Search Engine Details

The XperienCentral search engine is based on the popular open source search engine Lucene. The Lucene core code is used for the search service, querying and indexing. A lot of additional functionality has been added to deal with XperienCentral-specific content, structure, security and architecture. For more information on Apache Lucene see http://lucene.apache.org/

In This Topic

Overview

The following shows a schematic overview of the XperienCentral search engine:

Filling the index with documents is done in three steps: crawling (1) + (2), parsing (3) and indexing (4). Each of these steps will be covered in detail in the following sections. Retrieving search results from the index is done from a client such as a page in XperienCentral with a search element (5), or from the Setup Tool, or from a command line client such as the george-control (Linux) or george-client.bat (Windows).

Back to top

Crawling

Retrieving the documents starts with retrieving the URLs of all documents and pages. GX WebManager provides links to all items that should be indexed on one page: the indexer page. This page contains references to:

• Pages of all web initiatives
• Media items created in the last 5 days
• Documents (uploaded at a Download element or placed in the Object Manager)
• Special content types (such as government papers)

The URL of the indexer page is configured in the properties.txt file (parameter: ‘metaUrl’) and is the default start page in the Setup tool. Note that the setting in the properties.txt file is leading.

The search engine crawler retrieves all the URLs from the indexer page and creates requests for each of the documents. Requests are always sent to the front-end in order to benefit from existing caching and to take authorization and personalization into account. With a default configuration the crawler only indexes links from the indexer page and it does not follow links in the retrieved documents. This means the default index depth is 1. An index depth of 1 ensures that the indexing process is efficient and that no outbound links (links outside the web initiative) are found and indexed.

A request for a document is not only a direct request for the document - additional meta information is requested as well. This extra meta information is provided by the indexer page by requesting the indexer page with an additional ‘document=’ parameter. For example to index the homepage on local GX WebManager installation the crawler requests the URL http://localhost:8080/web/webmanager?id=39016&document= http%3A%2F%2F127.0.0.1%3A8080%2Fweb%2Fshow%2Fid%3D26111. The underlined value is the URL-encoded URL of the homepage. When this URL is requested a small XML result is returned which looks like this:

<document>
  <langid>42</langid>
  <contenttype>page</contenttype>
  <date>2007-12-24</date>
<webid>26098</webid>
<pagepath>p26111</pagepath>
  <pagepath_00_name>Home</pagepath_00_name>
  <pagepath_00_url>http://127.0.0.1:8080/web/Home.htm</pagepath_00_url>
</document>

Back to top

Parsing

All retrieved documents are parsed before they are stored in the index. All documents are converted to a plain text format. Office (Word, Excel) and PDF documents are converted with external programs. These programs are usually executable files and they are configured in <searchengine-directory>/conf/properties.txt. The mapping between document type and converter is configured in <searchengine-directory>/conf/parser.txt. The mapping can be based on both file extension and content-type, which is retrieved from the HTTP header. For more information about the parser.txt file see *** chapter 4.3.

Back to top

Indexing

The index is the storage of the indexed information. In fact it is a small database that is tuned for fast information retrieval. It is also similar to a database because it contains a set of documents (one indexed document is one document in the database) and every document has a set of fields. Notice the similarity between tables and table fields.

Before documents are stored in the index they go through three processes. The first is the normalizing process. In this process characters are converted to normal and lowercase equivalents. For example ‘ç’ is converted to ‘c’, and ‘C’ is converted to ‘c’. By doing this a user will get search results when searching for ‘barcelona’ instead of ‘Barçelona’.

The second process, tokenizing, breaks up words and sentences in different so called ‘tokens’. These tokens are counted and the amount of tokens inside and outside the document is stored. The number of tokens is one of the important factors for relevance. For example when a document contains 9 times the token ‘car’, and the rest of the website contains only one other ‘car’ token then this document is highly relevant when searching for ‘car’ or ‘cars’.

The third process is storing all the information in the index. Storing information is done by creating an entry for every document and filling all the fields of a document. Every field has a ‘relevance’ factor that indicates how important the field is for the document relevance. Some of these factors can be set by changing the corresponding parameters in the properties.txt file. The default settings are:

factor.title=10
factor.description=5
factor.keyword=10
factor.location=1

The title and the keyword are the most important fields.

Back to top