基于多特征的中文文本蕴含识别

摘要 提出了一种基于多特征的中文文本蕴含识别方法,首先对文本进行预处理、中文分词、词性标注、命名实体识别、依存分析等处理;然后提取字符串特征、句法特征、语义特征等,使用贝叶斯逻辑回归模型进行预测;最后再使用规则进行修正,得到最终的识别结果. 该方法在2014年RITE-VAL评测任务的CS数据上的MacroF1为0.625,超过目前最好的研究现状(MacroF1:0.615, BUPTTeam-CS-SVBC-05).
Recognizing textual entailment is an effective approach for computer to automatically identify semantic relation between texts with an important position in the field of natural language processing. A method using multi-feature was proposed. The new algorithm preprocess the raw text, Chinese characters segmentation, part-of-speech tagging, named entity recognition and dependency parser, string features, syntactic features and semantic features and uses Bayesian logistic regression model to predict the preliminary results, finally it uses the rules to filter the results. Experiments indicate that the algorithm's MacroF1 on RITE-VAL data is 0.625, outperform optimal value (MacroF1:0.615, BUPTTeam-CS-SVBC-05).