appch's Introduction

APPCH Corpus

The APPCH Corpus (Application Privacy Policies in Chinese) is a collection of application privacy policies (i.e., in natural language) with annotations, which reveals privacy information collection practice through nine dangerous Android PERMISSIONs.

PERMISSION Group	PERMSSIONs	LABEL
CALENDAR	READ_CALENDAR、WRITE_CALENDAR	CAL
CAMERA	CAMERA	CAM
CONTACTS	READ_CONTACTS、WRITE_CONTACTS、GET_ACCOUNTS	CON
LOCATION	ACCESS_FINE_LOCATION、ACCESS_COARSE_LOCATION	LOC
MICROPHONE	RECORD_AUDIO	MIC
PHONE	READ_PHONE_STATE、CALL_PHONE、READ_CALL_LOG、 WRITE_CALL_LOG、ADD_VOICEMAIL、USE_SIP、 PROCESS_OUTGOING_CALLS	PHO
SENSORS	BODY_SENSORS	SEN
SMS	SEND_SMS、RECEIVE_SMS、READ_SMS、 RECEIVE_WAP_PUSH、RECEIVE_MMS	SMS
STORAGE	READ_EXTERNAL_STORAGE、WRITE_EXTERNAL_STORAGE	STO

In manual annotation sets, each privacy policy was read and annotated by three annotators from Cyber Engneering School. By manual annotation, we got 98 labeled policies and a dictionary of 678 data practice phrases.
Based on the above permission phrase dictionary, we made use of the BMM (Bidirectional Maximum Matching) algorithm for automatic annotation to expand the size of the corpus.
Finally, APPCH Copus contained 1058 privacy policies and 948K dangerous permission phrases. We can build train/dev/test datasets based APPCH and then train NER(Named Entity Recognition) model to extract data practice from privacy policy.

The dataset is made available for research, teaching, and scholarship purposes only. You can find raw corpus in "spider" and labeled corpus in "annotations".

Special thanks to: [YEDDA]：https://github.com/jiesutd/YEDDA

Recommend Projects