Link Relationships Revisited, Part 1

A while ago, I wrote an entry entitled Link Relationships that detailed my ideas for additions to the values for the rev and rel attributes. It went a long way into describing many relationships for several categories including user feedback, quality, accuracy, accessibility, rating and endorsement.

Having received some feedback from many sources regarding the value of all these relationships, the fact that Tantek had implemented an alpha system for Vote Links in Technorati and now that Google, MSN Search and Yahoo have implemented rel="nofollow"; I intend to revisit the idea and explain exactly why I don’t think Technorati’s vote-for, vote-against and vote-abstain; and Google’s nofollow relationships are suitable. Following this, in part 2, I will discuss the major modifications to my original ideas that will both simplify and reduce the number of relationships, yet provide increased semantics in a way that implements the desired functionality of vote links (including nofollow) with many more advantages that are applicable to a wider variety of user agents.

Relationship Semantics

Firstly, I would like to explain a little about the semantics and purpose behind link relationships to indicate the link role. Each relationship should meet one or more of these criteria to indicate:

  • The semantic relationship from one resource to another (eg. Next or Prev in sequence),
  • What the resource is (eg. Contents, Stylesheet, etc.), and/or
  • The purpose of the link (eg. To provide Help for the user).

Additionally the name of the relationship should accurately represent its semantics, and all relationships should make some sense in the context of any number of user agents and for any user — the relationship should be named and designed independent of any particular type of UA. However, this does not mean that every UA should be able to do something sensible with the link; nor that every UA that does, should do the same thing with it.

All of the above is true for all HTML 4 link types, somewhat true for XFN, and for most of the relationships I suggested previously. However, the same is not entirely the case for either the vote-links or nofollow relationships.

In fairness, vote links (unlike nofollow) do express some semantics. ie. A vote-for relationship expresses that the author of document A approves of document B. In this way, a search engine, for example, may make use of the relationship during page-rank calculations and another UA may simply gather and list all such links for the user to select from easily. However, the relationships do not indicate any semantic relationship, what the resource is, or the purpose of the link.

In the Technorati Vote Links Wiki, the reasons given for using the yes/no/abstain model are given. However, although they do raise some good points that I will be incorporating into my link relationship suggestions later on, they fail to make a solid case for using them specifically. The general reasoning given is that this model has served well for politics, committees and user rating systems, such as that used on eBay, and will thus, due to the democratic nature of the web, serve well for page rank calculations; and that they were designed to be easy-to-remember to manually type in a hand-coding environment, and easy for authoring tools to implement.

Given that, however, the main reason for why vote-links are unsuitable is not just for lack of useful semantics, it is that they provide too much potential for abuse. Mark Pilgrim has given his reasons for why vote-for is useless and why vote-against is harmful.

  • rel=”vote-for” –> increase PageRank (the default, this is what all links do now)
  • rel=”vote-abstain” –> ignore for PageRank (like a Javascript link that Google can’t follow, or a hypothetical link-level NOFOLLOW meta tag)
  • rel=”vote-against” –> decrease PageRank

He also states that vote-abstain is the only useful value and compares it with the idea for a nofollow value which he approves of, and which Google was soon to implement.

Because vote-abstain is essentially equivalent to nofollow in that it states that the link should not be counted for page-rank calculations, it is harmful for the same reasons given below for nofollow, though it is better because its name more accurately represents its meaning.

The nofollow Relationship

The nofollow relationship has no semantics relating to the documents in any way whatsoever – its name simply states, in no uncertain terms, that a user agent should not follow the link; however, its intended meaning is that the link should not be counted towards page-rank (ie. This link is abstaining from voting). Compare this with the similarly named, yet semantically different nofollow value for the Robot’s Meta Tag. For the robot’s meta tag, this value means that no links within the document should be followed or indexed in any way; whereas the nofollow relationship, according to google’s announcement, may be followed, and indexed but shall not recieve any additional credit.

From now on, when Google sees the attribute (rel="nofollow") on hyperlinks, those links won’t get any credit when we rank websites in our search results.

Thus, as stated earlier, vote-abstain is a better name; though both are considered harmful for the following reasons.

nofollow makes no sense at all in any user agent, other than a search engine that uses the page-rank (or equivalent) algorithm; and is, in fact, a case where the relationship is not named or designed to be independent of a particular type of UA.

Additionally, this relationship opens up the potential for abuse. Much like an untrusted content element (and other similar suggestions) would be abused by marking the entire content of every page; and due to the complete lack of semantics, uneducated authors (who make up a large proportion of the web community) may begin to include the relationship within all external links so as not to help increase the page rank for any external site they link to, and may even do so as a result of company policy. Of course its appearance in company policies may seem like a wild assumption, but just consider other such linking policies that some companies have introduced. Therefore I am stating it as a possibility, not a certainty.

Steven Garrity has raised another interesting point in his Thoughts on Weblog Comment Spam Prevention:

Tagging all links in comments left by weblog readers means that none of these links will contribute to the great hive mind that is Google PageRank. There are loads of great and valuable links in weblog comments.

He then continues to say that it is likely that some software may not use the value for links added by registered users; but most comments left on blogs are not from registered users at all. Thus, this will, in fact, inadvertently harm legitimate users, and indeed search results, by losing the benefit of mass communication.

