Focusing on each country’s topmost destination/origin migration relation with other countries, this study builds top1 destination networks and top1 origin networks in order to understand their skeletal construction and community dynamics. Each top1 network covers approximately 50% of the complete migrant network stock for each decade between 1960 and 2000. We investigate the community structure by implementing the Girvan-Newman algorithm and compare the number of components and communities to illustrate their differences. We find that (i) both top1 networks (origin and destination) exhibited communities with a clear structure and a surprising evolution, although 80% edges persist between each decade; (ii) top1 destination networks focused on developed countries exhibiting shorter paths and preferring more advance countries, while top1 origin networks focused both on developed as well as more substantial developing nations that presented a longer path and more stable groups; (iii) only few countries have a decisive influence on community evolution of both top1 networks. USA took the leading position as a destination country in top1 destination networks, while China and India were the main Asian emigration countries in top1 origin networks; European countries and the Russian Federation played an important role in both.
Natural language understanding (NLU) has two core tasks: intent classification and slot filling. The success of pre-training language models resulted in a significant breakthrough in the two tasks. One of the promising solutions called BERT can jointly optimize the two tasks. We note that BERT-based models convert each complex token into multiple sub-tokens by wordpiece algorithm, which generates a mismatch between the lengths of the tokens and the labels. This leads to BERT-based models do not do well in label prediction which limits model performance improvement. Many existing models can be compatible with this issue but some hidden semantic information is discarded in the fine-tuning process. We address the problem by introducing a novel joint method on top of BERT which explicitly models the multiple sub-tokens features after wordpiece tokenization, thereby contributing to the two tasks. Our method can well extract the contextual features from complex tokens by the proposed sub-words attention adapter (SAA), which preserves overall utterance information. Additionally, we propose an intent attention adapter (IAA) to obtain the full sentence features to aid users to predict intent. Experimental results confirm that our proposed model is significantly improved on two public benchmark datasets. In particular, the slot filling F1 score is improved from 96.1 to 98.2 (2.1% absolute) on the Airline Travel Information Systems (ATIS) dataset.
23 March 2026 in Theoretical and Applied Climatology