mirror of
https://github.com/maxent-ai/zeroshot_topics.git
synced 2022-04-17 03:56:55 +03:00
69 lines
2.0 KiB
Python
69 lines
2.0 KiB
Python
import attr
|
|
from keybert import KeyBERT
|
|
from .utils import load_zeroshot_model
|
|
|
|
import nltk
|
|
nltk.download('wordnet')
|
|
|
|
from nltk.corpus import wordnet as wn
|
|
|
|
classifier = load_zeroshot_model()
|
|
|
|
|
|
@attr.s
|
|
class ZeroShotTopicFinder:
|
|
"""
|
|
Identifies topics in the given piece of text using wordnet and zero-shot classifiers.
|
|
|
|
step 1: Use transformer model to find the keywords form the text
|
|
step 2: Use the wordnet to find most similar words and expand the keyword list.
|
|
step 3: Find the hypernyms for all the keywords.
|
|
step 4: Pass the list of hypernyms and text to your choice of zero-shot classifier and get the top labels as topic.
|
|
"""
|
|
|
|
model = attr.ib(default='all-MiniLM-L6-v2')
|
|
|
|
def __attrs_post_init__(self):
|
|
self.model = KeyBERT(self.model)
|
|
|
|
def find_topic(self, text, n_topic=2):
|
|
"""
|
|
Infer the topic in a given string.
|
|
parameters
|
|
----------
|
|
|
|
text: str
|
|
pass the text for which you want to infer the topic.
|
|
|
|
n_topic: int
|
|
Define the maximum number of topic you want to identify.
|
|
|
|
|
|
Returns
|
|
-------
|
|
labels : list
|
|
List of topics identified.
|
|
|
|
"""
|
|
keyword = self.get_keyword(text)
|
|
labels = self.get_parent_words(keyword)
|
|
prediction = classifier(text, candidate_labels=labels)
|
|
labels = prediction['labels'][:n_topic]
|
|
labels = [i.replace("_", ' ').title() for i in labels]
|
|
return labels
|
|
|
|
def get_keyword(self, text):
|
|
kw = [i[0] for i in self.model.extract_keywords(text)]
|
|
return kw
|
|
|
|
def get_parent_words(self, keywords):
|
|
parents = []
|
|
for kw in keywords:
|
|
sym = wn.synsets(kw)[:2]
|
|
parents += [j.name().split('.')
|
|
for i in sym for j in i.hypernyms()]
|
|
parents = [i[0] for i in parents if i[1] != 'v']
|
|
parents = [i for i in parents if i not in keywords]
|
|
parents = list(set(parents))
|
|
return parents
|