter achieves a better understanding of the underlying time-series patterns. In particular, it captures the trend variations more accurately in several intervals with pronounced fluctuations, demonstrating stronger temporal reasoning and forecasting capability. :::

如果想更强调“对比优势”，可以把最后一句进一步改成：This highlights the effectiveness of iAdapter in leveraging cross-modal contextual information for robust time-series forecasting.