/WebGraph

Crawls the web and builds a Neo4J graph of pages and words.

Primary LanguageClojure

WebGraph

This tool crawls a website and generates a graph of pages and links. Additionally, a node is added for each word and edges are added from each word to the pages they are a part of.

Installation

Only Leiningen is required for building. Install Leiningen, then execute the following from the webGraph directory:

webGraph$ lein deps
webGraph$ lein compile

To actually do anything useful, you'll need access to a running Neo4J server. You can install this to your local machine and start an instance:

neo4j-community-1.6$ ./bin/neo4j start

Usage

Use the --help command line option to see the possible arguments:

webGraph$ lein run --help
Usage:

Switches               Default    Desc                                    
--------               -------    ----                                    
-h, --no-help, --help  false      Show this help message                  
--host                 localhost  Neo4J server host                       
--port                 7474       Neo4J server port                       
-s, --seed                        Fully qualified URL for the crawl       
-p, --max-pages        10         Maximum number of pages to crawl        
-d, --max-depth        5          Maximum crawl depth                     
-n, --politeness       1000       Time to wait between page requests (ms) 
-j, --threads          1          Number of crawler threads 

To crawl a domain specify the fully qualified domain name as the seed:

webGraph$ lein run --seed http://www.milk.com/

The crawler will follow links outside of the given domain.

Future Work

  • There needs to be a way to query the database outside of Neo4J's web interface. Either a command-line query tool or a custom web interface.
  • The web crawler library that is used (crawler4j) does a poor job at extracting plain-text from the HTML. Some words get squished together. This can be fixed by taking the raw HTML from the crawler and running it through something like jsoup's Jsoup.parse(html).text().
  • Currently, each distinct word from a crawled page requires either two or three calls to the server. This is very slow, but it can be sped up by using the batch API.
  • To run PageRank on the graph, the server must be taken off-line so a custom Gremlin distribution can access it directly. Maybe there's a better way?

License

Copyright (C) 2012 James Marble and John Russell

Distributed under the Eclipse Public License, the same as Clojure.