The Design And Implementation Of Driver Assistance Information System

Driver-car interaction should not affect driving. Compared to traditionalinteractive mode, voice provides a more secure and convenient way tointeract between driver and car. All your requirements with your car duringdriving can be met with your voice. Voice driver assistance system needs thecombination of voice recognition technology and natural languageunderstanding technology. It can greatly improve the driving experience.Based on related domestic and international research and work, an in-depthanalysis and research on the issue of how to understand queries efficientlyand accurately is carries out on this article. The design and implementation ofChinese driver assistance information system is based on the above analysisand research. The main work is as follow:(1) Based on extensive research on short text classification, conditionalrandom fileds, linear SVM and maximum likelihood are used to classify thewidely existed special nouns and organization names.(2) Lots of intentions are collected and they are divided into severalcategorizations. Then a database is built. As the intention inference is basedon words and part-of-speech information will also be used, an automaticparser based on conditional random fields is built. Given a query, the parser isbuilt in with both one-step and two-step approaches to output quite accuratesegmentation and part-of-speech tagging results.(3) Based on extensive research on natural language understanding andintention inference, an intention inference module based on maximumentropy model is built after the implementation of automatic parser. Duringthe feature selection procedure not only the unigram word features but alsosome bigram features are selected. The selected bigram features should haverelatively high information gain and kinds of meaningful syntax meaning.Higher accuracy is got when bigram features are added and it can basically meet practical requirements.The driver assistant information system described in this article is thecombination of the classifier for special nouns and organization names,automatic parser module and intention inference module. Given a query, theparser module will output its segmentation and part-of-speech tagging whichwill act as the input of intention inference module. The intention inferencemodule will then output the intention inference result and correspondingactions will be taken by the system to improve the driving experience. Andthe classifier can handle the widely existed special nouns and organizationnames quite efficiently to improve the performanece of the system.Experiment results shows this system can efficiently understand queries witha quite high accuracy.