Part 3: Web Scraping with Fess

<<This page is generated by Machine Translation from Japanese. Pull Request is welcome!>>

Last time we showed you how to add site search to your existing site. Fess has various functions, but this time I would like to introduce the Web scraping function.

In recent years, AI and machine learning are popular fields, but there are also fields where data can be used to do various things. There is a lot of information on the Internet in Web pages, and the technology to extract information from it is Web scraping. Gather information in the right way with Web scraping and apply machine learning to create even more value.

Fess has a powerful crawler, so you can extract specified parts from within a web page and save them in an index. By using Fess, it is possible to collect and analyze necessary information only by setting, without creating a Web scraping program.

In this and the next article, I will introduce how to collect information from news sites and create a model for classifying text.


I will consider creating a text classification model using the commentary / example articles of IT Search +. There are categories such as “server / storage” in the explanation / case articles of IT Search +, so which category does it belong to given an arbitrary sentence? Create a classification model that can judge. By changing the question settings, it will be possible to build a system that involves natural language processing, such as judging spam documents and automatically sorting articles as used in news applications.

Then, I will explain how to build a web scraping server using Fess in this problem setting. Please install Fess by referring to the procedure of :doc:`Part 1 </articles/1/document>`__.

Crawling target analysis

In order to do web scraping, you need to understand what is being crawled. If your crawl is like a news site,

  • Article list page: URL includes page number such as p=1 or page=1
  • Article page: Same upper part like …/article/<article name>

You can crawl the two types of pages and save only the latter to collect the information you need. The list page can be determined from the URL by clicking the pagination link in that page.

Next, consider extracting the necessary parts from the article page. Normally crawling with Fess will only index the text below the body tag as a search target, but in Fess you can save the value in the index by specifying the location you want to acquire with XPath. For example, if you want the title tag value, specify //TITLE to extract and save the value (in Fess, the XPath tag name must be specified in uppercase).

Creating a crawl configuration

Once the crawl target and extraction location are determined, create the crawl settings on the management screen. The web crawl settings this time are as follows.

Item Value
Name IT Search + Explanations/Case Studies
URLs to be crawled**/[0-9]+
URL to search*/ [0-9] +
Configuration parameter
field.xpath.article_category=//LI[@class=’genre_tag sp-none’]/SPAN
Depth 1
Interval 3000 milliseconds

First, check the list page to set to start crawl. When you check the second page of the list page, since the URL is specified in the format including the page number like all _2, set each list page to the URL. If there is a link from the article page to the article page, it will be crawled one after another unless the depth is specified, so by limiting the depth to 1, it will be linked directly from the list page group specified by the URL It is intended only for the article pages that are present. This setting is for crawling the articles on the list page up to the 10th page.

By specifying the URL to be crawled, it is restricted to crawl only the list page and the article page. Also, since it is not necessary to save the list page, it is specified to save only the article page with the URL to be searched.

In this setup, get the category information from the article page, save it in the article _category field, and save only the article body in the article _body field. In Fess, by writing field.xpath. [Field name] = [XPath] in the configuration parameter, the value extracted by the specified field name can be saved in the index.

Depending on the site, the canonical URL may be specified on the list page, and Fess will process it with the canonical URL. Therefore, canonical processing is disabled by leaving config.html.canonical.xpath empty in the configuration parameters.

Change of expiration date

Fess sets an expiration date for data indexed during crawl. By default, three days are set, so after three days the collected data will be deleted. To prevent deletion, set the value of “Delete previous documents” to -1 days in the settings of the “General” crawler from the administration screen.

Check crawl

Once you have configured the settings up to this point, start the Default Crawler job in the “scheduler” and perform a crawl. After the crawl job finishes, check the saved values. Fess sets the added fields so that they cannot be searched or displayed by default. Change the following property values ​​in app / WEB-INF / classes / fess to get the added article _category and article _body this time.

# When writing by JSP
# To include in JSON response

You need to restart Fess after changing fess After restarting, you can confirm that the value has been obtained by calling the JSON API as shown below.

curl -s "localhost:8080/json/?q=*" | \
  jq '.response.result[0] | {article_category: .article_category, article_body: .article_body[0:40]}'
  "article_category": "マーケティング",
  "article_body": "デジタル領域の4テーマについて、1日1テーマで開催された「Fujitsu Ins"

If you can’t get it, check logs/fess-crawler.log to see if crawling is running as expected.


This time, I introduced how to use Fess as a web scraping server. By using this function, it is possible to build an information gathering environment just by setting without writing code for scraping, and focus on the analysis and machine learning tasks that are the original objectives .

Next time, I will introduce how to create a classification model using data collected by Fess.