基于条件随机场的自然口语语义理解方法

Approach to Understand Chinese Oral Task for Mobile Terminals Based on Conditional Random Fields

采用条件随机场技术将面向智能手机用户的自然口语语义理解分为操作任务分类和语义组块提取两个主要步骤,收集并分析了口语语料库的特征,根据归纳出的任务种类和语义组块特征规律设计了任务分类标记集和语义组块标记集;通过基于规则的组块分析得到了中间语义表示格式,从而实现了对用户口语语义理解的目的.实验结果表明:任务分类准确率及语义组块提取平均正确率分别达到98.85%和94.53%,系统综合性能测试的准确率达到91.86%.

Conditional random fields technology is applied to this paper and semantic understanding of the natural spoken language of smart phone users is divided into two processes： classification of operating instructions and extraction of semantic chunking information.Characteristics of the spoken language corpus are collected and analyzed.Tag sets of task classification and semantic chunking are designed according to the inductive types of tasks and the rule of the semantic chunking characteristics.The middle semantic representation format are obtained through the chunking analysis based on rules, so as to realize the goal of the semantic understanding spoken language of the users.In the experiment, the accuracy rate of tasks classification reaches 98.85%, and the average accuracy rate of semantic chunking extract reaches 94.53%.In the end, the accuracy rate of system＆#39;＆#39;s comprehensive performance test reaches 91.86%.