arex

node.js实现自动提取文章正文，标题，发布日期。自动生成文章摘要.

Node版本

建议v6.17.0

Http 服务

运行

node server.js

客户端链接

curl -X POST -d '{"url":"https://export.shobserver.com/baijiahao/html/411796.html","size":100,"smooth":false}' http://localhost:3824|jq -r .

#安装

npm install arex

#使用例子:

var arex = require('arex');
//example 1, 给定网址自动抓取，提取正文，生成摘要
arex.get_article('http://finance.sina.com.cn/consume/puguangtai/2016-03-15/doc-ifxqhmve9227502.shtml',120,(err,result)=>{
                //120: 摘要长度为120，如果不需要生成摘要此参数传入false.
		//result: {"title":"...","content":"....", "summary":"...", "pubdate":"..."}
		console.log(result['content']);
});

//example 2, 给html内容，提取正文，生成摘要
result = arex.get_article_sync('<html.........</html>',120);//result: {"title":"...","content":"....", "summary":"...", "pubdate":"..."}

//example 3, 给html内容，生成摘要
//summarize(content, exptd_len=120, shingle=false, min=150, max=350, filter=[], title)
//shingle的意义: 以摘要长度的句子组合为单位计算权重，shingle为false则以自然句为单位计算权重, filter是过滤规则，符合规则的段落都会被过滤不作为摘要
var summary = arex.summarize('<html>.......</html>', 120, true);
var summary = arex.summarize('<html>.......</html>', 0.04, true, 100, 300);//摘要长度比例 4%, 最短 100, 最长 300

#测试

##获取源码

git clone https://github.com/ahkimkoo/arex.git

##测试某个网页的抽取

cd arex
npm install
node test/test.js http://finance.sina.com.cn/consume/puguangtai/2016-03-15/doc-ifxqhmve9227502.shtml 120

120表示期望文摘的长度

##算法说明

正文抽取: 基于行块密度分布来抽取正文，每个行块由若干自然段落组成。
标题抽取: 分别从正文附近抽取h1标签，从title标签取值，取最可能是标题的那一个。
发布日期抽取: 用正则表达式抽取正文附近的日期。（有误差）。
自动文摘: sentense rank算法，参照pagerank算法的实现，可以指定期望的文摘长度。优化点：加入了神经网络模型判断一句话是否适合作为摘要。

arex

node.js article extractor, automatic summarization.

#Install

npm install arex

#Usage:

var arex = require('arex');
//example 1
arex.get_article('http://finance.sina.com.cn/consume/puguangtai/2016-03-15/doc-ifxqhmve9227502.shtml',120,(err,result)=>{
                //120: summary limited, if you do not need summary set it to false.
		//result: {"title":"...","content":"....", "summary":"...", "pubdate":"..."}
		console.log(result['content']);
});

//example 2
result = arex.get_article_sync('<html.........</html>',120);//result: {"title":"...","content":"....", "summary":"...", "pubdate":"..."}

//example 3
//summarize(content, exptd_len=120, shingle=false, min=150, max=350, filter=[], title)
var summary = arex.summarize('<html>.......</html>', 120, true);
var summary = arex.summarize('<html>.......</html>', 0.04, true, 100, 300);//summary ratio 4%, min length 100, max length 300

#Test

##get source

git clone https://github.com/ahkimkoo/arex.git

##test link

cd arex
npm install
node test/test.js http://finance.sina.com.cn/consume/puguangtai/2016-03-15/doc-ifxqhmve9227502.shtml 120

##About algorithm

article extractor: based density of article blocks， a bock consists of a number of natual lines.
title extracor: h1 tag or title tag, choose the best one.
pubdate extractor: regex extraction nearby the begging or article.
summarizer: based sentense rank, similar pagerank. Optimization: neural network model to determine whether a sentence is suitable as a summary.