1，文本纠错(query纠错)，可用于爬取的新闻资讯等进行预处理，去掉错别字、可用于搜索业务中query词纠错、可用于对话中的智能改错。中文相关的纠错paper

两个指标：过纠率（FAR，也就误报率），召回率

过纠率：正确的句子被改错的比率（FAR=正确句子被错纠的个数/正确句子个数）；召回率：错误的句子被全部纠正的比率。较大的过纠率将会对系统和用户体验带来负面效果。因而，纠对句子数量远远大于被改错句子的数量，如果句子出错概率是K，则K*RECALL>>(1-K)*FAR。

github1，并不是说有了huggingface就可以拿来用，关键是在此基础上进行修改和设计模型，fine-tune，这才是水平和能力。

2，掩蔽词masked language model

from transformers import pipeline
unmasker = pipeline('fill-mask', model='bert-base-uncased')
unmasker("what a [MASK] blog.")
[{'score': 0.32585206627845764, 'token': 2307, 'token_str': 'great', 'sequence': 'what a great blog.'}, {'score': 0.04082264378666878, 'token': 3376, 'token_str': 'beautiful', 'sequence': 'what a beautiful blog.'}, {'score': 0.040087465196847916, 'token': 8403, 'token_str': 'lovely', 'sequence': 'what a lovely blog.'}, {'score': 0.03804076090455055, 'token': 6919, 'token_str': 'wonderful', 'sequence': 'what a wonderful blog.'}, {'score': 0.028535649180412292, 'token': 2204, 'token_str': 'good', 'sequence': 'what a good blog.'}]

获得sentence embedding，当然也有sentence Transform那个库，这里也可以的。

from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
model = BertModel.from_pretrained("bert-base-uncased")
text = "what a great blog I have seen."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
output.pooler_output#(1,768)
output.last_hidden_state.shape
torch.Size([1, 12, 768])

3，实体词识别NER

先说分词，这是基本操作，模型是上面的uncased，wordpiece级别分词

from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
sentence = "what a great blog I have seen in China!"
tokens = tokenizer.tokenize(sentence)
>>> tokens
['what', 'a', 'great', 'blog', 'i', 'have', 'seen', 'in', 'china', '!']
#全部改成了小写
ids=tokenizer.convert_tokens_to_ids(tokens)
>>> ids
[2054, 1037, 2307, 9927, 1045, 2031, 2464, 1999, 2859, 999]
strings=tokenizer.decode(ids)
>>> strings
'what a great blog i have seen in china!'

采用cased模型，当然其中的ids也不同，因为训练数据不同，那么词表也不同，故而ids不同

tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")
。。。
>>> tokens
['what', 'a', 'great', 'blog', 'I', 'have', 'seen', 'in', 'China', '!']
>>> ids
[1184, 170, 1632, 10679, 146, 1138, 1562, 1107, 1975, 106]
>>> strings
'what a great blog I have seen in China!'

NER 实体识别就是实体分类（entity token classification）

types of entities: location (LOC), organizations (ORG), person (PER) and Miscellaneous (MISC).

位置，组织，人名，杂乱的，这4个是实体识别的重点，又具体分为：

总之，前加B就是Beginning，I就是本体，I可以去掉，O就是啥都不是，非实体。

from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipelinetokenizer = AutoTokenizer.from_pretrained("dslim/bert-base-NER")
model = AutoModelForTokenClassification.from_pretrained("dslim/bert-base-NER")
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "My name is Xiaoming and I live in Beijing, I love you forver !"
ner_results = nlp(example)
print(ner_results)
[{'entity': 'B-PER', 'score': 0.9986733, 'index': 4, 'word': 'Xiao', 'start': 11, 'end': 15}, {'entity': 'I-PER', 'score': 0.9260187, 'index': 5, 'word': '##ming', 'start': 15, 'end': 19}, {'entity': 'B-LOC', 'score': 0.99963117, 'index': 10, 'word': 'Beijing', 'start': 34, 'end': 41}]

