Skip to main content

Table 6 Performance of state-of-the-art MIST de-identification system against our system, over the development and test sets

From: A system for de-identifying medical message board text

 

Development (BC)

Test (Arthritis)

  

Recall

Precision

Recall

Precision

MIST

Proper names

79.9%

80.5%

72.2%

74.9%

 

Usernames

49.5%

 

34.3%

 
 

All names

73.0%

 

54.6%

 

Proposed system

Proper names

98.9%

67.4%

99.0%

61.4%

 

Usernames

94.5%

 

90.3%

 
 

All names

98.1%

 

94.3%

Â