基于协同训练的意图分类优化方法

目的/意义]针对单纯使用统计自然语言处理技术对社交网络上产生的短文本数据进行意向分类时存在的特征稀疏、语义模糊和标记数据不足等问题,提出了一种融合心理语言学信息的Co-training意图分类方法。方法/过程]首先,为丰富语义信息,在提取文本特征的同时融合带有情感倾向的心理语言学线索对特征维度进行扩展。其次,针对标记数据有限的问题,在模型训练阶段使用半监督集成法对两种机器学习分类方法(基于事件内容表达分类器与情感事件表达分类器)进行协同训练(Co-training)。最后,采用置信度乘积的投票制进行分类。结论/结果]实验结果表明融入心理语言学信息的语料再经过协同训练的分类效果更优。

Intention Classification Optimization Method Based on Collaborative Training

Purpose/Significance]Aiming at the problems of feature sparseness, semantic ambiguity and mark data insufficiency caused by using single statistical natural language processing technology for intention classification of short text data generated on social networks, a psycholinguistic information based Co-training intention classification method was proposed.Method/Process]Firstly, in order to enrich the semantic information, the feature dimension was extended by extracting the features of the text while synthesizing the psycholinguistic clues with emotional tendencies.Secondly, aiming at the insufficiency of mark data, two machine learning classification methods(based on the event content expression classifier and the emotional event expression classifier)were used cooperatively for training the model. Finally, the classification was performed by using a voting system of confidence products.Conclusion/Results]The experimental results show that, by adding psycholinguistic information into the corpus, the cooperative training could provide better classification results.