Construction and Implementation of a High School Chinese Literature Imagery Deconstruction Model Based on Multimodal Transformers

This study addresses the challenges of semantic abstraction and fragmented cultural context in the teaching of classical poetry and prose in high school. We designed and implemented a multimodal imagery deconstruction model (MT-PoemDecoder). By integrating three types of heterogeneous data: the semantics of high school Chinese texts (classical poetry and prose), visual arts, and temporal-spatial contexts (dynasties, geographical tags), the model employs a dual-stream Transformer architecture and a cross-modal attention mechanism. It deconstructs imagery across three dimensions: superficial symbolic representations (e.g., “lonely boat,” “falling leaves”), emotional metaphor mappings (e.g., the “moon” associated with homesickness), and historical allusion origins (e.g., “Golden Crow” referring to the sun), generating an interactive dynamic imagery knowledge graph (F1-score=0.93, a 22% improvement over unimodal BERT). In practical teaching scenarios, the model powers AR scenario tools, enabling poetry-painting spatial translation (e.g., the virtual-real mapping of “Jiang Xue” and Song Dynasty landscape paintings), emotional density comparisons (Li Bai's “bold imagery” at 35.2% vs. Du Fu's “sombre imagery” at 27.8%), and multi-perspective situational explorations (the dual perspectives of the traveler and the longing spouse in “Spring River Flower Moon Night”). A simulated control experiment across 12 high schools (N=1,200) indicated a 21.7% increase in the accuracy of imagery interpretation among students in the experimental group, along with a significant enhancement in cultural identity (p<0.01). This research provides an innovative paradigm for integrating artificial intelligence into traditional high school Chinese education.