Discussion Extraction API

The Discussion API automatically structures and extracts entire threads or lists of reviews/comments from most discussion pages, forums, and similarly structured web pages.

Request

To use the Discussion Thread API, perform a HTTP GET request on the following endpoint:

https://api.diffbot.com/v3/discussion

Provide the following arguments:

ArgumentDescription
token
Developer token
url
Web page URL of the discussion to process (URL encoded)
Optional arguments
fields
Used to specify optional fields to be returned by the Discussion API. See the Fields section below.
timeout
Sets a value in milliseconds to wait for the retrieval/fetch of content from the requested URL. The default timeout for the third-party response is 30 seconds (30000).
callback
Use for jsonp requests. Needed for cross-domain ajax.
maxPages
Set the maximum number of pages in a thread to automatically concatenate in a single response. Default = 1 (no concatenation). Set maxPages=all to retrieve all pages of a thread regardless of length. Each individual page will count as a separate API call.

Response

The Discussion API returns data in JSON format.

Each V3 response includes a request object (which returns request-specific metadata), and an objects array, which will include the extracted information for all objects on a submitted page. The Discussion API returns all post data in a single object within the objects array.

Within the Article and Product APIs (to extract comments or review data), discussion data will be returned within the nested discussion object.

The Discussion API objects / discussion response will include the following fields:

FieldDescription
type
Type of object (always discussion).
pageUrl
URL of submitted page / page from which the discussion is extracted.
resolvedPageUrl
Returned if the pageUrl redirects to another URL.
title
Title of the discussion.
numPosts
Number of individual posts in the thread.
posts
Array of individual posts.
type
Type of element (always post).
id
ID of the individual post. The first post of a thread will have an ID of 0.
parentId
ID of the parent, if the post is a reply or response.
text
Full text of the extracted post.
html
Diffbot-normalized HTML of the extracted post. Please see the HTML Specification for a breakdown of elements and attributes returned.
tags
If the post is long enough, an array of tags generated from its specific content.
humanLanguage
Spoken/human language of the post, using two-letter ISO 639-1 nomenclature.
images
If any images are detected within post content, they will be returned in a separate array. Individual array fields are the same as the Article API's images array.
date
Date of post, normalized in most cases to RFC 1123 (HTTP/1.1).
author
Name/username of the post author.
authorUrl
URL of the author profile page, if available.
pageUrl
URL of the page on which the post was found.
diffbotUri
Internal ID used for indexing.
tags
Array of tags/entities as generated from analysis of all extracted posts and cross-referenced with DBpedia and other data sources.
participants
Number of unique participants in the discussion thread or comments.
numPages
Number of pages in the thread concatenated to form the posts response. Use maxPages to define how many pages to concatenate. More on automatic concatenation.
nextPage
If discussion spans multiple pages, nextPage will return the subsequent page URL.
nextPages
Array of all page URLs concatenated in a multipage discussion. More on automatic concatenation.
provider
Discussion service provider (e.g., Disqus, Facebook), if known.
humanLanguage
Spoken/human language of the discussion / comment thread, using two-letter ISO 639-1 nomenclature.
rssUrl
URL of the discussion's RSS feed, if available.
diffbotUri
Unique object ID. The diffbotUri is generated from the values of various Discussion fields and uniquely identifies the object. This can be used for deduplication.
Optional fields, available using fields= argument
sentiment
Returns a sentiment score of each individual post, a value ranging from -1.0 (very negative) to 1.0 (very positive). For English-language posts only.
links
Returns a top-level object (links) containing all hyperlinks found on the page.
meta
Returns a top-level object (meta) containing the full contents of page meta tags, including sub-arrays for OpenGraph tags, Twitter Card metadata, schema.org microdata, and -- if available -- oEmbed metadata.
querystring
Returns any key/value pairs present in the URL querystring. Items without a discrete value will be returned as true.
breadcrumb
Returns a top-level array (breadcrumb) of URLs and link text from page breadcrumbs.

