在自定义数据集上训练 Spacy NER 出错

Question

I am trying to train spacy NER model on custom dataset. Basically I want to use this model to extract Name, Organization, Email, phone number etc from resume.

Below is the code I am using.

import json
import random
import spacy
import sys
import logging
from sklearn.metrics import classification_report
from sklearn.metrics import precision_recall_fscore_support
from spacy.gold import GoldParse
from spacy.scorer import Scorer
from sklearn.metrics import accuracy_score
from spacy.gold import biluo_tags_from_offsets
def convert_dataturks_to_spacy(dataturks_JSON_FilePath):

    try:
        training_data = []
        lines=[]
        with open(dataturks_JSON_FilePath, encoding='utf-8') as f:
            lines = f.readlines()
        for line in lines:
            data = json.loads(line)
            text = data['content']
            entities = []
            for annotation in data['annotation']:
                #only a single point in text annotation.
                point = annotation['points'][0]
                labels = annotation['label']
                if not isinstance(labels, list):
                    labels = [labels]

                for label in labels:
                    entities.append((point['start'], point['end'] + 1 ,label))


            training_data.append((text, {"entities" : entities}))
        return training_data
    except Exception as e:
        logging.exception("Unable to process " + dataturks_JSON_FilePath + "\n" + "error = " + str(e))
        return None

def reformat_train_data(tokenizer, examples):
    output = []
    for i, (text, entity_offsets) in enumerate(examples):
        doc = tokenizer(text.strip())
        ner_tags = biluo_tags_from_offsets(tokenizer(text), entity_offsets['entities'])
        words = [w.text for w in doc]
        tags = ['-'] * len(doc)
        heads = [0] * len(doc)
        deps = [''] * len(doc)
        sentence = (range(len(doc)), words, tags, heads, deps, ner_tags)
        output.append((text, [(sentence, [])]))
    print("output",output)
    return output
################### Train Spacy NER.###########
def train_spacy():
    TRAIN_DATA = convert_dataturks_to_spacy("C:\\Users\\akjain\\Downloads\\Entity-Recognition-In-Resumes-SpaCy-master\\traindata.json")
    nlp = spacy.blank("en")  
    if 'ner' not in nlp.pipe_names:
        ner = nlp.create_pipe('ner')
        nlp.add_pipe(ner, last=True)    
    # add labels
    for _, annotations in TRAIN_DATA:
        for ent in annotations.get('entities'):
            ner.add_label(ent[2])

    def get_data(): return reformat_train_data(nlp.tokenizer, TRAIN_DATA)
    optimizer = nlp.begin_training(get_data)
    for itn in range(10):
        print("Starting iteration " + str(itn))
        random.shuffle(TRAIN_DATA)
        losses = {}
        for text, annotations in TRAIN_DATA:
            nlp.update(
                [text],  # batch of texts
                [annotations],  # batch of annotations
                drop=0.2,  # dropout - make it harder to memorise data
                sgd=optimizer,  # callable to update weights
                losses=losses)
        print(losses)
train_spacy()

I am getting the below error. Also, I came across a link ( https://github.com/explosion/spaCy/issues/3558 ) with some suggestion to fix this code. But even after implementing that I am still getting error.

I am using Python 3.6.5 and Spacy 2.2.3

Dataset:

