Vehicle Instructions Extraction Based On Natural Language Understanding

Vehicle travel instructions extraction which extracts structured information from an unstructured vehicle traveling text, is an application of information extraction in the field of human-computer interaction.For a vehicle travel instruction text, if we want to extract some structural information which can able to guide vehicle travel correctly, such as direction and speed which is used to control vehicle to travel, it is critical to position these words or phrases and mark one description which belongs to their own semantic description.We build an automatic vehicle travel instructions extraction system (VIAES) which is based on natural language processing. The system consists of four modules:①semantic classification module:marked with a label which is belong to its own category for words or phrases in the instruction text;②detect whether a instruction text is vehicle instruction module:to detect a instruction text that user inputs whether it is related to vehicle instruction;③structure segmentation module:to cut a instruction which own multiple sub-instructions into a plurality of structure;④structuralization module:to represent a semantic classification sequence with six elements that can control vehicle travelling.This paper proposes a semantic classification method which combining CRF, self-training and Dictionary. The experiment results show that our method can be effective in semantic classification for a vehicle travel instruction text,the overall correct rate is92%, and the method of this article propose has a good portability.According semantic classification sequences obtained, we use some classification algorithms to detect whether a text is belong to the domain of vehicle. The experiment shows that the GMM model is best.A vehicle travel instruction of Natural language description may be composed by multiple sub-instructions, such as:“车辆向前行驶500米右转”,is composed by two structures. Structure1:“向前行驶500米”,structure2:“右转”.The structural segmentation can be seen as a sequence labeling problem. We use CRF to do structure segmentation for semantic classification sequences. The experiment shows that accuracy rate of structure segmentation reach to96.89%.The structuralization module contains two parts, one is the start-end location identification, such as“从蓟门桥向前行驶到知春路”where“蓟门桥”is the start location, and"知春路”is the end location. We use label propagation algorithm to solve the problem of start-end position identification. The experiment results show that the accuracy rate of this model is92.78%. Another is structuralization, which is used to represent a semantic classification sequence (after start-end location identification) with six elements that can control vehicle travelling. The six elements are:the start location, speed, distance, direction, action and the end location.Vehicle travel instructions automatic extraction system(VIAES), based on natural language processing, not only is a specific application in the field of human-computer interaction, but also is an exploration for intelligent travel. The overall accuracy of VIAES is89.7%.