StumbleUpon | CastorQuinn's blog posts

http://CastorQuinn.stumbleupon.com/review/35980278/

Thu, 10 Sep 2009 19:03:11 -0700

Have a good, relaxing day everyone.

Especially those millions of people who die around the world every day from preventable poverty, lack of access to medical resources, food and clean water. Especially those killed by civil wars, foreign aggression and home-grown insanity. Especially those with a sense of perspective.

I'd say 'lest we forget', but we'd have to know about these tragedies before forgetting them would become an issue.

http://CastorQuinn.stumbleupon.com/review/25808183/

Thu, 25 Sep 2008 19:33:26 -0700

I know that it's bad form in flowcharts to have a loop with no escape vector, but spend any amount of time in the cores and you'll realise that really is how it works. Esteban knows it.

Hosted at ImageShack.

http://CastorQuinn.stumbleupon.com/review/22618733/

Wed, 18 Jun 2008 18:05:15 -0700

THE CATEGORY SYSTEM

Every page that enters the SU database is placed into a category, and that category influences which pages you stumble.

One common misapprehension is that the SU category system is a taxonomy. It is not. The list of categories is not exhaustive, it has differing levels of specificity in different areas, and categories have no internal hierarchy (despite the broad grouping you see on the topics page, which is purely presentational and has no real impact on the system).

So if the categories are not taxonomic, what are they? The answer is that they are fuzzy sets. This is where you take a broad conceptual space and, rather than dividing it into rigidly defined areas with clear boundaries, you instead define certain points, or characteristics, within that space and then group items based on how similar they are to the various characteristics.

Now that's all lovely, but what does it mean in terms of SU? SU categories aren't defined by dictionary definitions or criteria that must be met for a page to qualify to belong to a certain topic. Instead we use the characteristic of similarity of interest - that is, would someone who likes the content of this category also like this page?

Let me give an example. SU has a Pornography category. Pornography is defined in dictionaries as a depiction of intercourse or genitalia for the purpose of titillation or the inducement of sexual arousal. If SU categories were taxonomic, that would mean that any page which doesn't satisfy the criteria of including either the sex act or genitalia for the purpose of sexual arousal wouldn't belong in that category. However the Pornography category includes a broader range of content than this: we have pictures of naked women where the genitalia is not visible; we have pictures of people on their own, not having sex; we even have pictures of people fully clothed but in erotic poses. These aren't technically pornography, so why are they in there?

Because the Pornography category is a fuzzy set grouped around the idea of similarity of interest. If you are interested in pictures of people having sex, you would also be interested in arousing pictures of a semi-clothed woman, or an evocative picture of a woman with enormous breasts threatening to - but not actually - bursting out of her sheer, scanty blouse. SU categories exist to allow SU to show you content you might like, so rather than sticking to a rigid definition and saying "Sorry, we can't show you this, technically it's not pornography", SU instead says "This is very similar to the sorts of things you're looking at so we reckon this might be up your alley".

Another example might be the division between Humor and Adult Humor. If you like ribald jokes and sexual innuendo, there's a good chance you'd enjoy a video of a remote control flying dildo. However flying dildos don't necessarily indicate an interest in knock knock jokes or Garfield comics. So Humor and Adult Humor are split into two categories based on the idea of a similarity of interest.

The same principle is used to separate content that a rigid definition would actually put together. For instance we have a Judaism category. Anti-semitic pages are technically about Judaism - if SU categories were taxonomic then these pages would go in the Judaism category. But obviously people looking for websites about Judaism don't want anti-semitic content, so the fuzzy sets actually separate this content into two different categories, based again on similarity (or dissimilarity) of interest.

This is the principle behind all the categories, and to which categories pages are assigned. The human brain is designed to box things up, to categorise them and group them - it's how we learn language and form societies. But if you keep the idea of fuzzy sets at the back of your mind, the category system makes a lot more sense.

http://CastorQuinn.stumbleupon.com/review/22349987/

Wed, 18 Jun 2008 18:04:51 -0700

STUMBLEUPON INFOPOSTS

Following the example of several other stumblers, I'm beginning an archive of posts about how SU actually works, since a lot of the systems and features around here aren't very well understood by most stumblers. Watch this space, as I'll add to it whenever I come across a topic that needs explanation.

The SU Algorithm
SU Friends - How They Work
SU Friends - Common Questions
The Category System
Categories and Miscatting
Similar Stumblers.

Opinion Posts
These posts are not so much information about SU as my perspectives on the place and how it does - or should - operate.
SU Friends - The Reality

Infoposts by Other Stumblers
Xineann's Archive

SU Help Resources
The Official FAQ
The Help Forum
Mr Helpful - a guide to formatting and content on SU pages.

