基于极性转移和LSTM的树结构网络与句子分类

长短期记忆网络(long short term memory，LSTM)是一种能长久储存序列信息的循环神经网络，在语言模型、语音识别、机器翻译等领域都得到了广泛的应用。先研究了前人如何将LSTM中的记忆模块拓展到语法树得到LSTM树结构网络模型，以获取和储存句子深层次的语义结构信息；然后针对句子词语间的极性转移在LSTM树结构网络模型中添加了极性转移信息提出了极性转移LSTM树结构网络模型，更好获取情感信息来进行句子分类。实验表明在Stanford sentiment tree-bank数据集上，提出的极性转移LSTM树结构网络模型的句子分类效果优于LSTM、递归神经网络等模型。

Tree-structured networks based on polarity shifting and LSTM for sentences classification

Long short term memory (LSTM) is a recurrent neural network (RNN) which has outstanding ability to preserve sequence information for a long time. It has been widely used in the language modeling, machine translation, speech recognition and other fields. Firstly, this paper explored how predecessors extend the memory module in LSTM to the syntax tree to get Tree-Structured LSTM networks model which can obtain and store the semantic structure information of the sentences; Then, according to the polarity shifting information between the words of sentences, it proposed a Polarity Shifting Tree-Structured LSTM networks model to capture the sentiment information for better sentences classification. The proposed model show a better performance than LSTM, recursive neural networks and other models for sentences classification on Stanford Sentiment Tree-bank dataset.