Yixuan Liu · ITM Web of Conferences 2026 · 2026
DOI: 10.1051/itmconf/20269301023
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
In this paper, for the task of low resource text classification in the medical, industrial and other Chinese vertical fields, aiming at the problems of the existing research experimental standards are different and the lack of suitable Chinese feature theory, 9 authoritative Chinese and English literatures from 2019 to 2026 are selected to carry out a detailed domain review based on transformer pre training model. This paper studies the establishment of standardized literature screening rules, builds a hierarchical evaluation framework based on Bert base Chinese, analyzes four mainstream optimization schemes from the principle level, including prompt learning, domain incremental pre training, graph fusion model and comparative learning fine-tuning, and compares the advantages and limitations of various methods. Combined with the theoretical analysis of the validation of conventional and extremely small sample public data sets, the constraints of the number of labels on the applicable boundary of the method are clarified, and the core challenges such as the limited generalization ability of templates, the scarcity of industry corpus, the high computational cost of graph models, and the lack of field evaluation standards are summarized.
No comments yet — start the discussion below.