Developing an Amazon Movie Scraper
#1
I have recently started using XBMC (on a Mac) and found that while the IMDB scraper works well enough, there are many DVDs not on IMDB that are on Amazon.

[Note: While the examples below use a film title of "Soylent Green" I have manually searched IMDB using a browser to confirm other titles are definitely not listed.]

Surprisingly there is no existing Amazon scraper. As part of an effort to make one myself I started off by looking at the existing scrapers to see how they worked, and following on from this I made some initial efforts to convert the current FilmAffinity scraper to use English results rather than Spanish results (you can download a copy here if you are interested http://homepage.mac.com/jelockwood/.Publ...nityen.zip).

While I have not yet got an Amazon scraper even partially working yet, I have found some important information about the format of the various URLs that Amazon uses.

1. Amazon itself normally replaces spaces in Title searches with a plus (+) symbol, however it does seem to also work with a space (or %20).

A search URL like the following entered in a web-browser all work

Code:
http://www.amazon.com/s/ref=nb_ss_d?url=search-alias=dvd&field-keywords=soylent+green&x=0&y=0
Code:
http://www.amazon.com/s/ref=nb_ss_d?url=search-alias=dvd&field-keywords=soylent green&x=0&y=0
Code:
http://www.amazon.com/s/ref=nb_ss_d?url=search-alias=dvd&field-keywords=soylent%20green&x=0&y=0

and indeed also the slightly shorter

Code:
http://www.amazon.com/s/ref=nb_ss_d?url=search-alias=dvd&field-keywords=soylent%20green

2. The URL of a result is normally a rather messy and complicated format like this

Code:
http://www.amazon.com/Soylent-Green-John-Barclay/dp/B0016I0AJG/ref=sr_1_1?ie=UTF8&s=dvd&qid=1217077050&sr=1-1

as you can see there would appear to be two different ID numbers plus a text field. However I have been able to determine that the following much simpler form of the URL also works.

Code:
http://www.amazon.com/dp/B0016I0AJG/

Therefore we just need to extract the ID number beginning with a B (they all seem to begin with a B).

3. The thumbnail image normally has a URL of the form

Code:
http://ecx.images-amazon.com/images/I/51bU-puSlkL._SL500_AA240_.jpg

and the large image a URL of the form

Code:
http://ecx.images-amazon.com/images/I/51bU-puSlkL._SS500_.jpg

as you can see the ID number is totally different to anything previously used. However I have also found that the following URL produces the same large image and uses the main ID number from the original URL

Code:
http://ecx.images-amazon.com/images/P/B0016I0AJG.01.L.jpg

or the older alternative host name

Code:
http://images.amazon.com/images/P/B0016I0AJG.01.L.jpg

Note these forms of the URL must use a P rather than an I.

Based on all the above, would anyone care to assist by coming up with an initial Scraper by coding up the CreateSearchUrl and GetSearchResults sections? I will then try scraping the info fields.

PS. On a different topic, if one has a VIDEO_TS folder in a folder representing the name of the film one can use this folder name for IMDB scraping, however as mentioned not all the DVDs are listed on IMDB, I can see it should be possible to use an NFO file to provide at least some metadata but I am unsure of the correct naming and placement in this scenario.

e.g. /DVDs/Soylent Green/VIDEO_TS/

What should the NFO file be called and in which of the three possible folders (DVDs, Soylent Green, or VIDEO_TS) should it be placed?
Reply
#2
There are more information on the wiki about scrapers.
Reply
#3
blittan Wrote:There are more information on the wiki about scrapers.

Been there and read it already.

It may have enough information for people already expert in writing regex but for the majority of us it still does not help. What is really needed is some examples in the Wiki.

What I would suggest would be that for the CreateSearchUrl it should include an example Url and describe how it is created.

Likewise, for the GetSearchResults it should show a URL returned as a result of using the CreateSearchUrl and describe how the example regex extracts the needed information.

I am not expecting it to cover every eventuality or source, but it does not have any examples at all. This is why I found it much more helpful to look at an existing scraper (the FilmAffinity one) and compare the xml code with what appears in a web-browser doing the same thing.

Don't get me wrong I think the developers of XBMC have done a great job but like most/all open-source projects (especially Linux people) they assume all the users have the same level of programming expertise as themselves. I am not a full-time programmer but I have found bugs in some open-source projects and submitted successful fixes and I still find the documentation less than desirable.

Hmm, just had a thought, part of the difficulty is that the only way to 'test' a scraper is to use it within XBMC, and you don't get any detailed feedback, it either works or it does not. However I do have a utility to test regex against sample text, this might at least help me test the extracting portion.

