Table 2.
The fault model and results of its application to Johnson et al.’s erroneous outputs. The rows in bold are the subtotaled percentages of the broad categories of errors in relation to all errors. The non-bolded rows indicate the percentages of the subtypes of errors in relation to the broad category that they belong to. The counts for the subtypes of text processing errors exceed the total text processing count because multiple types of text processing errors can contribute to one erroneously matched relationship.
| Inter-judge agreement | ||||
|---|---|---|---|---|
| Type of error | Percent | Count | pre-resolution | post-resolution |
| Lexical ambiguity errors | ||||
| biological polysemy | 56% | (105/186) | 86% | 98% |
| ambiguous abbreviation | 44% | (81/186) | 96% | 99% |
| Lexical Ambiguity Total | 38% | (186/481) | ||
|
| ||||
| Text processing errors | ||||
| Stemming | 6% | (29/449) | 100% | 100% |
| digit removal | 51% | (231/449) | 100% | 100% |
| punctuation removal | 27% | (123/449) | 100% | 100% |
| stop word removal | 14% | (65/449) | 99% | 100% |
| Text Processing Total | 60% | (290/481) | ||
|
| ||||
| Matched Metadata Total | 1% | (5/481) | 100% | 100% |
| Total | 99% | (481/481) | 95% | 99% |