区别于现有类别型方法，维度型文本情感分析方法可以提供更为准确细致的情感信息。目前针对中文文本的维度型研究仍面临词典和语料资源覆盖性差，现有分析方法效率及准确度低等问题。项目基于Valence-Arousal(VA)空间，研究从词汇、短语、句子和文本等多个层次进行中文维度型文本情感分析，具体包括三个方面的内容：1) 通过将情感信息引入词嵌入的学习过程，形成基于情感信息的词向量修正模型，解决现有词向量无情感信息的问题；2) 在短语层次，实现基于自动权重学习的修饰词短语情感强度预测模型，整合实词的情感强度以及修饰词的修饰权重，解决现有基于规则方法中准确率低的问题；3) 在文本层次，利用基于残差堆叠的长短期记忆模型，构建一个六层神经网络实现文本层次的VA值预测，解决深层模型的退化问题。项目希望形成一套系统性的维度型中文文本情感分析方法，构建中文VA情感词典，实现中文文本情感分析的原型系统。

Compared with existing categorical approach, the dimensional approach can provide more fine-grained sentiment analysis. However, there are still several difficulties in current relevant research works for Chinese text, such as poor coverage of affective lexicons and corpora resource, and low performance and accuracy of existing analysis methods. Based on continuous valence-arousal (VA) space, this project proposes to research on dimensional sentiment analysis for Chinese text on word-, phrase-, sentence- and text-level. Main content include:. (1) By introducing affective information into the training procedure, a word vector refinement model will be proposed. This model is expected to tackle the problem of lacking affective information in current word vectors.. (2) On phrase-level, an automatic weight learning model will be proposed to predict sentiment intensity for modifier phrase. By combining the intensity of the content word and the shifting weight of the modifier within the phrase, the model can address the low accuracy problem in existing rule-based method.. (3) On text-level, a six-layer neural network model will be proposed to predict VA values for given texts. Based on stacked residual LSTM units, this model can solve the degradation and low performance problems in existing deep models for dimensional sentiment analysis.. Overall, this project seeks to provide a systematic framework of dimensional sentiment analysis, build a Chinese VA lexicon, and implement a prototype system of sentiment analysis for Chinese Text.