【摘要】：針對情感分類問題中長句和短句具有不同的建模特點,提出了一種基于聯(lián)合深度學(xué)習(xí)模型的情感分類方法。該方法融合長短期記憶(long-short term memory,LSTM)模型與卷積神經(jīng)網(wǎng)絡(luò)(convolutional neural network,CNN)對影視評論數(shù)據(jù)進行情感極性判別;采用LSTM對上下文進行建模,通過逐詞迭代得到上下文的特征向量;采用CNN模型從詞向量序列中自動發(fā)現(xiàn)特征,抽取局部特征并整合成全局特征來提高分類效果。所提出的方法在COAE2016評測的任務(wù)2的情感極性分類任務(wù)中,取得最高的系統(tǒng)準(zhǔn)確率。
[Abstract]:Aiming at the different modeling characteristics of long sentence and short sentence in affective classification, a method of emotion classification based on joint deep learning model is proposed. This method combines long-short term memory LSTM model with convolutional neural network (convolutional neural network to judge the emotional polarity of video review data, uses LSTM to model the context, and obtains the feature vector of the context by word iteration. The CNN model is used to automatically discover features from word vector sequences, extract local features and integrate them into global features to improve the classification effect. The proposed method achieves the highest system accuracy in the classification of emotional polarity of task 2 evaluated by COAE2016.
【作者單位】： 哈爾濱工業(yè)大學(xué)計算機科學(xué)與技術(shù)學(xué)院機器智能與翻譯實驗室;
【基金】：國家自然科學(xué)基金資助項目(61402134)
【分類號】：TP391.1

【相似文獻】

相關(guān)碩士學(xué)位論文 前1條

1 陳浩;基于中層語義特征表達的物體檢測方法研究[D];北京工業(yè)大學(xué);2016年

，

本文編號：2137326