赞
踩
文本摘要是自然语言处理(NLP)的一种应用,随着人工智能的发展文本提取必将对我们的生活产生巨大的影响。随着网络的发展我们处在一个信息爆炸的时代,通读每天更新的海量文章/文档/书籍会占用我们大量的时间,所以用一种算法帮我们提取一篇文章的关键信息是非常高效的。谢天谢地,这项技术已经出现了。你有没有遇到过inshorts的手机应用?这是一款创新的新闻应用程序,可以将新闻文章转换成60字的摘要。这正是我们在这篇文章中要学习的——自动文本摘要提取。
自动文本摘要早在20世纪50年代就引起了人们的注意。汉斯•彼得•鲁恩(Hans Peter Luhn)在20世纪50年代末发表了一篇研究论文,题为《文学文摘的自动创作》(the automatic creation of literature abstracts)。该论文利用词频和短语频等特征,从文本中提取重要句子进行总结。
另一项重要的研究是Harold P Edmundson在20世纪60年代末所做的,该研究利用线索词的出现、出现在文章标题中的词以及句子的位置等方法,提取出有意义的句子进行文本总结。从那时起,许多重要和令人兴奋的研究已经发表,以解决自动文本摘要的挑战。
下面来看TextRank算法进行网球类文章的摘要提取实例
一、TextRank算法流程
我们选取网球类文章来进行我们的文本摘要提取实战,我们将以多篇文章作为输入,并生成单个项目符号摘要。本文不讨论多域文本摘要,但您可以在文章末尾尝试它。
数据集地址:https://s3-ap-south-1.amazonaws.com/av-blog-media/wp-content/uploads/2018/10/tennis_articles_v4.csv
三、python代码
1、导入库
- import numpy as np
- import pandas as pd
- import nltk
- nltk.download('punkt') # one time execution
- import re
2、读取并查看数据
- df = pd.read_csv("tennis_articles_v4.csv")
- df.head()
3、将文本分成句子
- from nltk.tokenize import sent_tokenize
- sentences = []
- for s in df['article_text']:
- sentences.append(sent_tokenize(s))
-
- sentences = [y for x in sentences for y in x] # flatten list
4、下载GloVe词嵌入
GloVe词嵌入是词的向量表示。这些词的嵌入将被用来为我们的句子创建向量。我们也可以使用单词包或TF-IDF方法为句子创建特征,但是这些方法忽略了单词的顺序(特征的数量通常相当大)。我们将使用预先培训的维基百科2014 + Gigaword5 GloVe矢量,这些单词嵌入的大小是822 MB。
- !wget http://nlp.stanford.edu/data/glove.6B.zip
- !unzip glove*.zip
5、提取单词嵌入或单词向量
- # Extract word vectors
- word_embeddings = {}
- f = open('glove.6B.100d.txt', encoding='utf-8')
- for line in f:
- values = line.split()
- word = values[0]
- coefs = np.asarray(values[1:], dtype='float32')
- word_embeddings[word] = coefs
- f.close()
6、文本处理
对文本数据做一些基本的文本清理以尽可能避免文本数据的噪音对摘要提取的影响。
- # remove punctuations, numbers and special characters
- clean_sentences = pd.Series(sentences).str.replace("[^a-zA-Z]", " ")
-
- # make alphabets lowercase
- clean_sentences = [s.lower() for s in clean_sentences]
- nltk.download('stopwords')
- from nltk.corpus import stopwords
- stop_words = stopwords.words('english')
- # function to remove stopwords
- def remove_stopwords(sen):
- sen_new = " ".join([i for i in sen if i not in stop_words])
- return sen_new
- # remove stopwords from the sentences
- clean_sentences = [remove_stopwords(r.split()) for r in clean_sentences]
7、句子的向量表示
- # Extract word vectors
- word_embeddings = {}
- f = open('glove.6B.100d.txt', encoding='utf-8')
- for line in f:
- values = line.split()
- word = values[0]
- coefs = np.asarray(values[1:], dtype='float32')
- word_embeddings[word] = coefs
- f.close()
- sentence_vectors = []
- for i in clean_sentences:
- if len(i) != 0:
- v = sum([word_embeddings.get(w, np.zeros((100,))) for w in i.split()])/(len(i.split())+0.001)
- else:
- v = np.zeros((100,))
- sentence_vectors.append(v)
8、创建相似矩阵
为了找出句子之间的相似点,我们将使用余弦相似法来解决这个问题。让我们为这个任务创建一个空的相似性矩阵,并用句子的余弦相似性填充它。
- # similarity matrix
- sim_mat = np.zeros([len(sentences), len(sentences)])
- from sklearn.metrics.pairwise import cosine_similarity
- for i in range(len(sentences)):
- for j in range(len(sentences)):
- if i != j:
- sim_mat[i][j] = cosine_similarity(sentence_vectors[i].reshape(1,100), sentence_vectors[j].reshape(1,100))[0,0]
9、实验TextRank算法
在这里我们将相似矩阵sim_mat转换为图形。图中的节点表示句子,边表示句子之间的相似度得分。在这个图中,我们将使用PageRank算法得到句子的排名。
- import networkx as nx
- nx_graph = nx.from_numpy_array(sim_mat)
- scores = nx.pagerank(nx_graph)
-
- #Summary Extraction
- ranked_sentences = sorted(((scores[i],s) for i,s in enumerate(sentences)), reverse=True)
- # Extract top 10 sentences as the summary
- for i in range(10):
- print(ranked_sentences[i][1])
输出结果
- When I'm on the courts or when I'm on the court playing, I'm a competitor and I want to beat every single person
- whether they're in the locker room or across the net.So I'm not the one to strike up a conversation about the
- weather and know that in the next few minutes I have to go and try to win a tennis match.
-
- Major players feel that a big event in late November combined with one in January before the Australian Open will
- mean too much tennis and too little rest.
-
- Speaking at the Swiss Indoors tournament where he will play in Sundays final against Romanian qualifier Marius
- Copil, the world number three said that given the impossibly short time frame to make a decision, he opted out of
- any commitment.
-
- "I felt like the best weeks that I had to get to know players when I was playing were the Fed Cup weeks or the
- Olympic weeks, not necessarily during the tournaments.
- Currently in ninth place, Nishikori with a win could move to within 125 points of the cut for the eight-man event
- in London next month.
- He used his first break point to close out the first set before going up 3-0 in the second and wrapping up the
- win on his first match point.
- The Spaniard broke Anderson twice in the second but didn't get another chance on the South African's serve in the
- final set.
- "We also had the impression that at this stage it might be better to play matches than to train.
-
- The competition is set to feature 18 countries in the November 18-24 finals in Madrid next year, and will replace
- the classic home-and-away ties played four times per year for decades.
-
- Federer said earlier this month in Shanghai in that his chances of playing the Davis Cup were all but non-existent.
具体代码和数据参看github:https://github.com/prateekjoshi565/textrank_text_summarization
结语:
更多机器学习算法的学习欢迎关注我们。对机器学习感兴趣的同学欢迎大家转发&转载本公众号文章,让更多学习机器学习的伙伴加入公众号《python练手项目实战》,在实战中成长。
Copyright © 2003-2013 www.wpsshop.cn 版权所有,并保留所有权利。