[BUG] Exact matching failing for some records in affiliation endpoint

[adambuttrick]

Version v1, v2

Describe the bug In the affiliation endpoint, when using as input the primary name of an organization, some names fail to return an exact match. This does not appear specific to records in a single release, as examples (all from v1.45) below demonstrates.

To Reproduce Steps to reproduce the behavior:

Example 1:

1. Query the affiliation endpoint using the display name for https://ror.org/02xhx4j26 - Centre for Marine Socioecology in v1 and v2: https://api.ror.org/v1/organizations?affiliation=Centre%20for%20Marine%20Socioecology https://api.ror.org/v2/organizations?affiliation=Centre%20for%20Marine%20Socioecology
2. 0 results are returned in v1 and v2

Example 2:

1. Query the affiliation endpoint using the display name for https://ror.org/00arpt780 - Institute for Marine and Antarctic Studies in v1 and v2: https://api.ror.org/v1/organizations?affiliation=Institute%20for%20Marine%20and%20Antarctic%20Studies https://api.ror.org/v2/organizations?affiliation=Institute%20for%20Marine%20and%20Antarctic%20Studies

2. 1 results is returned in v1 and 2 results in v2, with no chosen=True in either.

Example 3:

1. Query the affiliation endpoint using the display name for https://ror.org/026xeq875 - Tasmanian Behavioural Lab in v1 and v2: https://api.ror.org/v1/organizations?affiliation=Tasmanian%20Behavioural%20Lab https://api.ror.org/v2/organizations?affiliation=Tasmanian%20Behavioural%20Lab

2. 1 results is returned in v1 and in v2, both with chosen=True for the ROR ID

Expected behavior Exact matches on all records where the ROR display or any other names on the record are used.

[adambuttrick]

@dtkaczyk took a look at this and this appears to be the result of our logic that maps US states names to the US as a country. Since the similarity between “marine” and “Maine” is above the matching score (https://github.com/ror-community/ror-api/blob/94bad807f188f22c80a3234fcbf7b48e52f01818/rorapi/common/matching.py#L111), the strategy recognizes this affiliation as being from the US. Since the country for the correct records doesn’t match (all examples are from Australia), they're rejected. We should perhaps adjust the state matching logic to be exact vs. fuzzy to account for this.

[adambuttrick]

Adding word boundaries matchers around the name value appears to fix this, e.g.:

def get_country_codes(string):
    """Extract the country codes from the string,  
    if the country names are mentioned."""

    string = unidecode.unidecode(string).strip()
    lower = re.sub(r"\s+", " ", string.lower()) 
    lower_alpha = re.sub(r"\s+", " ", re.sub("[^a-z]", " ", string.lower()))
    alpha = re.sub(r"\s+", " ", re.sub("[^a-zA-Z]", " ", string))
    codes = []
    for code, name in COUNTRIES:
        if re.search("[^a-z]", name):
            score = fuzz.partial_ratio(name, lower)
        elif len(name) == 2:  
            score = max([fuzz.ratio(name.upper(), t) for t in alpha.split()])
        else:
            # Add word boundary matchers around the name 
            regex = r'\b' + re.escape(name) + r'\b'
            score = 100 if re.search(regex, lower_alpha) else 0
        if score >= 90:
            codes.append(code.upper())
    return list(set(codes))