Setting up multiple, related topics

Yes, that is exactly the way to think about it. The Relevance engine should be much better at distinguishing the different categories with training than if you try to do it with keywords.

You should expect a few articles that do get classified as good in multiple categories as training is not 100% accurate. Also, an article that talks equally about Solar and Wind energy would naturally show up in both Solar and Wind categories.

I would be interested in how it works out. When you have some results, you can use the Contact Us menu item on our site to send an update if you have time.

Thanks much, Mark. I’ll be happy to report back.

Mark, here’s an update on our experience thus far.

1. Using no keywords whatsoever seems to be a hard way to go. Training alone seems to not narrow articles enough or it takes a good while to do so. that’s true even if you just look at good articles but of course even worse if you want to check not sure or bad articles.

2, There doesn’t seem to be much benefit or advantage to dividing the search by topic, As you note, we get a lot of the same articles and that just adds steps and multiplies the total amount of articles we need to scroll through.

So, given the natural overlap among our topics, I’m wondering if there is any reason not to use just one primary topic to search all of our sources (leave first keyword field blank but combine lists for the 2nd keyword field). Will the relevance engine “learn” just as well if it is getting told that a somewhat broader mix of articles are “good”?

We could still have a secondary catch-all topic that we could continue to try to train without keywords, but we would have to do that just once not multiple times.

We would then organize the resulting round-up compilation of articles by appropriate topic at the end, rather than trying to do so from the beginning.

Does that sound about right? Thanks, Jan

Hi Jan,

1. Training does not affect the volume of articles as you’ve found. They all still appear. The classification does give you a way to ‘prioritize’ your work with good articles being the first to review, then the other classifications less often.

Using keyword filters will reduce the volume. Articles that don’t meet the filter will not show up on the training page. If you are confident in your keywords, this will certainly help. If there are ways to write good articles without using your keywords, you may miss some good content.

2. One primary topic with a lot of keywords in the 2nd field should work. The more keywords in the 2nd field, the higher the volume of articles that will make it through the filter.

Training across multiple related topics should still yield some benefit. You will get more ‘good’ articles than a narrow topic. As a way to prioritize, you should still see that the good articles are usually more on target than the bad.

A trick that may help is to ‘bias’ the training. If you train a lot more bad than good, the classification will tend to find more ‘bad’ articles. Similarly if you overweight the training with good articles it will tend to find more good ones.

Finally, doing the round-up at the end will certainly work. You’ll have to remember which of all of the multi articles belong to which post.

In a week or so we are coming out with Notebooks in our next release. This should make your round-ups much easier. You can save articles in multiple notebooks, then use them in a post just like the multi. It should be much easier to organize your round-ups that way.

Mark