The real question, however, which everyone will be asking is: will this stop comment spam? On the surface, it seems like a brilliant idea. It attacks the very reason behind a lot of comment spam which is designed to increase page rank for spammer’s web sites. However, to answer this question, one really needs to look at the current methods used to fulfil the same function. Not all spam prevention methods need to be looked at, just those that prevent a search engine from indexing the link and counting it towards page rank. These methods include:

  • A link redirector, such as that used by Blogger.
  • Plain text-links.
  • Moderation and deleting spammer comments.

I’ve have only listed these techniques, and avoided others like spam filters, black lists, etc. because, while those are designed to stop spam before they even get published, these are designed (like nofollow) to reduce the benefit of the link for the spammer in the event that the comment is published.

It’s not hard to see that while each of these methods have been and will continue to be used in many blogs, none of them have succeeded in stopping the flood of spam. Again, Mark Pilgrim has explained this phenomenon in regard to the use of plain text links by saying that spammers don’t care.

Spammers have it in their heads now that weblog comments are a vector to exploit.  They don’t look at individual results and tweak their software to stop bothering individuals.  They write generic software that works with millions of sites and goes after them en masse.  So you would end up with just as much spam, it would just be displayed with unlinked URLs.

That point does not only apply to plain text links, but to any method with the same effect. Spammers don’t read blogs, they don’t know who implements what features and who doesn’t; and they don’t know which spam attempts will be successful, yet they continue to automate and spam anyone and everyone they can, in the hope that some small percentage will get through.

For this method to be successful, it would require that every single web site, that provides user feedback mechanisms, to implement it. If only some do, spammers won’t care. If most do, but some don’t, spammers will just try harder to hit those that don’t, by striking more blogs harder, and faster, than ever. However, I guarantee that no method will be completely fool proof and ever succeed in being used by 100% of web sites; and as a result, spam will continue. Of course, there’s no need to be completely pessimistic about all this, it is another tool in a spam-fighter’s toolbox that may have some impact; it just won’t stop it entirely.

So, in summary, nofollow is harmful because it:

Update: changed the following list to an ordered list for easier referencing.

  1. Fails to indicate any semantic relationship between resources.
  2. Does not indicate what the resource is.
  3. Does not state the purpose of the link.
  4. Makes no sense for any user agent or user, other than a search engine using the page-rank (or equivalent) algorithm.
  5. May be abused to prevent any external links, regardless of their context, from a web site being counted towards page-rank, perhaps due to company policy.
  6. Has the potential to harm the benefits of mass communication among weblogs and their comments.
  7. Its name does not accurately represent its semantics.
  8. It will fail to stop comment spam completely.

Because of all this, I will not be endorsing nor using this relationship for any links on my site (assuming blogger gives me a choice in the matter), and I encourage all authors, weblog CMS vendors and search engines to seriously consider whether the small possibility of a slight reduction in comment spam is really worth causing so much devistation to the web community as a whole. Next, in part 2, I plan to discuss major revisions to my earlier link relationships proposal by incorporating all of the feedback I recieved and the ideas I have presented here to meet the criteria for well designed and useful relationships.

2 thoughts on “Link Relationships Revisited, Part 1

  1. In your conclusion you name 8 reasons why nofollow is harmful. Points 1, 2, 3, 4 and 7 are exactly the same argument: nofollow does not have a semantic meaning. This is true, and I agree that they could have chosen a better name. However, de meaning of nofollow is absolutely clear for the huge target group of this tool: the bloggers with comment spam.

    In point 5 you state that nofollow might be abused, but I would rather say ‘use’ and see it as a positive point. I have read nowhere that nofollow MUST only be used for comments, and neither is it common for web specifications to limit the use; rather they specify a platform. Robert Scoble already explained a possible additional use of nofollow: being able to write about negative experiences without leveraging the importance of the sites linked to.

    In point 6 you state that it will harm the benefits of mass communication. I’d rather state that it will undo part of the harm that has already been done, and bring the inflated importance of weblogs back to normal proportions. Many bloggers might not like it when they will be listed lower in the search results, but it will benefit the people using search engines to look for information.

    Finally you state in point 8 that nofollow will fail to stop comment spam completely. You are absolutely right about this, and all the parties involved agree with you. But is is a measure that is very easy to take, and will reduce comment spam on the long term.

  2. Jeroen,
    Points 1, 2, 3, 4 and 7 are not the same as I clearly addressed each point individually throughout the article, however you are correct in that each of those reasons fall back on the idea of semantics, which, as you have probably noticed, is really the core of the argument.

    Regarding point 5, you are correct that nofollow was not designed to be used purely for user feedback; however, my point was essentially that it may be over used for cases where there is really no reason for doing so; which is essentially due to the fact that its use cases, and thus semantics, are not clearly defined – only its function for one particular type of user agent is defined.

    The “inflated importance of weblogs” is not caused soley by the fact that links from comments currently increase page-rank for sites, but because weblogging as a whole has taken web communication to a whole new level of sharing newer, fresher, up-to-date ideas faster than any method has done in the past. Keep in mind that link counting is only one aspect of the whole page-rank algorithms; it’s just a shame that comment spammers don’t realise this, because it essentially renders their efforts futile anyway.

    I stand by my statement that it will harm the benefits of mass communication because it does unfairly punish the vast majority of legitimate users that provide comments, most of which are highly relevant to the article, including any links they provide.

    At least we agree that it won’t stop comment spam completely, though in fairness I did state in the article that it is another tool that may have some positive impact.

Comments are closed.