Mouse interaction data contain a lot of interaction information between users and Search Engine Result Pages (SERPs), which can be useful for evaluating search satisfaction. Existing studies use aggregated features or anchor elements to capture the spatial information in mouse interaction data, which might lose valuable mouse cursor movement patterns for estimating search satisfaction. In this paper, we leverage regions together with actions to extract sequences from mouse interaction data. Using regions to capture the spatial information in mouse interaction data would reserve more details of the interaction processes between users and SERPs. To modeling mouse interaction sequences for search satisfaction evaluation, we propose a novel LSTM unit called Region-Action LSTM (RALSTM), which could capture the interactive relations between regions and actions without subjecting the network to higher training complexity. Simultaneously, we propose a data augmentation strategy Multi-Factor Perturbation (MFP) to increase the pattern variations on mouse interaction sequences. We evaluate the proposed approach on open datasets. The experimental results show that the proposed approach achieves significant performance improvement compared with the state-of-the-art search satisfaction evaluation approach.

鼠标交互数据包含用户与搜索引擎结果页面（SERP）之间的大量交互信息，这对于评估搜索满意度很有用。现有研究使用聚合的特征或锚点元素来捕获鼠标交互数据中的空间信息，这可能会丢失宝贵的鼠标光标移动模式以用于估计搜索满意度。在本文中，我们利用区域以及从鼠标交互数据中提取序列的动作。使用区域捕获鼠标交互数据中的空间信息将保留用户和SERP之间的交互过程的更多细节。为了对鼠标互动序列进行建模以进行搜索满意度评估，我们提出了一种新型的LSTM单元，称为Region-Action LSTM（RALSTM），它可以捕获区域和动作之间的交互关系，而不会使网络受到更高的训练复杂性。同时，我们提出了一种数据增强策略多因素摄动（MFP），以增加鼠标交互序列上的模式变化。我们对开放数据集评估提出的方法。实验结果表明，与最新的搜索满意度评估方法相比，该方法可显着提高性能。