【目的】在總結當前引文元數據抽取方法的基礎上,結合語義學知識和機器學習方法,對引文元數據的自動抽取方法進行探索。【方法】實驗中采用神經網絡模型對人工分割過的語料進行詞向量訓練。利用相同類型的元數據會相對集中地出現在向量空間中某一位置的現象,通過支持向量機分類算法實現對元數據的自動歸類和標注。【結果】在以外文引文數據作為測試集的實驗中,本文方法取得了較高的準確率和召回率,特別是針對引文中含有多種語言和縮寫的現象,具有較好的處理能力。【局限】在對于引文元數據時間內容的細粒度抽取中存在一定的局限性。【結論】實驗結果表明,此方法在引文元數據的自動發現和標注上具有良好的效果,并能很大程度地提高方法的適用性和容錯率。

[Objective]This paper proposes a new method to automatically extract bibliographic metadata, with the help of semantic knowledge and machine learning technologies.[Methods]We used the neural network model to create word vectors from manually split data, and then found that same type of metadata is relatively concentrated at certain locations in the vector space. Thus, we proposed a new SVM classification algorithm to classify and annotate the bibliographic metadata automatically.[Results]The proposed method achieved high recall and precision rates with citation data,especially for citations with various languages and abbreviations.[Limitations]The fine-grained extraction of the time related content could be improved.[Conclusions]The proposed method could effectively detect and tag bibliographic metadata, and improve the system's compatibility and fault tolerance ability.