知识增强图Transformer的方面级情感分析

郑文军,黎志昆,韩守飞

Aspect-based sentiment analysis via knowledge-enhanced graph Transformer

Wenjun ZHENG,Zhikun LI,Shoufei HAN

表 2 在5个数据集上与基线模型的对比结果

Tab.2 Comparison result with baseline model on five datasets

嵌入方式 模型 Restaurant14 Laptop14 Twitter Restaurant15 Restaurant16 Acc F1 Acc F1 Acc F1 Acc F1 Acc F1 GloVe SenticGCN (2021) 0.8134 0.7307 0.7602 0.7207 0.7254 0.7046 0.8063 0.6511 0.8896 0.7096 SSEGCN (2022) 0.8293 0.7539 0.7848 0.7453 0.7612 0.7479 0.8026 0.6525 0.9058 0.7579 KGAN (2023) 0.8325 0.7526 0.7828 0.7452 0.7727 0.7619 0.8076 0.6233 0.8922 0.7151 DualGCN (2024) 0.8266 0.7445 0.7674 0.7356 0.7518 0.7371 0.8026 0.6685 0.8912 0.7416 TextGT (2024) 0.8329 0.7664 0.7737 0.7342 0.7548 0.7409 0.7934 0.6161 0.8734 0.7273 DAGCN (2024) 0.8204 0.7394 0.7579 0.7185 0.7667 0.7542 0.8137 0.6578 0.8718 0.6855 K-GTNet (本文) 0.8427 0.7760 0.7816 0.7485 0.7693 0.7550 0.8147 0.6641 0.8998 0.7681 BERT SenticGCN (2021) 0.8580 0.7840 0.7774 0.7336 0.7428 0.7329 0.8413 0.6803 0.9075 0.7461 SSEGCN (2022) 0.8518 0.7742 0.8049 0.7683 0.7703 0.7599 0.8487 0.7261 0.9237 0.7654 GMF-SKIA (2023) 0.8419 0.7836 0.7869 0.7557 0.7412 0.7321 — — — — KGAN (2023) 0.8618 0.8085 0.8109 0.7777 0.8026 0.7939 0.8584 0.7294 0.9281 0.8141 TextGT (2024) 0.8561 0.7994 0.8117 0.7771 0.7740 0.7616 0.8598 0.7075 0.9156 0.7762 DAGCN (2024) 0.8195 0.7010 0.8085 0.7722 0.7341 0.7207 0.8284 0.6334 0.9237 0.7901 WordTransABSA (2024) 0.8554 0.7911 0.7840 0.7483 0.7488 0.7651 — — — — LSOIT (2024) 0.8607 0.8037 0.7899 0.7558 0.7760 0.7651 — — — — K-GTNet (本文) 0.8642 0.8115 0.8201 0.7862 0.7787 0.7674 0.8569 0.7213 0.9275 0.8116