基于雙重對齊領域自適應的文本分類模型
首發時間:2021-08-24
摘要:為了有效的進行不同領域之間的遷移學習,考慮了領域間的特征分布差異和標簽分布差異,提出基于雙重對齊領域自適應的文本分類模型。首先,模型使用相關對其算法(CORAL)對源域與目標域所提出的特征進行特征對齊,將對齊后的特征輸入到特征提取器中進一步提取特征;然后輸入到分類器和判別器中進行兩個領域的特征在正負類別和域類別上達到雙重對齊。該方法在亞馬遜數據集上進行實驗,實驗結果證明了模型對的有效性。
For information in English, please click here
Double Alignment Domain Adaptation Based Text Classification Model
Abstract:In order to perform transfer learning between different domains effectively, we consider the differences in feature distribution and label distribution between both domains, and propose a double alignment domain adaptation text classification model. First, the correlation alignment (CORAL) loss is utilized to align the feature distributions between the source and target domain. The aligned features are input to the feature extractor to further extract features, and then are input to classifier and discriminator, forcing the features of the source and target domains to have clear positive and negative category boundaries and domain category boundaries. Through training, the features of the two domains are double aligned on the positive and negative categories and the domain categories. This model was tested on Amazon review dataset. Experimental results demonstrate the effectiveness of our model.
Keywords: text classification domain adaptation correlation alignment distribution discrepancy
基金:
引用
No.****
動態公開評議
共計0人參與
勘誤表
基于雙重對齊領域自適應的文本分類模型
評論
全部評論