Probably the most common technique used traditionally to extract data from web pages this is to cook up some regular expressions that match the pieces you want (eg, URL's and link titles). Our screen-scraper software actually started out as an application written in Perl for this very reason. In addition to regular expressions, you might also use some code written in something like Java or Active Server Pages to parse out larger chunks of text. Using raw regular expressions to pull out the data can be a little intimidating to the uninitiated, and can get a bit messy when a script contains a lot of them. At the same time, if you're already familiar with regular expressions, and your scraping project is relatively small, they can be a great solution.
Other techniques for getting the data out can get very sophisticated as algorithms that make use of artificial intelligence and such are applied to the page. Some programs will actually analyze the semantic content of an HTML page, then intelligently pull out the pieces that are of interest. Still other approaches deal with developing "ontologies", or hierarchical vocabularies intended to represent the content domain.
There are a number of companies (including our own) that offer commercial applications specifically intended to do screen-scraping. The applications vary quite a bit, but for medium to large-sized projects they're often a good solution. Each one will have its own learning curve, so you should plan on taking time to learn the ins and outs of a new application. Especially if you plan on doing a fair amount of screen-scrapping it's probably a good idea to at least shop around for a screen-scraping application, as it will likely save you time and money in the long run.
So what's the best approach to data extraction? It really depends on what your needs are, and what resources you have at your disposal. Here are some of the pros and cons of the various approaches, as well as suggestions on when you might use each one:
Raw regular expressions and code
— If you're already familiar with regular expressions and at least one programming language, this can be a quick solution.
— Regular expressions allow for a fair amount of "fuzziness" in the matching such that minor changes to the content will not break them.
— You likely do not need to learn any new languages or tools (again, assuming you're already familiar with regular expressions and a programming language).
— Regular expressions are supported in almost all modern programming languages. Heck, even VBScript has a regular expression engine. It's also nice because the various regular expression implementations do not vary too significantly in their syntax.
— They can be complex for those that do not have a lot of experience with them. Learning regular expressions is not like going from Perl to Java. It's more like going from Perl to XSLT, where you have to wrap your mind around a completely different way of viewing the problem.
— They're often confused to analyze. Take a look through some of the regular expressions people have created to match something as simple as an email address and you'll see what I mean.
— If the content you're trying to match changes (eg, they change the web page by adding a new "font" tag) you'll likely need to update your regular expressions to account for the change.
— The data discovery portion of the process (traversing various web pages to get to the page containing the data you want) will still need to be handled, and can get fairly complex if you need to deal with cookies and such.
When to use this approach: You'll most likely use straight regular expressions in screen-scraping when you have a small job you want to get done quickly. Especially if you already know regular expressions, there's no sense in getting into other tools if all you need to do is pull some news headlines off of a site.
Ontologies and artificial intelligence
— You create it once and it can more or less extract the data from any page within the content domain you're targeting.
— The data model is generally built in. For example, if you're extracting data about cars from web sites the extraction engine already knows what the make, model, and price are, so it can easily map them to existing data structures (eg, insert the data into the correct locations in your database).
— There is relatively little long-term maintenance required. As web sites change you likely will need to do very little to your extraction engine in order to account for the changes.
— It's relatively complex to create and work with such an engine. The level of expertise required to even understand an extraction engine that uses artificial intelligence and ontologies is much higher than what is required to deal with regular expressions.
— These types of engines are expensive to build. There are commercial offerings that will give you the basis for doing this type of data extraction, but you still need to configure them to work with the specific content domain you're targeting.
— You still have to deal with the data discovery portion of the process, which may not fit as well with this approach (meaning you may have to create an entirely separate engine to handle data discovery). Data discovery is the process of crawling web sites such that you arrive at the pages where you want to extract data.
When to use this approach: Typically you'll only get into ontologies and artificial intelligence when you're planning on extracting information from a very large number of sources. It also makes sense to do this when the data you're trying to extract is in a very unstructured format (eg, newspaper classified ads). In cases where the data is very structured (meaning there are clear labels identifying the different data fields), it may make more sense to go with regular expressions or a screen-scraping application.
— Abstracts most of the complicated stuff away. You can do some pretty sophisticated things in most screen-scraping applications without knowing anything about regular expressions, HTTP, or cookies.
— Dramatically reduces the amount of time required to set up a site to be scraped. Once you learn a particular screen-scraping application the amount of time it requires to scrape sites vs. other methods is significantly lowered.
— Support from a commercial company. If you run into trouble while using a commercial screen-scraping application, chances are there are support forums and help lines where you can get assistance.
— The learning curve. Each screen-scraping application has its own way of going about things. This may actually learn a new scripting language in addition to familiarizing yourself with how the core application works.
— A potential cost. Most ready-to-go screen-scraping applications are commercial, so you'll probably be paying in dollars as well as time for this solution.
— A proprietary approach. Any time you use a proprietary application to solve a computing problem (and proprietary is obviously a matter of degree) you're locking yourself into using that approach. This may or may not be a big deal, but you should at least consider how well the application you're using will integrate with other software applications you currently have. For example, once the screen-scraping application has extracted the data how easy is it for you to get to that data from your own code?
When to use this approach: Screen-scraping applications vary widely in their ease-of-use, price, and suitability to tackle a broad range of scenarios. Chances are, though, that if you do not mind paying a bit, you can save yourself a significant amount of time by using one. If you're doing a quick scrape of a single page you can use just about any language with regular expressions. If you want to extract data from hundreds of web sites that are all formatted differently you're probably better off investing in a complex system that uses ontologies and / or artificial intelligence. For just about everything else, though, you may want to consider investing in an application specifically designed for screen-scraping.
As an aside, I thought I should also mention a recent project we've been involved with that has actually required a hybrid approach of two of the aforementioned methods. We're currently working on a project that deals with extracting newspaper classified ads. The data in classifieds is about as unstructured as you can get. For example, in a real estate ad the term "number of bedrooms" can be written about 25 different ways. The data extraction portion of the process is one that lends itself well to an ontologies-based approach, which is what we've done. However, we still had to handle the data discovery portion. We decided to use screen-scraper for that, and it's handling it just great. The basic process is that screen-scraper traverses the various pages of the site, pulling out raw chunks of data that measure the classified ads. These ads then get passed to code we've written that uses ontologies in order to extract out the individual pieces we're after. Once the data has been extracted we then insert it into a database.