CGIR: a Model of Cross Language Information Retrieval based on Concept Graph by Fusing Attention Mechanism
Gang Liu, Jiang Wen-hua, Yulin Hu, Kai Zhan, Tongli Wang · 2023
Cross language information retrieval faces challenges such as language differences, data scarcity, contextual disparities, and machine translation errors. To enhance retrieval accuracy and effectiveness, this paper proposes a similarity evaluation framework called Concept Graph Information Retrieval (CGIR). The framework includes the creation of concept graphs, quantized representations of these graphs, and retrieval processes. The construction process of CGIR incorporates an attention mechanism, significantly boosting its performance and accuracy. Through this fusion, concept graphs offer a comprehensive representation of texts, capturing the essence of the entire content while minimizing the displayed information, all while preserving the original meaning of the text to the fullest extent possible. The experimental results clearly demonstrate that the generated concept graphs effectively function as semantic representations of the entire texts. In comparison to keyword-based, ontology-based, and term-based retrieval methods, CGIR exhibits a remarkable improvement in accuracy, surpassing them by over 10%.