Sememe is the smallest semantic unit for describing real-world concepts, which improves the interpretability and performance of Natural Language Processing (NLP). To maintain the accuracy of the sememe description, its knowledge base needs to be continuously updated, which is time-consuming and labor-intensive. Sememes predictions can assign sememes to unlabeled words and are valuable work for automatically building and/or updating sememeknowledge bases (KBs). Existing methods are overdependent on the quality of the word embedding vectors, it remains a challenge for accurate sememe prediction. To address this problem, this study proposes a novel model to improve the performance of sememe prediction by introducing synonyms. The model scores candidate sememes from synonyms by combining distances of words in embedding vector space and derives an attention-based strategy to dynamically balance two kinds of knowledge from synonymous word set and word embedding vector. A series of experiments are performed, and the results show that the proposed model has made a significant improvement in the sememe prediction accuracy. The model provides a methodological reference for commonsense KB updating and embedding of commonsense knowledge.

Legal judgment prediction is a common basic task in the field of Legal AI, aimed at using deep domain models to predict the outcomes of judicial cases, such as charges, legal provisions, and other related tasks. This task has practical applications in environmental law, including legal decision assistance and legal advice, offering a promising and broad prospect. However, most previous studies focus on using high-quality labeled data for strong supervised training in criminal justice, often neglecting the rich external knowledge contained in various charges and laws. This approach fails to accurately simulate the decision-making steps of judges in real scenarios, overlooking the semantic information in case descriptions that significantly impacts judgment results, leading to biased outcomes. In judicial environmental protection, the high overlap and similarity between different charges can cause confusion, and there is a lack of relevant judicial decision labeling datasets. To address this, we propose the External Knowledge-Infused Cross Attention Network (EKICAN), which leverages the robust semantic understanding capabilities of large models. By extracting information such as fact descriptions and court opinions from documents of criminal, civil, and administrative cases related to judicial environmental protection, we construct the Judicial Environmental Law Judgment Dataset (JELJD). We address data imbalance in this dataset using the text generation capabilities of judicial large models. Finally, EKICAN fuses semantic information from different parts with external knowledge to output prediction results. Experimental results show that EKICAN achieves state-of-the-art performance on the JELJD compared to advanced models.