替换成bert-base-cased模型，结果都是Label-0，和label-1，这种0/1分类误差太大，意义不大。

当然也有较为基础的model，直接来用 （如果没有实体词，则为[]）

from transformers import pipeline
ner_pipe = pipeline("ner")
sequence = """I'm not the best but the great in recommondation system at Beijing, now I'm looking forward to your kind reply about the offer, if you have any question about job or my work, please contact me without hesitation, I will give you answer in time. 
Any other problem you can join in the QQ group 277356808 !"""
>>> ner_pipe(sequence)
[{'entity': 'I-LOC', 'score': 0.99978286, 'index': 17, 'word': 'Beijing', 'start': 59, 'end': 66}]

4，情感分析（sentiment analysis），发现仅仅就是分类问题，就是分词后对其中具有positive，negative的词进行标记，分类，这就是情感分析，仅此而已。

斯坦福NLP2013，NLTK

from transformers import pipeline
classifier = pipeline("sentiment-analysis")
result = classifier("I love you forver !")
>>> result
[{'label': 'POSITIVE', 'score': 0.9998611211776733}]

稍微复杂点的code，分类也多了1个中性

from transformers import AutoModelForSequenceClassification
from transformers import TFAutoModelForSequenceClassification
from transformers import AutoTokenizer, AutoConfig
import numpy as np
from scipy.special import softmax
def preprocess(text):new_text = []for t in text.split(" "):t = '@user' if t.startswith('@') and len(t) > 1 else tt = 'http' if t.startswith('http') else tnew_text.append(t)return " ".join(new_text)
MODEL = f"cardiffnlp/twitter-roberta-base-sentiment-latest"
tokenizer = AutoTokenizer.from_pretrained(MODEL)
config = AutoConfig.from_pretrained(MODEL)
# PT
model = AutoModelForSequenceClassification.from_pretrained(MODEL)
#model.save_pretrained(MODEL)
text = "what a great blog I have ever seen in China!"
text = preprocess(text)
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
scores = output[0][0].detach().numpy()
scores = softmax(scores)
ranking = np.argsort(scores)
ranking = ranking[::-1]
for i in range(scores.shape[0]):l = config.id2label[ranking[i]]s = scores[ranking[i]]print(f"{i+1}) {l} {np.round(float(s), 4)}")1) positive 0.9859
2) neutral 0.0116
3) negative 0.0025

5，文本摘要生成summarization，参考1

from transformers import AutoModelWithLMHead, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("mrm8488/t5-base-finetuned-summarize-news")
model = AutoModelWithLMHead.from_pretrained("mrm8488/t5-base-finetuned-summarize-news")
def summarize(text, max_length=150):input_ids = tokenizer.encode(text, return_tensors="pt", add_special_tokens=True)generated_ids = model.generate(input_ids=input_ids, num_beams=2, max_length=max_length,  repetition_penalty=2.5, length_penalty=1.0, early_stopping=True)preds = [tokenizer.decode(g, skip_special_tokens=True, clean_up_tokenization_spaces=True) for g in generated_ids]return preds[0]
>>> summarize('After the sound and the fury, weeks of demonstrations and anguished calls for racial justice, the man whose death gave rise to an international movement, and whose last words — “I can’t breathe” — have been a rallying cry, will be laid to rest on Tuesday at a private funeral in Houston.George Floyd, who was 46, will then be buried in a grave next to his mother’s.The service, scheduled to begin at 11 a.m. at the Fountain of Praise church, comes after five days of public memorials in Minneapolis, North Carolina and Houston and two weeks after a Minneapolis police officer was caught on video pressing his knee into Mr. Floyd’s neck for nearly nine minutes before Mr. Floyd died. That officer, Derek Chauvin, has been charged with second-degree murder and second-degree manslaughter. His bail was set at $1.25 million in a court appearance on Monday. The outpouring of anger and outrage after Mr. Floyd’s death — and the speed at which protests spread from tense, chaotic demonstrations in the city where he died to an international movement from Rome to Rio de Janeiro — has reflected the depth of frustration borne of years of watching black people die at the hands of the police or vigilantes while calls for change went unmet.', 80)"at a private funeral in Houston on Tuesday. Floyd, who was 46, died of multiple organ failure last month. A Minnesota police officer was caught on video pressing his knee into Mr. Floyd’s neck for nearly nine minutes before he died. A Minneapolis police officer has been charged with second-degree murder and manslaughter. Floyd's bail was set at $1"

愿我们终有重逢之时，而你还记得我们曾经讨论的话题