Example Response

The following response shows the extracted contents from this Hacker News discussion thread:

{
  "request": {
    "pageUrl": "https://news.ycombinator.com/item?id=5608988",
    "api": "discussion",
    "version": 3
  },
  "objects": [
    {
      "numPages": 1,
      "humanLanguage": "en",
      "confidence": 0.34803802321952554,
      "diffbotUri": "discussion|3|-110040828",
      "pageUrl": "https://news.ycombinator.com/item?id=5608988",
      "numPosts": 1,
      "type": "discussion",
      "title": "Show HN: Analysis of 2.5 Years of Frontpage Articles",
      "posts": [
        {
          "humanLanguage": "en",
          "author": "miket",
          "authorUrl": "https://news.ycombinator.com/user?id=miket",
          "diffbotUri": "post|3|1167619016",
          "html": "<table> <tbody><tr><td><table> <tbody><tr><td> <a href=\"https://news.ycombinator.com/user?id=johncoogan\">johncoogan</a> <a href=\"https://news.ycombinator.com/item?id=5609164\">1132 days ago</a>   <br> Huge fan of DiffBot and awesome projects like this. Really cool analysis, thanks for posting.<p>Would be possible for you to post / send me the original data? I have been very interested in working on more longitudinal analysis using DiffBot data and this seems like a fun and interesting place to start. I'm happy to open-source / clearly attribute DiffBot's contribution to whatever I find / hack together, and would feel a lot more comfortable about integrating DiffBot into larger projects in the future.</p><p>Please email me (in my profile) if this is a possibility. Thanks!</p> <p> -----  </p></td></tr> </tbody></table></td></tr> <tr><td><table> <tbody><tr><td> <a href=\"https://news.ycombinator.com/user?id=tswaterman\">tswaterman</a> <a href=\"https://news.ycombinator.com/item?id=5609252\">1132 days ago</a>   <br> Great idea! We'd be happy to share/help. If more people are interested, we'll figure out a good way to distribute the dataset generally. But in fact, you can extract the same data set, and add whatever other smart things you want along with it, using the Diffbot APIs. Everything we did to get this information is explained on our blog at<pre><code>    http://blog.diffbot.com/diffbots-hackernews-trend-analyzer/\n</code></pre> Sounds like you're already using the Diffbot service, but for anyone who's not, they can sign up for a free access token on our 'pricing' page at <a href=\"http://www.diffbot.com/\">http://www.diffbot.com/</a> It's a few hundred thousand pages you'd need to analyze to get this, which doesn't quite fit under the free plan. You might not want to analyze as many years worth of stuff as we did for this demo, though.<p>All the pieces and services we used for this, including all the text extraction, topic detection, and crawling, are available to any user.</p><p>Have fun with it, and keep us informed about whatever cool stuff you build with it, and of course tell us about any features or capabilities you wish Diffbot can provide.</p> <p> -----  </p></td></tr> </tbody></table></td></tr> <tr><td><table> <tbody><tr><td> <a href=\"https://news.ycombinator.com/user?id=tliou\">tliou</a> <a href=\"https://news.ycombinator.com/item?id=5609057\">1132 days ago</a>   <br> Had to figure out how to use it ... but interesting once you do! Android vs IPhone on Hackernews frontpage shows spike in iphone on launch dates, but mediocre to no activity for android. is it because android is less interesting and not as innovative? or not as fun to talk/read about?<p><a href=\"http://diffbot.com/robotlab/hackernews/#type=tags&item=IPhone&item=Android%20(operating%20system)&item=\">http://diffbot.com/robotlab/hackernews/#type=tags&amp;item=I...</a></p> <p> -----  </p></td></tr> </tbody></table></td></tr> <tr><td><table> <tbody><tr><td> <a href=\"https://news.ycombinator.com/user?id=mayank\">mayank</a> <a href=\"https://news.ycombinator.com/item?id=5609116\">1132 days ago</a>   <br> Funny, I just built a HN article catcher that uses Diffbot to collect and classify submissions from the /new page [1]. I've been a Diffbot fan for quite a while now (although their entity recognition/tag classifier needs a bit of work as you can see from the categorization on my catcher page below).<p>[1] <a href=\"http://lahiri.me/more\">http://lahiri.me/more</a></p><p>I should add that their API is fantastic, and far better than using BeautifulSoup/NLTK for extracting textual content from webpages.