If anyone has any topics they'd like explained or which would benefit from a broad explanation, drop me a message and if I can I'll post it up. Many questions are already covered by the existing help resources, so try there first if you need help with something. Note that this is an unofficial resource, based on my own observations and tests of SU and discussions some of us have been having about this place since it started; it's not based on any insider information from the devs, it's just a best guess about how things work around here.

http://CastorQuinn.stumbleupon.com/review/22553685/

Mon, 16 Jun 2008 17:42:23 -0700

SU FRIENDS - THE REALITY

So we all by now have our heads around how the SU friends system works mechanically. Theoretically we all add and remove friends based on how good an informant they make to our stumbles, right?

Yeah, right.

In reality, no matter how mechanical our friends list is, we take it personally. I know I do. I try to only add people to my list if I think they will make good stumble informants, but removing friends is a completely different matter. I have friends on my list that, no matter how much mechanical sense it makes, I just can't bring myself to remove. Maybe you have some of these categories of friends on your list:

The Real Friend - Some of us are lucky enough to have made real, solid friends here at SU, people we count as important to us. And some of us have really good friends whom we have coerced into joining SU, only to find out that they don't stumble often, or their commentary sucks, or their stumbles are a bit meh. But we keep them on our list because ... well, because they're mates, and we want them linked to us, however shitty or infrequent their stumbles may become.

The Gorgeous Person - Some people are just beautiful people, really kind, nice, innocent - the sorts of people who remind us how wonderful life can be. But they tend to stumble kitschy, glittery, cute images and thoughts that, quite honestly, don't gel with the cynic inside us. So why do we keep them around? Because defriending them would be like kicking a puppy, and besides, sometimes it's nice to just see them on your list, to remind you that there is more out there than porn, lolcats and bizarre/oddities.

The Long-Term Mutual - Some of us have been here for many years, and have friends from back when we first started stumbling. Those friends remind us of what this place used to be like, and even if they have stopped stumbling, we don't want to disassociate ourselves from them.

The Dearly Departed - Some stumblers go away. Maybe with notice, maybe without, but they are people that we valued as friends back in the day, and we don't want to remove them because, you never know, this may be the day that they finally come back to us, and we want to be here with open arms to welcome them home.

The Friend of a Friend - Sometimes we don't want to defriend someone because doing so would offend someone else. Classic example: You have a boyfriend and girlfriend duo in your mutuals. You can't defriend one without defriending the other.

As well as these mainstays, it is hard not to be offended when someone defriends you, especially someone who has been a long-term mutual. Even though we all know it is just mechanical, it's still as if they are saying: "Your stumbles - which are the basis of our friendship - no longer interest me". For someone who takes pride in their stumbles and their commentary, this can be quite a blow to the SU ego - especially coming from someone whose tastes you consider very good.

So what's the soln? Currently there isn't one. What we need is for the devs to split social friends from SU informants. Lend your voice in the Features forum if you feel the same as I do. Here's hoping this change will be implemented soon.

http://CastorQuinn.stumbleupon.com/review/22350684/

Mon, 09 Jun 2008 21:40:22 -0700

CATEGORIES AND MISCATTING

How are page categories assigned?
There are two systems that kicked into action when a new page enters the database: the pop-up discovery window, and the autocatter.

Ideally, when someone discovers a page (ie they are the first stumbler to thumb that page into the database) that discoverer gets to select the topic they think is most appropriate for the page. This option appears on the pop-up discovery window, where the discoverer gets to enter the page title, their comments about the page, their tags, the page category, and whether or not the page is adult.

If the discoverer does not select a category for the page from the discovery window, SU has an autocatter, a script which attempts to guess the category of the page. Obviously the autocatter is not hugely accurate. The autocatter seems to be based on the words used in the page title more than anything else (though it may also look at the textual content of a page), and usually only categorises pages into the broader categories (humor, weblogs, computers, arts, and so on), simply because it doesn't have the information to be able to understand the finer categorisations. Also, there is a greater chance of accuracy when looking at the broader categories.

The 'adult' Radio Buttons
On the discovery window are two radio buttons, allowing you to nominate a page as adult or not adult. Be aware that these buttons are badly designed and very misleading.

If you select a non-adult category for a page, but then click the 'adult' radio button, the page will automatically be sent into the Pornography category.

The rationale behind this is that if it is unclear whether a page is adult or non-adult, it is better to err on the side of adult, and thus remove this content from the potential stumbles of children. The Pornography category is seen as the most general adult category, and so all these questionable pages end up there. This is, by the way, the reason why the Pornography category has so much junk in it - the developers consider degradation of the porn category to be acceptable, as porn-stumbling seems to be considered non-core use of the service.

