Penalize relationship scores in cases for each non-match

[paulalbert1]

Problem

Some articles are scoring too highly, mostly because there are hundreds of coauthors, and by sheer chance, one or several of those co-authors matches a known relationship.

For example, CWID=pas2026 , and PMID=31031568

Suggested fix

Penalize each relationship non-match a small amount, say, -0.06. As a whole, set a minimum total score, say, -2.

Existing

        "relationshipEvidence": [
          {
            "relationshipNameArticle": {
              "firstName": "Erika L",
              "firstInitial": "E",
              "lastName": "Abramson"
            },
            "relationshipNameIdenity": {
              "firstName": "Erika",
              "firstInitial": "E",
              "lastName": "Abramson"
            },
            "relationshipType": [
              "Co-investigator"
            ],
            "relationshipMatchType": "verbose",
            "relationshipMatchingScore": 2.2,
            "relationshipVerboseMatchModifierScore": 0.6,
            "relationshipMatchModifierMentor": 0,
            "relationshipMatchModifierMentorSeniorAuthor": 0,
            "relationshipMatchModifierManager": 0,
            "relationshipMatchModifierManagerSeniorAuthor": 0
          },
          {
            "relationshipNameArticle": {
              "firstName": "Sameer",
              "firstInitial": "S",
              "lastName": "Malhotra"
            },
            "relationshipNameIdenity": {
              "firstName": "Sameer",
              "firstInitial": "S",
              "lastName": "Malhotra"
            }
        ],

Proposed

Add these values to application.properties

strategy.knownrelationships.relationshipMinimumTotalScore=-2
strategy.knownrelationships.relationshipNonMatchScore=-0.06

Change the feature generator output as follows.

        "relationshipEvidence": [
          {
            "relationshipNonMatchCount": 51,
            "relationshipNonMatchScore":-0.06,
            "relationshipMinimumTotalScore":-2
          },
          {
            "relationshipNameArticle": {
              "firstName": "Erika L",
              "firstInitial": "E",
              "lastName": "Abramson"
            },
            "relationshipNameIdenity": {
              "firstName": "Erika",
              "firstInitial": "E",
              "lastName": "Abramson"
            },
            "relationshipType": [
              "Co-investigator"
            ],
            "relationshipMatchType": "verbose",
            "relationshipMatchingScore": 2.2,
            "relationshipVerboseMatchModifierScore": 0.6,
            "relationshipMatchModifierMentor": 0,
            "relationshipMatchModifierMentorSeniorAuthor": 0,
            "relationshipMatchModifierManager": 0,
            "relationshipMatchModifierManagerSeniorAuthor": 0
          },
          {
            "relationshipNameArticle": {
              "firstName": "Sameer",
              "firstInitial": "S",
              "lastName": "Malhotra"
            },
            "relationshipNameIdenity": {
              "firstName": "Sameer",
              "firstInitial": "S",
              "lastName": "Malhotra"
            "relationshipType": [
              "Co-investigator"
            ],
            "relationshipMatchType": "verbose",
            "relationshipMatchingScore": 2.2,
            "relationshipVerboseMatchModifierScore": 0.6,
            "relationshipMatchModifierMentor": 0,
            "relationshipMatchModifierMentorSeniorAuthor": 0,
            "relationshipMatchModifierManager": 0,
            "relationshipMatchModifierManagerSeniorAuthor": 0
          }
        ],

Here, the revised relationship score would be: 2.2 + 2.2 + (51 * -0.06) = 1.34.

Now, let's suppose that the number of co-authors were 800. 2.2 + 2.2 + (800 * -0.06) = -43.6. Because -2 > -43.6, we go with -2.

1. Compute the scores for positive matches.
2. Compute the scores for negative matches.
3. Positive match + negative match = TotalProvisionalRelationshipScore
4. If TotalProvisionalRelationshipScore < relationshipMinimumTotalScore, then relationshipMinimumTotalScore. Else, TotalProvisionalRelationshipScore.

[sarbajitdutta]

@paulalbert1 Would this affect publications with less authors?

[paulalbert1]

Yes, but the effect would be minimal. If you have four authors and no relationship matches, the decrement in score would be only -0.18 (3 non-target authors). Since this would affect all such articles consistently, it wouldn't have a significant relative effect. The greatest effect would be on the articles with a gazillion authors.

[sarbajitdutta]

OK gotcha would you want to include this in the ReCiter release 1.1? @paulalbert1

[paulalbert1]

That would be great!