Similar Stories

[RetroFan]

At the end of each story on Literotica, there is a section where similar stories to the one you have just read are listed. Has anyone else ever been completely perplexed about how this works, when you have found that these similar stories are nothing like the one you have just read, or if it is your own story, written?

For example, one of my stories is called 'Sally and the Sailor', a simple enough first time story about a tomboy in World War 2 who loses her virginity to the young man next door who has enlisted in the US Navy. One would assume that similar stories to this would involve stories set during the Second World War, in the 1940s in general, about tomboys, sailors or other members of the armed forces, or perhaps pretty petite blondes like the the narrator Sally. But one would assume incorrectly.

At the other end of the scale, have you ever written two unrelated stories that deal with the same unusual themes (such as fetish) that should appear as similar stories, but do not?

 

[KeithD]

I would have thought that stories with near-identical tags would be used to determine this category. Not that I can tell, however.

 

[GroverLang]

The impression I’ve got is the ratings of other stories are around the rating of the story you’ve read. As for similarity, and quality, it can vary. Perhaps they’re chosen because of a combination the tags and ratings.

 

[NotWise]

At the end of each story on Literotica, there is a section where similar stories to the one you have just read are listed. Has anyone else ever been completely perplexed about how this works, when you have found that these similar stories are nothing like the one you have just read, or if it is your own story, written?

I seem to remember being perplexed by the "similar stories" selection on my own stories. I just grabbed one of my stories and looked at the list. I was pleased. "Words on Skin" by PacoFear, a story by 8letters, one by silkstockingslover, and "One Who Understands" by BurntRedstone.

I like the company the site gave me.

 

[Bramblethorn]

At the end of each story on Literotica, there is a section where similar stories to the one you have just read are listed. Has anyone else ever been completely perplexed about how this works, when you have found that these similar stories are nothing like the one you have just read, or if it is your own story, written?

As far as I know, there are two main ways to build recommender systems.

One is "content-based": for Literotica, that'd mean you fill in a detailed description of every story (tags for content, category, length, whatever seems like it might be relevant). Then you use that info to find stories that are very similar in content, and recommend based on that.

The other is "collaborative": you let users rate the content they read, and then you use those ratings to identify patterns of similarity. If the readers who liked Story A also tended to like B but not C, then you set the "similar stories" recommendations for A accordingly.

Collaborative is often easier for sites with lots of content, since it harnesses the readership to help produce recommendations instead of inflicting all that admin work on the site ownership. I would guess that Literotica is using some sort of collaborative approach based on ratings.

One weakness of a simple recommender system is that it can only figure out "if you liked Story A, you'll probably also like B" after several people have rated both those stories. On a site with hundreds of thousands of stories, that's not guaranteed. A lot of similar stories will be missed, and since individual voters can have diverse tastes, the fact that Jane Doe five-starred both "Regency Ravishment" and "Bob's Bear Hunt" may generate bad recommendations. And before the first votes come in, you don't have any information of that kind, so you'd have to use a fallback method, maybe tag-based.

There are more sophisticated approaches that can partially mitigate those problems, but I think many of them were developed after Literotica launched and I don't know what it's using.

If my guesses are right, then the quality of recommendations would be very dependent on the number of votes and/or favourites a story has. But they're only guesses.

 

[Kojak01]

One weakness of a simple recommender system is that it can only figure out "if you liked Story A, you'll probably also like B" after several people have rated both those stories. On a site with hundreds of thousands of stories, that's not guaranteed. A lot of similar stories will be missed, and since individual voters can have diverse tastes, the fact that Jane Doe five-starred both "Regency Ravishment" and "Bob's Bear Hunt" may generate bad recommendations. And before the first votes come in, you don't have any information of that kind, so you'd have to use a fallback method, maybe tag-based.

My "problem" with the similar stories is that there are always the same maybe 25 stories per category which show up for me.

 

[RetroFan]

The impression I’ve got is the ratings of other stories are around the rating of the story you’ve read. As for similarity, and quality, it can vary. Perhaps they’re chosen because of a combination the tags and ratings.

In some cases yes, but some story matches are just flat out weird.

For example, my Christmas 2017 story was a romance story called 'Take Cover From Tracy', and is about a young man and woman who have a chance meeting in the Australian city of Darwin at Christmas 1974 and fall hard for each other, only to be caught up in Cyclone Tracy, which destroyed Darwin in the early hours of Christmas morning.

For months, one of the similar stories that showed up against this one was a fan fiction fetish piece called 'Hayden Panettiere's Feet'. Oddly enough, the female protagonist in my story in my mind at least did look a bit like this actress, although I didn't think of this when writing the story. And my story did use the tag feet, with the lead female character's feet often described in scenes. Foot fetishists may have liked this aspect of my story.

But given the amount of foot fetish stories and fan fiction on Literotica, it seems odd that a fan fiction foot fetish story would link as similar to a historical romance piece set many years ago and half a world away in Australia.

 

[UsuallyPresent]

The clumpiness of the data - lots of small numbers of ratings, a few with a lots of ratings, tons with no data - generally drives most analysis systems to focus results in the small number of data points with lots of responses. Given how ratings, favorite stories, and comments appear to be spread on this site - lots of data on a relatively few highly popular stories, some data on many more, and almost no data on the vast majority - it seems likely that the dataset Lit has to work with is damn clumpy!

This is not to say there couldn't be some or lots of hidden metadata the site could keep & use to get better results. Time-per-page for instance might be a quite interesting & useful metric, although it would be difficult to gather 100% of the time (browsers today tend not to provide a clean farewell message when leaving a page, especially if moving to an external site.)

Amazon has unsurprisingly done a lot of thinking & trying things in this area, unsurprisingly. This article has a reasonable overview from outside Amazon, and this article is how Amazon chooses to resell the capability. These can be useful to start to scope the discussion, and there's LOTS of academic work available as well.

 

[Voboy]

I don't know how they do it, but I know the effect. Suddenly, a story out of my back catalog will get a bunch of new favorites and I'll get a little uptick in followers. I reckon it's because that story's been tagged in something good that's just posted.

If I cared, I'd go find out which new story. But I don't care. It drives traffic, and that's the intent.

 

[gordo12]

At the end of each story on Literotica, there is a section where similar stories to the one you have just read are listed. Has anyone else ever been completely perplexed about how this works, when you have found that these similar stories are nothing like the one you have just read, or if it is your own story, written?

I have no idea how that process doesn't work but it doesn't. I gave up a long time ago looking at or reading any of those stories. If it's based on tags or keywords it's only using a small sampling of them in comparison. The stories are often in other categories I never read. I doubt I'm ever going from LW to Lesbian or BDSM or Gay.

I've always considered it one of the many broken parts of the site.