Enhancing the interpretability of text classification models can help increase the reliability of these models in real-world applications. Currently, most researchers focus on extracting task-specific words from inputs to improve the interpretability of the model. The competitive approaches exploit the Variational Information Bottleneck (VIB) to improve the performance of word masking at the word embedding layer to obtain task-specific words. However, these approaches ignore the multi-level semantics of the text, which can impair the interpretability of the model, and do not consider the risk of representation overlap caused by the VIB, which can impair the classification performance. In this paper, we propose an enhanced variational word masks approach, named E-VarM, to solve these two issues effectively. The E-VarM combines multi-level semantics from all hidden layers of the model to mask out task-irrelevant words and uses contrastive learning to readjust the distances between representations. Empirical studies on ten benchmark text classification datasets demonstrate that our approach outperforms the SOTA methods in simultaneously improving the interpretability and accuracy of the model.

Current works about false information detection based on conversation graphs on social networks focus primarily on two research streams from the standpoint of topic distribution: in-topic and cross-topic techniques, which assume that the data topic distribution is identical or cross, respectively. This signifies that all test data topics are seen or unseen by the model. However, these assumptions are too harsh for actual social networks that contain both seen and unseen topics simultaneously, hence restricting their practical application. In light of this, this paper develops a novel open-topic scenario that is better suited to actual social networks. In this open-topic scenario, we empirically find that the existing models suffer from impairment in the detection performance for seen or unseen topic data, resulting in poor overall model performance. To address this issue, we propose a novel Contrastive Adversarial Learning Network, CALN, that employs an unsupervised topic clustering method to capture topic-specific features to enhance the model’s performance for seen topics and an unsupervised adversarial learning method to align data representation distributions to enhance the model’s generalisation to unseen topics. Experiments on two benchmark datasets and a variety of graph neural networks demonstrate the effectiveness of our approach.

Extracting keyphrases that summarize the main points of a document is a fundamental task in natural language processing. Supervised approaches to keyphrase extraction(KPE) are largely developed based on the assumption that the training data is fully annotated. However, due to the difficulty of keyphrase annotating, KPE models severely suffer from incomplete annotated problem in many scenarios. To this end, we propose a more robust training method that learns to mitigate the misguidance brought by unlabeled keyphrases. We introduce negative sampling to adjust training loss, and conduct experiments under different scenarios. Empirical studies on synthetic datasets and open domain dataset show that our model is robust to incomplete annotated problem and surpasses prior baselines. Extensive experiments on five scientific domain datasets of different scales demonstrate that our model is competitive with the state-of-the-art method.