</p> <p> -----  </p></td></tr> </tbody></table></td></tr> <tr><td><table> <tbody><tr><td> <a href=\"https://news.ycombinator.com/user?id=tswaterman\">tswaterman</a> <a href=\"https://news.ycombinator.com/item?id=5609278\">1132 days ago</a>   <br> Cool! How many articles, or what time period, did you use for this? It looks like you're using only a subset of the topic tags -- did you make a list of 'interesting stuff' to filter against? <p> -----  </p></td></tr> </tbody></table></td></tr> <tr><td><table> <tbody><tr><td> <a href=\"https://news.ycombinator.com/user?id=mayank\">mayank</a> <a href=\"https://news.ycombinator.com/item?id=5609731\">1132 days ago</a>   <br> It's been running for about a week I think, and I'm just taking the top 80 or so tags by article count. Glad you like it :) <p> -----  </p></td></tr> </tbody></table></td></tr> <tr><td><table> <tbody><tr><td> <a href=\"https://news.ycombinator.com/user?id=minimax\">minimax</a> <a href=\"https://news.ycombinator.com/item?id=5609100\">1132 days ago</a>   <br> Neat! Wish I could select by just domain name (i.e. just nytimes.com rather than dozen or so whatever.nytimes.com subdomains). <p> -----  </p></td></tr> </tbody></table></td></tr> </tbody></table>",
          "pageUrl": "https://news.ycombinator.com/item?id=5608988",
          "id": 0,
          "text": "johncoogan 1132 days ago\nHuge fan of DiffBot and awesome projects like this. Really cool analysis, thanks for posting.\nWould be possible for you to post / send me the original data? I have been very interested in working on more longitudinal analysis using DiffBot data and this seems like a fun and interesting place to start. I'm happy to open-source / clearly attribute DiffBot's contribution to whatever I find / hack together, and would feel a lot more comfortable about integrating DiffBot into larger projects in the future.\nPlease email me (in my profile) if this is a possibility. Thanks!\n\n\n-----\n\n\n\n\ntswaterman 1132 days ago\nGreat idea! We'd be happy to share/help. If more people are interested, we'll figure out a good way to distribute the dataset generally. But in fact, you can extract the same data set, and add whatever other smart things you want along with it, using the Diffbot APIs. Everything we did to get this information is explained on our blog at\nhttp://blog.diffbot.com/diffbots-hackernews-trend-analyzer/ Sounds like you're already using the Diffbot service, but for anyone who's not, they can sign up for a free access token on our 'pricing' page at \nhttp://www.diffbot.com/ It's a few hundred thousand pages you'd need to analyze to get this, which doesn't quite fit under the free plan. You might not want to analyze as many years worth of stuff as we did for this demo, though.\nAll the pieces and services we used for this, including all the text extraction, topic detection, and crawling, are available to any user.\nHave fun with it, and keep us informed about whatever cool stuff you build with it, and of course tell us about any features or capabilities you wish Diffbot can provide.\n\n\n-----\n\n\n\n\ntliou 1132 days ago\nHad to figure out how to use it ... but interesting once you do! Android vs IPhone on Hackernews frontpage shows spike in iphone on launch dates, but mediocre to no activity for android. is it because android is less interesting and not as innovative? or not as fun to talk/read about?\nhttp://diffbot.com/robotlab/hackernews/#type=tags&item=I...\n\n\n-----\n\n\n\n\nmayank 1132 days ago\nFunny, I just built a HN article catcher that uses Diffbot to collect and classify submissions from the /new page [1]. I've been a Diffbot fan for quite a while now (although their entity recognition/tag classifier needs a bit of work as you can see from the categorization on my catcher page below).[1] http://lahiri.me/more\nI should add that their API is fantastic, and far better than using BeautifulSoup/NLTK for extracting textual content from webpages.\n\n\n-----\n\n\n\n\ntswaterman 1132 days ago\nCool! How many articles, or what time period, did you use for this? It looks like you're using only a subset of the topic tags -- did you make a list of 'interesting stuff' to filter against?\n\n\n-----\n\n\n\n\nmayank 1132 days ago\nIt's been running for about a week I think, and I'm just taking the top 80 or so tags by article count. Glad you like it :)\n\n\n-----\n\n\n\n\nminimax 1132 days ago\nNeat! Wish I could select by just domain name (i.e. just nytimes.com rather than dozen or so whatever.nytimes.com subdomains).\n\n\n-----",
          "type": "post",
          "title": "Show HN: Analysis of 2.5 Years of Frontpage Articles"
        }
      ],
      "tags": [
        {
          "score": 0.52,
          "count": 2,
          "label": "Ellipsis",
          "uri": "http://dbpedia.org/resource/Ellipsis",
          "rdfTypes": [
            "http://www.w3.org/2002/07/owl#Thing"
          ]
        }
      ],
      "participants": 1
    }
  ]
}