This page is in the wrong category. What can I do?
If you go to the review page for the page you are looking at (click the speech bubble icon in your toolbar), down the bottom of the right-hand sidebar you'll see this text:

Is this site not about [category]? Report it

If you click the Report it link, this will give you the opportunity to say what the category for the page should be, and also give a short explanation as to why the category should be changed. This will then send the page to the miscatters, who will review your suggestion and, if your suggestion is in line with the category guidelines, make the change.

If you are the discoverer of the page, and noone else has thumbed the page yet, the Report it line of text will be replaced with a drop-down menu, allowing you to change the page category right away. Once the page has entered circulation though (been thumbed by others) the category is locked in place and can only be changed by the miscatters.

How does miscatting work?
Basically, SU maintains a list of all the sites that stumblers have reported as being in the wrong category. The miscatters are volunteers who go through this list and change any categories that need to be changed.

I reported a page a long time ago. How long does it take for a page to be recategorised?
It can take a while. There are very few miscatters, the miscatting process itself is very badly designed, very clunky, and makes the process very labour-intensive. The devs don't appear inclined to improve the system any time soon (despite constant prodding for several years), so the process will likely remain slow

http://CastorQuinn.stumbleupon.com/review/22349722/

Mon, 09 Jun 2008 20:57:05 -0700

SU FRIENDS - Common Questions

Should I remove a friend who is inactive?
Not necessarily. If someone is inactive they won't be adding new content to the database or to their pages. But the SU algorithm isn't based on new content, it's based on similarities across the whole database of pages. So long as that person has thumbed a good number of pages (a couple of thousand or more) then they still function as a friend - they still help SU decide which pages you are likely to like.

However there is very little benefit in adding a friend who ceased stumbling before you started stumbling; you simply won't have enough pages in common for them to be a useful informant, unless you decide to go through and thumb their stumbles manually. They will still impact your stumbles, but far less significantly.

How many friends should I have?
This is a contentious question. You need about ten friends just for the SU algorithm to work, and the general consensus is that 50-60 friends is about the right number to give you really good results, assuming you have broad interests and your friends are carefully selected to reflect your interests. The more friends you add, however, the less impact they will have on your stumbles. This is just a matter of statistics: The more people you select from the community, the less specific your sample is; the larger the sample, the more it represents the average of the whole community, and if your friends as a group represent the average, that's the same as having no input from friends at all. People generally report a degradation of the quality of their stumbles when they have 150+ friends, assuming their friends are all carefully selected; for those stumblers who select friends randomly or select friends with whom they have little in common, this degradation can occur much sooner.

So ideally, assuming you select the majority of your friends because of their stumbles, 50-130 seems to be the optimum number of friends to have.

Why can't I have more than 200 friends?
SU has capped the number of friends you can have at 200. It's assumed that the primary reason for this is the one I've just mentioned: that large numbers of friends average out to give the same input as having no friends at all, and so the cap prevents people from degrading their stumbles through selecting too many friends.

There may also be a technical reason: Depending on how the SU algorithm is actually written, it may have to calculate the input from all your friends every time you hit the stumble button, in which case having large numbers of friends across the system would drastically increase the computational power needed to provide stumbles.

How can I tell if someone would make a good friend?
There are a few ways:
- Look through the first few pages of their blog. Do you see lots of pages there that interest you, that you would like to stumble? If you are opening 50% or more of the pages they have thumbed, there's a good chance you'd enjoy the rest of their content.
- Look at their tags on the sidebar. The larger the tag appears, the more pages they have tagged in that category. Are their largest tags things you find interesting? If half their tags are technical and half are artistic, and you are only interested in artistic content, they may not make a good informant, but if the majority of their larger tags are things you find interesting, consider adding them as a friend.
- Do your existing friends have someone listed as a friend? If so they may make a good informant for you, since they have the same or similar interests to someone already on your friends list.

http://CastorQuinn.stumbleupon.com/review/22349613/

Mon, 09 Jun 2008 20:52:44 -0700

SU FRIENDS - How They Work

The "friends" feature is one of the more confusing and contentious of the systems we work with at SU. This is mostly due to the fact that it is presented as a social feature when it is actually a mechanical part of stumbling, and also because the naming of the different aspects of the friends system makes these differences unclear.

Let's start by explaining what the terms mean:

A friend is someone you have added as a friend - you have clicked a button on their profile page and added them to the list of your friends; you have become their fan.
A fan is someone who has added you as a friend - they have clicked a button on your profile page to add you to their list of friends.
A mutual friend is when two stumblers have added each other as friends - they are both fans of each other.

