基于时序知识图谱的智能感知人机交互系统研究

Intelligent sensing human-computer interaction system based on temporal knowledge graph

社会信息化程度提高的背景下,社会对人机交互技术需求同样增加。为提高人机交互系统的效率和用户体验,并扩大人机交互系统的应用场景,在时序知识图谱的基础上引入知识表示学习Trans系列模型对其中的关键技术进行优化。经实验分析可知,GraphQ数据集中研究算法的拟合度最高,达到98.2%,其中无监督学习方法出现欠拟合的情况。研究方法下词义理解、句法分析、语境理解的准确度均在95%以上。线下场馆智能问答应用中,研究系统的平均精确度达到97.6%,比基于循环神经网络与基于无监督学习的系统的精确度分别高25.3%和31.6%。研究系统的机器人在执行多目标对话请求下的最佳规划路径长度为153.4 m。综上可知,研究的人机交互对话系统具有高拟合度和高准确性。

Under the background of the improvement of social informatization,the demand for human-computer interaction technology is also increasing.In order to improve the efficiency and user experience of human-computer interaction system and expand the application scenarios of human-computer interaction system,knowledge representation learning Trans series model is introduced to optimize the key technologies on the basis of temporal knowledge graph.Through experimental analysis,it can be seen that the fit degree of the research algorithm in GraphQ dataset is the highest,reaching 98.2%,in which the unsupervised learning method has underfit.The accuracy of word meaning understanding,syntax analysis and context understanding under the research method is above 95%.The average accuracy of the research system is 97.6%,which is 25.3%and 31.6%higher than that of the system based on recurrent neural network and unsupervised learning,respectively.The optimal path length of the robot in the research system is 153.4m when executing multi-objective dialogue requests.To sum up,the human-computer interactive dialogue system studied has high fitting degree and high accuracy.