融合N-Gram的水產(chǎn)養(yǎng)殖長(zhǎng)文本實(shí)體關(guān)系聯(lián)合抽取

打開(kāi)文本圖片集

關(guān)鍵詞：水產(chǎn)養(yǎng)殖；長(zhǎng)文本；實(shí)體關(guān)系聯(lián)合抽取;N-Gram算法；多模型融合算法DOI：10. 15938/j. jhust. 2025. 02. 010中圖分類號(hào)：TP391.1;S951.2 文獻(xiàn)標(biāo)志碼：A 文章編號(hào)：1007-2683（2025）02-0091-13

Abstract：Tosolvetheproblemofmisjudgmentandlossofvalidinformationcausedbyalargeamountof irelevantinformationin aquaculturelongtext，ajointextractionmethodofentityrelationsbasedonN-Gramfusionwasproposed.Firstly，themulti-model fusionalgorithmisused toextract thetextmatrixfeaturemapbasedonBERTiitialization，andthenthecascadingBiLSTMisusedto extractedepfeatures.Afterthat，thefeaturesofthelongtextslcematrixpreprocessedbyfusionN-Gramalgorithmareeracted layerbylayer，andtherelativeandabsolutepositionsofslicematrixaremodeled.Theexperimentalresultsontheself-constructed aquaculturelong textdatasetandSKE publicdataset show significant improvementscompared withthe benchmark model.The experimentalresultsshowthatthismethodcanfullacquireandprocessthesemanticinformation inaquaculturelongtext，and effectively improve the accuracy and integrity of entity relation extraction.

Keywords：aquaculture；long text;joint extractionof entityrelations；N-Gram algorithm；multi-model fusion algorithm

0 引言

隨著我國(guó)經(jīng)濟(jì)的高速發(fā)展，各行各業(yè)都在向智慧化方向轉(zhuǎn)型[1-3]，水產(chǎn)養(yǎng)殖業(yè)也在向精準(zhǔn)化和集約化的養(yǎng)殖方式轉(zhuǎn)變。（剩余18864字）