{"content": "Nida Khan\nTech Support Executive - Teleperformance for Microsoft\n\nJaipur, Rajasthan - Email me on Indeed: indeed.com/r/Nida-Khan/6c9160696f57efd8\n\n• To be an integral part of the organization and enhance my knowledge to utilize it in a productive\nmanner for the growth of the company and the global.\n\nINDUSTRIAL TRAINING\n\n• BHEL, (HEEP) HARIDWAR\nOn CNC System& PLC Programming.\n\nWORK EXPERIENCE\n\nTech Support Executive\n\nTeleperformance for Microsoft -\n\nSeptember 2017 to Present\n\nprocess.\n• 21 months of experience in ADFC as Phone Banker.\n\nEDUCATION\n\nBachelor of Technology in Electronics & communication Engg\n\nGNIT institute of Technology -  Lucknow, Uttar Pradesh\n\n2008 to 2012\n\nClass XII\n\nU.P. Board -  Bareilly, Uttar Pradesh\n\n2007\n\nClass X\n\nU.P. Board -  Bareilly, Uttar Pradesh\n\n2005\n\nSKILLS\n\nMicrosoft office, excel, cisco, c language, cbs. (4 years)\n\nhttps://www.indeed.com/r/Nida-Khan/6c9160696f57efd8?isid=rex-download&ikw=download-top&co=IN","annotation":[{"label":["Email Address"],"points":[{"start":872,"end":910,"text":"indeed.com/r/Nida-Khan/6c9160696f57efd8"}]},{"label":["Skills"],"points":[{"start":800,"end":857,"text":"Microsoft office, excel, cisco, c language, cbs. (4 years)"}]},{"label":["Graduation Year"],"points":[{"start":676,"end":679,"text":"2012"}]},{"label":["College Name"],"points":[{"start":612,"end":640,"text":"GNIT institute of Technology "}]},{"label":["Degree"],"points":[{"start":552,"end":609,"text":"Bachelor of Technology in Electronics & communication Engg"}]},{"label":["Companies worked at"],"points":[{"start":420,"end":448,"text":"Teleperformance for Microsoft"}]},{"label":["Designation"],"points":[{"start":395,"end":417,"text":"\nTech Support Executive"}]},{"label":["Email Address"],"points":[{"start":106,"end":144,"text":"indeed.com/r/Nida-Khan/6c9160696f57efd8"}]},{"label":["Location"],"points":[{"start":66,"end":71,"text":"Jaipur"}]},{"label":["Companies worked at"],"points":[{"start":35,"end":63,"text":"Teleperformance for Microsoft"}]},{"label":["Designation"],"points":[{"start":10,"end":32,"text":"Tech Support Executive "}]},{"label":["Designation"],"points":[{"start":9,"end":31,"text":"\nTech Support Executive"}]},{"label":["Name"],"points":[{"start":0,"end":8,"text":"Nida Khan"}]}]}

Answer 1

The problem is you are feeding training data to model optimizer.

As mentioned in https://github.com/explosion/spaCy/issues/3558 , use the following function to remove leading and trailing white spaces from entity spans.

def trim_entity_spans(data: list) -> list:
    """Removes leading and trailing white spaces from entity spans.

    Args:
    data (list): The data to be cleaned in spaCy JSON format.

    Returns:
    list: The cleaned data.
    """
    invalid_span_tokens = re.compile(r'\s')

    cleaned_data = []
    for text, annotations in data:
        entities = annotations['entities']
        valid_entities = []
        for start, end, label in entities:
            valid_start = start
            valid_end = end
            # if there's preceding spaces, move the start position to nearest character
            while valid_start < len(text) and invalid_span_tokens.match(
                    text[valid_start]):
                valid_start += 1
            while valid_end > 1 and invalid_span_tokens.match(
                    text[valid_end - 1]):
                valid_end -= 1
            valid_entities.append([valid_start, valid_end, label])
        cleaned_data.append([text, {'entities': valid_entities}])
    return cleaned_data

Then use the following function for training:

def train_spacy():

    TRAIN_DATA = convert_dataturks_to_spacy("C:\\Users\\akjain\\Downloads\\Entity-Recognition-In-Resumes-SpaCy-master\\traindata.json")
    TRAIN_DATA = trim_entity_spans(TRAIN_DATA)
    nlp = spacy.blank('en')  # create blank Language class
    # create the built-in pipeline components and add them to the pipeline
    # nlp.create_pipe works for built-ins that are registered with spaCy
    if 'ner' not in nlp.pipe_names:
        ner = nlp.create_pipe('ner')
        nlp.add_pipe(ner, last=True)


    # add labels
    for _, annotations in TRAIN_DATA:
         for ent in annotations.get('entities'):
            ner.add_label(ent[2])

    # get names of other pipes to disable them during training
    other_pipes = [pipe for pipe in nlp.pipe_names if pipe != 'ner']
    with nlp.disable_pipes(*other_pipes):  # only train NER
        optimizer = nlp.begin_training()
        for itn in range(10):
            print("Statring iteration " + str(itn))
            random.shuffle(TRAIN_DATA)
            losses = {}
            for text, annotations in TRAIN_DATA:
                nlp.update(
                [text],  # batch of texts
                [annotations],  # batch of annotations
                drop=0.2,  # dropout - make it harder to memorise data
                sgd=optimizer,  # callable to update weights
                losses=losses)
            print(losses)