Anyway, if any one else is interested, the original Amazon information I posted may assist in a group effort.
Reply
#4
Totally agreed about the documentation. The problem is that it's only ever developers or very experienced users that bother writing any.

I'll see if I can get C-Quel to make some comments in this thread as to what the functions do and hopefully it can be used as a base or example that others can see and get some good use out of.

Cheers,
Jonathan
Always read the XBMC online-manual, FAQ and search the forum before posting.
Do not e-mail XBMC-Team members directly asking for support. Read/follow the forum rules.
For troubleshooting and bug reporting please make sure you read this first.


Image
Reply
#5
afaik c-quel already has a amazon scraper that is almost finished.

i answer all specific questions. but you need to grasp the principles from examples.

Code:
<CreateSearchUrl dest="3">
  <RegExp input="$$1" output="http://www.amazon.com/s/ref=nb_ss_d?url=search-alias=dvd&field-keywords=\1" dest="3">
  <expression noclean ="1"/>
</RegExp>
</CreateSearchUrl>

this is as basic as it gets - we get our url encoded search string passed in $$1, we select it all and append it at the appropriate spot to create the search url.
Reply
#6
jmarshall Wrote:Totally agreed about the documentation. The problem is that it's only ever developers or very experienced users that bother writing any.

I'll see if I can get C-Quel to make some comments in this thread as to what the functions do and hopefully it can be used as a base or example that others can see and get some good use out of.

Cheers,
Jonathan

Just to make it clearer (for others) what I was referring to by examples (since there is a bit in the Wiki). For the example Regexc given, it should show what the resulting URL actually looks like that has been generated from the Scraper and Regexc.

Note to others the relevant Wiki page is http://www.xbmc.org/wiki/?title=Scraper.xml

On that page it shows an example GetSearchResults using JadedVideo but does not show what the full real URL was so you cannot compare before and after. Also it would help if they use the same example source for both the CreateSearchUrl and the GetSearchResults (they use different ones).
Reply
#7
jelockwood Wrote:Been there and read it already.

It may have enough information for people already expert in writing regex but for the majority of us it still does not help. What is really needed is some examples in the Wiki.

What I would suggest would be that for the CreateSearchUrl it should include an example Url and describe how it is created.

Likewise, for the GetSearchResults it should show a URL returned as a result of using the CreateSearchUrl and describe how the example regex extracts the needed information.

I am not expecting it to cover every eventuality or source, but it does not have any examples at all. This is why I found it much more helpful to look at an existing scraper (the FilmAffinity one) and compare the xml code with what appears in a web-browser doing the same thing.

Don't get me wrong I think the developers of XBMC have done a great job but like most/all open-source projects (especially Linux people) they assume all the users have the same level of programming expertise as themselves. I am not a full-time programmer but I have found bugs in some open-source projects and submitted successful fixes and I still find the documentation less than desirable.

Hmm, just had a thought, part of the difficulty is that the only way to 'test' a scraper is to use it within XBMC, and you don't get any detailed feedback, it either works or it does not. However I do have a utility to test regex against sample text, this might at least help me test the extracting portion.

Anyway, if any one else is interested, the original Amazon information I posted may assist in a group effort.
what utility is it that you're using to "test" the scraper?
Reply
#8
there is no such utility available. there was one (/tools/Scrap) but "somebody" forgot to commit parts of the code and consequently it was lost in a hdd crash or something along that. at one point hopefully somebody will redo the utility as it is really very handy :/
however i feel it's a waste of my time personally, i have just this much coding time to put into the project and i much rather use it improving xbmc.

you can use any regular expression evaluator to test the extraction part, i have been using 'the regexp coach' myself (when on win32) or kregexpeditor in linux.
Reply
#9
That "somebody" was me, sorry about this.
Always read the XBMC online-manual, FAQ and search the forum before posting.
Do not e-mail XBMC-Team members directly asking for support. Read/follow the forum rules.
For troubleshooting and bug reporting please make sure you read this first.
Reply
#10
spiff Wrote:there is no such utility available. there was one (/tools/Scrap) but "somebody" forgot to commit parts of the code and consequently it was lost in a hdd crash or something along that. at one point hopefully somebody will redo the utility as it is really very handy :/
however i feel it's a waste of my time personally, i have just this much coding time to put into the project and i much rather use it improving xbmc.

you can use any regular expression evaluator to test the extraction part, i have been using 'the regexp coach' myself (when on win32) or kregexpeditor in linux.