Friends: The most important of these three terms is the first one - "friend". Adding someone as a friend - becoming their fan - has several effects on your use of StumbleUpon:
- They appear on your Friends page;
- They have a blue (or blue and red) icon next to their avatar when they appear on your visitor list or on review pages;
- They may appear in the sidebar on your profile (visible only to stumblers who visit your page, not to you unless you View page as others see it), as part of a selection of your friends;
- When a friend thumbs up and reviews a page, that review appears on your What's New page;
- When viewing the review page of a site, any comments written by your friends appear at the top, after the discoverer, not just in their chronological position;
- Most importantly, any pages thumbed up or down by a friend are more heavily weighted by the SU algorithm when considering what pages to send you as stumbles.

As you can see, these are all mechanical features, not social features. Even though friends implies "social", adding someone as a friend has no impact on whom you can message and who can send you messages, who can see your profile or your stumbles, with whom you interact in the forums or in groups, with whom you can interact at SU, or anything else that is social. Friends only relate to stumbling.

Fans: You have no control over who adds you as a friend - over who becomes your fan. However someone becoming your fan has no impact on anything whatsoever at SU.

Mutual friends: If someone becomes a mutual friend, this makes no difference to your stumbling experience. Becoming a mutual simply means that someone of whom you are already a fan has added you as a friend; it is mechanically identical to gaining a new fan.

So what does all this mean?
In a nutshell, all this means is that you should add and remove friends based on their stumbles, and that if someone adds or removes you as a friend you should take this as a reflection of their interest in your stumbles, not of their opinion of you as a person or of how much they like you. Adding someone as a friend has a definite impact on the quality of your stumbles. If you see someone whose stumbles you like - whose thumbs match yours, whose pages interest you, whose commentary you wish to see - add them as a friend. If there's someone on your friends list whose interests have drifted away from yours, or who's causing you to see pages you have no real interest in, remove them. Make that decision based on their stumbles, and on how their stumbles impact on your stumbles.

http://CastorQuinn.stumbleupon.com/review/22349425/

Mon, 09 Jun 2008 20:45:35 -0700

THE SU ALGORITHM

When you click the Stumble button, SU doesn't just send you a random page. It sends you a page from its database which it thinks you might like. The exact algorithm that is used to calculate the probability of you liking a particular page is proprietary, and secret, but it sits at the centre of SU's functionality and so there are some educated guesses we can make about how it functions.

You click the stumble button and the algorithm kicks in to figure out what page to send you as a stumble. The algorithm considers everything in the total SU database of pages as a starting point, but cuts out anything you have already thumbed (whether you stumbled it or not), as well as your own profile page. It then starts weighting the remaining pages - giving more or less weight to each page based on the factors listed below. Once the weightings have been assigned, the algorithm has determined which pages you are more likely to like, and which pages you are less likely to like. Some sort of random function kicks in at this point to shuffle up the results while preserving the general weightings. That's how the system basically works, so let's look at which factors impact on the weightings.

The algorithm considers these factors when assigning weightings:
- The categories you select as your interests
- The general popularity of a page
- Whether similar stumblers thumbed the page
- Whether your friends thumbed the page

It firstly looks at the categories you have selected as interests and gives pages from those categories a heavier weighting; this weighting must be significant, because generally stumblers do not stumble pages outside the categories they have selected, except in cases where they have selected only a few topics, or have exhausted the pages within their chosen topics.

The algorithm also factors in the popularity of a page. We don't know whether this is overall popularity, or current popularity ("buzz"), but it is a factor, as buzz pages do circulate more quickly and more frequently than older content.

The algorithm now considers how similar you are to other stumblers. SU maintains these similarity ratings separately, and you have a similarity rating with every other stumbler based on the similarity of your thumbs. If a similar stumbler has thumbed a page up, that page is given more weighting by the algorithm; if a similar stumbler has thumbed a page down, that page is given less weighting by the algorithm. The more similar another stumbler is, the more weighting they give to pages.

This is where your SU "friends" come into the equation. Your friends add weightings to pages in the same way that similar stumblers do. The weighting of pages based on your friends seems to be the most significant of the weightings, as changing your selection of friends has an immediate and significant impact on the stumbles you see.

Things that don't effect which pages are sent to you as stumbles:
- Who your fans are, the number of fans you have.
- Whether you are a sponsor or not (except insofar as sponsors have the option to turn off sponsored stumbles).
- How many thumbs up or down your profile/blog has received from other stumblers.
- The way that pages have been tagged by other users, or by you.

http://CastorQuinn.stumbleupon.com/review/15500423/

Sun, 23 Dec 2007 14:33:14 -0800

Happy mid-financial year marketing day, consumerist whores. May the European saint that Coca-Cola rebranded to sell its product bring you all the material goods you think you need to be happy in the company of your loved ones.

And please, think about the kids this Christmas.