Multivariate time series is ubiquitous in real-world applications, yet it often suffers from missing values that impede downstream analytical tasks. In this paper, we introduce the Long Short-Term Memory Network based Recurrent Autoencoder with Imputation Units and Temporal Attention Imputation Model (RATAI), tailored for multivariate time series. RATAI is designed to address certain limitations of traditional RNN-based imputation methods, which often focus on predictive modeling to estimate missing values, sometimes neglecting the contextual impact of observed data at and beyond the target time step. Drawing inspiration from Kalman smoothing, which effectively integrates past and future information to refine state estimations, RATAI aims to extract feature representations from time series data and use them to reconstruct a complete time series, thus overcoming the shortcomings of existing approaches. It employs a dual-stage imputation process: the encoder utilizes temporal information and attribute correlations to predict and impute missing values, and extract feature representation of imputed time series. Subsequently, the decoder reconstructs the series from the feature representation, and the reconstructed values are used as the final imputation values. Additionally, RATAI incorporates a temporal attention mechanism, allowing the decoder to focus on highly relevant inputs during reconstruction. This model can be trained directly using data that contains missing values, avoiding the misleading effects on model training that can arise from setting initial values for missing values. Our experiments demonstrate that RATAI outperforms benchmark models in multivariate time series imputation.

Named entity recognition, a fundamental task in natural language processing, faces challenges related to the sequence labeling framework widely used when dealing with nested entities. The span-based method transforms nested named entity recognition into span classification tasks, which makes it an efficient way to deal with overlapping entities. However, too much overlap among spans may confuse the model, leading to inaccurate classification performance. Moreover, the entity mentioned in the training dataset contains rich information about entities, which are not fully utilized. So, in this paper, a span-prototype graph is constructed to improve span representation and increase its distinction. In detail, we utilize the entity mentions in the training dataset to create a prototype for each entity category and add prototype loss to adapt the span to its similar prototype. Then, we feed prototypes and span into a graph attention network (GAT), enabling span to automatically learn from different prototypes, which integrate the information about entities into the span representation. Experiments on three common nested named entity recognition datasets, including ACE2004, ACE2005, and GENIA, show that the proposed method achieves 87.28%, 85.97%, and 79.74% F1 scores on ACE2004, ACE2005, and GENIA, respectively, performing better than baselines.