Unfortunately for me I am using Mac OS X and it seems the range of Regexp tools is not as good. I have tried using RegExhibit - pretty much the only one for Mac, but it does not seem to use exactly the same rules (as it appears Scrapers.xml use).

For example, the following genuine search result from Amazon which I believe to be what the scraper needs to search for -

Code:
<div class="productTitle"><a href="http://www.amazon.com/Soylent-Green-John-Barclay/dp/B0016I0AJG/ref=sr_1_1?ie=UTF8&s=dvd&qid=1217775403&sr=1-1"> Soylent Green </a><span class="binding"> ~ John Barclay, Whit Bissell, Jan Bradley, and Chuck Connors</span><span class="binding"> (<span class="format">DVD</span> - 2008)</span></div>

Seems with RegExhibit to need regex like this

Code:
<div class="productTitle"><a href="http://www.amazon.com/.*/dp/(B[a-zA-Z0-9]*)/ref=sr_1_[0-9]*\?ie=UTF8&s=dvd&qid=[0-9]*&sr=1-[0-9]*"> .* </a><span class="binding">?.~ .*</span><span class="binding"> ?\(<span class="format">DVD</span> - .*\)</span></div>

to get the important ID number (in some places [a-zA-Z0-9] does not seem to work), and like this

Code:
<div class="productTitle"><a href="http://www.amazon.com/.*/dp/B[a-zA-Z0-9]*/ref=sr_1_[0-9]*\?ie=UTF8&s=dvd&qid=[0-9]*&sr=1-[0-9]*"> (.*) </a><span class="binding">?.~ .*</span><span class="binding"> ?\(<span class="format">DVD</span> - .*\)</span></div>

to get the movie title.

RegExhibit can be obtained here http://homepage.mac.com/roger_jolly/software/index.html
Reply
#11
jelockwood Wrote:Unfortunately for me I am using Mac OS X and it seems the range of Regexp tools is not as good.

Might be of help......

I have used some tools in the past for checking my regular expressions, many of the ones i found were internet/flash based (and i, assume), platform independant, sorry, i can't remember the names/places, but google will.

Shorty
Reply
#12
remember; the scraper is a xml file so any special chars needs to be xml'ized for the regexp to function properly in the scraper (or rather, for the scraper xml to load correctly in the first place). this is why there are all these &lt; &amp; etc stuffs in the other scrapers (which i assume you use for reference)
Reply
#13
spiff Wrote:remember; the scraper is a xml file so any special chars needs to be xml'ized for the regexp to function properly in the scraper (or rather, for the scraper xml to load correctly in the first place). this is why there are all these &lt; &amp; etc stuffs in the other scrapers (which i assume you use for reference)

Yep, I spotted that on the Wiki and have taken that in to consideration when trying RegExhibit. It was the fact that in some places [a-zA-Z0-9] works and others I have had to use .* that makes it difficult to know if my regex will be right. Also escaping characters like ~ (tilde), a question mark ?, a space character, and ( ) <- not actual regex but real parenthesis is confusing. These are not listed in the Wiki.

While I am at it, here are some questions I feel the Wiki does not adequately answer.

1. I am using folder names as the search criteria. The folder names include the year of the film, e.g. "Soylent Green (1973)". While IMDB works very well with all of that as a search string, Amazon does not like the year being included (with or without parenthesis) if you type that using a web-browser, suggesting it would equally not like it being sent by a scraper.

Does your typical scraper when using folder names like this, include the year when creating a search URL, or does it strip it off, if so how?

2. When a scraper looks at the search results, XBMC displays a list of titles found, but the scraper has to use the ID number to generate the URL to access the selected result. I am not clear from the Wiki where these two different steps are done, and how the results are linked. As you saw from my last post, I have found the relevant html code returned by Amazon and somewhat got regex code that can extract either the ID or the title.

(Oops, I thought I had already posted this reply, but came back later and discovered this message still open for editing in a tab in my web-browser.)
Reply
#14
1) i'm not completely sure but afaik its done by the scraper. if not i will remedy that
2) no you dont have to use the id. you are free to list whatever url you want in the returned results from GetSearchResults. flow is simply;

we call CreateSearchUrl with buffers 1 set to the cleaned title (url encoded) and iirc buffer 2 set to the year - if this is not so that i will fixed. xbmc grabs the url, then calls GetSearchResults with buffer 1 set to the returned html. this function returns a xml list with the obtained results. remember, that's ALL the scraper does - it is translating some general html into a fixed xml format
Reply
#15
The only way to test a scraper is for it to work almost completely. A bit of a catch22.

