zshot spaCy ignoring email and phone entities?

[jakesno]

Hi All,

We are using spaCy for entity and keyword extraction and noticed that it generally ignores phone numbers and emails within our text data. I have included a basic, redacted example below where would have expected the phone number to be extracted.

Example, "Our Air Conditioning Installation service in XXXX offers top-quality solutions to keep your home or office cool and comfortable. Call us ###-###-#### or email at [email protected]"

Anything im missing?

[svlandeg]

Hi!

The pretrained NER model from the English models doesn't have specific labels for Email or Phone_number. It does however recognize Cardinal and Ordinal entities, which might be relevant for your use-case. They might not capture exactly what you want though.

For this type of information extraction, I would recommend using builtin attributes that spaCy defines on the token-level, such as like_num and like_email. These can then be used to design custom matcher rules, see here.

So for instance, you can write something like:

    text = "Hello Jake is your email still [email protected] and your phone number 0342-42-66?"
    nlp = spacy.load("en_core_web_lg")
    doc = nlp(text)
    for ent in doc.ents:
        print(ent.text, ent.label_)
    print()
    for token in doc:
        if token.like_email:
            print(token, "Email")

    matcher = Matcher(nlp.vocab)
    matcher.add("Phone Number", patterns=[[{"LIKE_NUM": True}, {"IS_PUNCT": True}, {"LIKE_NUM": True}, {"IS_PUNCT": True}, {"LIKE_NUM": True}]])

    matches = matcher(doc, as_spans=True)
    for match in matches:
        print(match.text, match.label_)

and you'll get

Jake PERSON
0342-42 CARDINAL
[email protected] Email
0342-42-666 Phone Number