The invention discloses a human-computer interaction risk
perception method for vehicle function safety, and the method comprises the steps: firstly, automatically unpacking an apk file of a vehicle-mounted app through an apktool, and obtaining all
xml files; analyzing all
xml files in the
layout, analyzing resource reference, generating a structuralized. Txt description, analyzing the structuralized description of a page by using LLM in combination with a current App
core function, and prompting a
large model to judge a main page; the method comprises the following steps of: independently analyzing a security problem of each main page according to an established standard
library through LLM concurrent worker on the basis of a retrieval enhanced
large model technology; then reasonably inferring a jump relation between pages through LLM according to AndroidManifest, and analyzing potential risks in a navigation process,
dialog box popping, associated permission and the like in combination with a standard
library; finally, safety problems found in the previous two steps are subjected to LLM duplicate removal and aggregation, sorting is carried out according to danger levels, modification suggestions are given, and a comprehensive report is formed and output. According to the invention, the vehicle-mounted App interaction safety detection efficiency and accuracy can be greatly improved.