Despite the very poor documentation, it seems clear enough there are three main sections to a scraper.

1. Create the Search URL
2. Process the results to list the returned movies and let a user choose one
3. For the chosen movie get the meta data fields

I am reasonably clear on CreateSearchUrl. However the documentation for GetSearchResults sucks big time, especially as I now believe it has at least one major error in the documentation. So having sat down and looked at two existing working scrapers (IMDB and FilmAffinity) and what documentation there is in the Wiki I am going to break down the steps as I so far understand them and ask for confirmation and clarification of my understanding.

I will use the FilmAffinity GetSearchResults code as the example here. So firstly here is the exact code in that scraper.

Code:
<GetSearchResults dest="8">
    <RegExp input="$$5" output="&lt;?xml version=&quot;1.0&quot; encoding=&quot;iso-8859-1&quot; standalone=&quot;yes&quot;?&gt;&lt;results&gt;\1&lt;/results&gt;" dest="8">
      <RegExp input="$$1" output="\1" dest="7">
        <expression>&lt;img src="http://www.filmaffinity.com/imgs/movies/full/[0-9]*/([0-9]*).jpg"&gt;</expression>
      </RegExp>
      <RegExp dest="5" input="$$1" output="&lt;entity&gt;&lt;title&gt;\1 (\2)&lt;/title&gt;&lt;url&gt;http://www.filmaffinity.com/en/film$$7.html&lt;/url&gt;&lt;id&gt;$$7&lt;/id&gt;&lt;/entity&gt;">
        <expression noclean="1">&lt;title&gt;([^&lt;]*)\(([0-9]*)\) - FilmAffinity</expression>
      </RegExp>
      <RegExp input="$$1" output="\1" dest="4">
        <expression noclean="1">(&lt;b&gt;&lt;a href="/en/film.*)</expression>
      </RegExp>
      <RegExp dest="5+" input="$$1" output="&lt;entity&gt;&lt;title&gt;\2 (\3)&lt;/title&gt;&lt;url&gt;http://www.filmaffinity.com/en/film\1.html&lt;/url&gt;&lt;id&gt;\1&lt;/id&gt;&lt;/entity&gt;">
        <expression repeat="yes" noclean="1,2">&lt;a href="/en/film([0-9]*).html[^&gt;]*&gt;([^&lt;]*)&lt;/a&gt;[^\(]*\(([0-9]*)</expression>
      </RegExp>
      <expression noclean="1"></expression>
    </RegExp>
  </GetSearchResults>

It seems fairly clear that the first RegExp line outputs the following

Code:
<?xml version="1.0" encoding="iso-8859-1" standalone="yes"?>
<results>
\1
</results>

where \1 is the content of variable 1

It is not clear what the second RegExp is doing other than it seems to be related to the unique ID number of the film on the website, here is what the second one seems to translate to.

Code:
<img src="http://www.filmaffinity.com/imgs/movies/full/[0-9]*/([0-9]*).jpg">

Interestingly, while this URL is valid and loads the film thumbnail, it does not seem to exist in the results returned by FilmAffinity.

The third RegExp is returning

Code:
<entity>
   <title>\1 (\2)</title>
   <url>http://www.filmaffinity.com/es/film$$7.html</url>
   <id>$$7</id>
</entity>

Here we see a case that seems to clearly point to an error in the Wiki. The Wiki suggests

Code:
<entity>
   <title>?</title>
   <url>?</url>
   <url>?</url>
</entity>

and makes no mention of </id>?</id> at all. Note: both the IMDB and FilmAffinity scrapers do use the ID tags.

I would guess that <title>?</title> is the title of the film as returned by the website, <url>?</url> is the URL to access details for that selected film, and <id>?</id> is a unique ID number of that film as used by the website.

I have no idea what purpose the fourth RegExp has, it appears in this case to return

Code:
(<b><a href="/en/film.*)

The fifth RegExp seems almost exactly the same as the third one.

Code:
<entity>
   <title>\2 (\3)</title>
   <url>http://www.filmaffinity.com/es/film\1.html</url>
   <id>\1</id>
</entity>

So while it seems clear that the 3rd and 5th RegExp fill in the returned XML, it is still not clear to me what bit does the actual searching of the results to find the list of films returned by the website.

I would particularly like an explanation of what the second and fourth RegExp bits do.
Reply

Logout Mark Read Team Forum Stats Members Help
Developing an Amazon Movie Scraper1