(Liz Crampton) #1

Hey everyone! Does anyone know of anyone doing work on predictive algorithms on their community content?

Here at Mind, we’re working with a think tank and data scientist to see if we can recognise patterns in ‘crisis’ posts (eg. particular words eg. ‘suicide’ ‘end it’ etc. or rapid succession posts). The goal of which is obviously to decrease moderation time but also to prioritise posts better, as at the moment we’re reading each of them and it’s not scalable.

Finally, is anyone using algorithms in another wya eg. to suggest very relevant content to members or even friends (LinkedIn, Facebook style etc.).

Would love your thoughts!


(Angela Seeley) #2

Hiya. We’re in an algorithm design phase as we speak. Our community is for English learners. They’re big pain point is they want to practice/use English more…but whatever they do, it has to solve a big tension - their desire to BOTH have fun and feel like it helps them progress. Magic sauce.

Ultimately, we want our platform dishing up content/people/interactions that solve this tension for the member sitting in front of the screen.

The step we’re on now is understaing what components belong in the logic, attempting to ignore our gut and do proper analysis to confirm what actually belongs (impacts if members click) and what has no bearing whatsover even though we may like to think it does because teachers always know best, right??? (vomit-emoji) :slight_smile: :slight_smile: :slight_smile:

So far this step involves looking at member behavior data and running click tests to identify how different member clusters make choices about how to spend their time (what are the components of “fun and useful”?) and what gives most regular and prolonged use.

So, that’s a thought for you I guess - confirm what goes in before you start hard coding in logic :open_mouth:

(Nick Emmett) #3

Hi @lizcrampton I think some of the larger platforms are starting to move this in to their feature set in the near future.
We use Salesforce’s Community Cloud platform and they have a Recommendations feature, where I can specify Content to be displayed in certain situations but also I can specify various “audiences” to be shown content types. Not sure about other platforms but I imagine it’s something coming on to the radar. Or this may be a slightly different thing!

(christopher w) #4

Very interesting question @lizcrampton and in terms of predicting behaviour, one of my interests.

Have you considered adapting recommender systems and machine learning to predict user’s behaviour?

Recommender systems are used in a number of online industries, notably retail for point of sale product suggestion and video on demand for predicting the types of videos a viewer might want to watch next.

Many of the recommendations being made are based on rules input by humans behind the scenes, although some algorithms use machine learning to spot behaviours and base their recommendations, at least partly, on behaviours which are observed and analysed in real time.

With this in mind here’s a paper discussing Machine Classification and Analysis of Suicide-Related Communication on Twitter.

And an excerpt:

In this paper we report the results of a number of machine classifiers built with the aim of classifying text relating to suicide on Twitter. The classifier distinguishes between the more worrying content, such as suicidal ideation, and other suicide-related topics such as reporting of a suicide, memorial, campaigning and support. It also aims to identify flippant references to suicide.

And here’s another although dealing with the more general issue of threat analysis - but a useful backgrounder: Recommender Systems for Intelligence Analysts.

Hope the above is helpful.


(Liz Crampton) #5

Thanks @Nick_Emmett and @Angie_J for your replies, they’re really interesting.

@ccdw That’s exactly what I’m trying to get at - thank you! Hadn’t heard of recommender systems, so thanks for that. Really useful links, I will check all of those out. If you have any contacts or further ideas about how we could apply any of that, I’d love to chat more!

(christopher w) #6

H @lizcrampton

Happy to chat on the phone if you’d find that helpful.

Also, you might find sentiment analysis services useful too - check out my comments here:

Ideas on analyzing sentiments:

Best wishes,