Authentication and Custom Headers

You can supply Diffbot with basic authentication credentials or custom HTTP headers (see below) to access intranet pages or other sites that require a login.

Basic Authentication

To access pages that require a login/password (using basic access authentication), include the username and password in your url parameter, e.g.: url=http%3A%2F%2FUSERNAME:PASSWORD@www.diffbot.com.

Custom HTTP Headers

You can supply Diffbot APIs with custom HTTP headers that will be passed along when making requests to third-party sites. These can be used to define specific Referer, User-Agent, Cookie or any other values.

Custom headers should be sent as HTTP headers in your request to https://api.diffbot.com, and prepended with X-Forward-.

For instance, to send custom User-Agent, Referer and My-Custom-Header header values:

Desired HeaderSend to api.diffbot.com
User-Agent:DiffbotX-Forward-User-Agent:Diffbot
Referer:diffbot.comX-Forward-Referer:diffbot.com
My-Custom-Header:CustomValueX-Forward-My-Custom-Header:CustomValue

Custom Javascript

This functionality is currently in beta.

Using the X-Evaluate custom header (sent as X-Forward-X-Evaluate), you can inject your own Javascript code into web pages. Custom Javascript will be executed once the DOM has loaded.

Custom Javascript should be provided as a text string and contained in its own function. You must also include the special functions start() and end() to indicate the beginning and end of your custom script. Once end() fires, the updated document will be processed by your chosen extraction API.

It's recommended that your end() function be offset using setTimeout (see JavaScript Timing Events) in order to accommodate your primary function processing. Additionally, if your custom Javascript requires access to Ajax-delivered content, it may be necessary to offset your entire function using setTimeout in order to delay the start of your processing.

The following sample X-Evaluate header waits one-half second after the DOM has loaded, enacts a click on the a.loadMore element, then waits 800 milliseconds before signaling the end():

function() {
    start();
    setTimeout(function() {
        var loadMoreNode = document.querySelector('a.loadMore');
        if (loadMoreNode != null) {
            loadMoreNode.click();
            setTimeout(function() {
                end();
            }, 800);
        } else {
            end();
        }
    }, 500);
}

Delivered as a string value as a custom header:

"X-Forward-X-Evaluate": "function() {start();setTimeout(function(){var loadMoreNode=document.querySelector('a.loadMore');if (loadMoreNode != null) {loadMoreNode.click();setTimeout(function(){end();}, 800);} else {end();}},500);}"

Posting Content

If your content is not publicly available (e.g., behind a firewall), you can POST markup directly to the Discussion API endpoint for analysis:

https://api.diffbot.com/v3/discussion?token=...&url=...

Please note that the url argument is still required, and will be used to resolve any relative links contained in the markup.

Provide the content to analyze as your POST body, and specify the Content-Type header as text/html.