This repository contains the MiniDetector system, designed for detecting consistency issues within Mini-Apps. It provides tools for analyzing data flow, extracting entities from privacy policies, and conducting consistency detection.

• data

  • groundtruth: contains the groundtruth of our approach, including 70 mini-apps (data flow and privacy policy). Due to the size restriction of the GitHub repo, only 5 samples are included. For more samples, please download from Google Drive. (Update 2026/02: Sorry to permission management restrictions (under the research group’s asset account), the dataset previously required an application for access. It has now been adjusted and is publicly accessible.)
  • policies: contains 100 privacy policies from 50 popular apps (both Android and mini-apps).

• ontology: includes two ontologies, covering data types and operations.

• flow: analyzes data flow. This tool performs taint analysis for data flow in WeChat mini-apps.

• entity: entity extraction in privacy policies.

  • llm.py: retrieves practical entities from privacy policies.
  • extract.py: extracts privacy policies from static program code.

• consistency: consistency detection module.

• Step 1. Use flow to obtain data flow results.

  • For analyzing a single mini program, its entry file is main.js, and you can modify the corresponding configuration properties __cfg__, such as the mini program directory.
  • For batch analysis, please refer to the batch.js file.
  • After the analysis is completed, a __res__ result file will be generated in the directory.

• Step 2. Use entity to extract privacy policies from code or data practices from privacy policies.

  • The file llm.py contains the main code for querying data practices using a LLM. You need to configure the corresponding API token and select an appropriate model.

• Step 3. Utilize consistency for consistency detection.

  • The file detection.py contains a predefined list corresponding to the API-data practice. You can use this file to compare and analyze the data stream results with the LLM query results, ultimately obtaining the consistency comparison results for the mini-app.

• Node.js 12.19.0
• esprima
• Python 3.6.13

For further inquiries or assistance, feel free to contact us.

Wang, [email protected]