Better UX for dual (and more?) NGO and project search @ ProPoor?

Hi,

I came across propoor.org some weeks ago and as I clicked my way through the search form, I contemplated the idea of improving it.

Here is a quick prototype:
http://matthieubergel.com/propoor/

I might take it down at some point so here is a screenshot for future reference:

As it happened, I spent most of the time on two things:

  • getting my head around the best way to scrape a useful dataset
  • UX

Two columns is still acceptable but the propoor search can theoretically expand to up to 6 indexes, which, quota usage concerns set aside (6 queries per keystroke can get easily quite expensive), is starting to feel a bit crammed. I am wondering to what extent multi-index results are still a good idea cognitively speaking (https://birchbox.fr/) and when they become an overwhelming tech parade.

Another option would be to have less results per index but displayed in a grid, spanning the results on both columns and rows.

Interested to hear your thoughts on this.

Thanks!

Matthieu

1 Like

Hi Matthieu,
I think your question makes a lot of sense, and highlights that the whole challenge of a great search UX is to be able to display enough results to the users, so they can find whatever they are looking for, but without overwhelming them with too much data to process.

My believe on that is that it can probably be solved with the right UI.
One of the approach I like a lot, is to

  • create an “overview” default page that shows the most relevant and popular results for each type of data
  • and then to add to it specific views for each of those data types

By clicking on “Show more” or selecting one of the data type names in the tab list, they can access to a view that displays results for the selected data type + potential filters (very useful for browsing).

Of course all data type view doesn’t have to look the same and can be tailored to the type of data you want to display: some may display results mainly with pictures, some others as a list, and maybe another one with a map.

Benefits

  • at first the user gets a general overview of what are the different type of data he can search for
  • each view is tailored for a given dataset, enhancing the experience the user can have
  • the same approach can be easily used on both Desktop and Mobile website or applications

@mlbrgl What do you think of that approach? Do you think it could work for your project?

1 Like

Hi Alexandre, thanks for your comprehensive and documented reply. Yes that could definitely work, I very much like the idea of offering a custom interface tailored to the needs of each content type, if only to avoid “facets overload”.

Regarding the default results screen, I somehow think that they tend to create biases that get reinforced overtime, in the same way that businesses whose name started with “A” had a better chance to grow stronger overtime through getting more visibility in the phone book. From my perspective there is no way of knowing what is relevant for the user before he / she starts typing (profiling aside) and I would rather display random results than giving a few hits an “unfair advantage”. I have no data to back this up though, this is just a theoretical hunch. Maybe you have relevant data that could help seeing this from another perspective?

In the world of e-commerce, this is somewhat different because there might be incentives to promote a product over another and the self-fulling prophecy might actually be a good thing (promoted products get more visibility and generate more sales). But I am wondering to what extent this applies to search results which have no business value over one another, and where relevance to the user trumps any other criteria. Would the search as a whole still benefit from exposing arbitrary records vs random ones, or even none?

Another thing I noticed is that the terminology used by ProPoor might require a few lines of explanation (what projects and organizations are we searching through? Who are those individuals?). I am wondering how useful it would be to tell a short story to try and present these concepts, while still offering the general search as an alternative (which would then display the results in your suggested grid template).
Clicking on the concepts would take the user to the specific search interfaces:

Matthieu

1 Like

Hi Matthieu,
You just made me realize that I forgot to add the search query to the search field. Which indeed makes think that results / suggestion are showed without any query. Agree with you that it can create some bias but done well it could also be very powerful, since it can help surfacing popular content.

Regarding the default results screen, I somehow think that they tend to create biases that get reinforced overtime, in the same way that businesses whose name started with “A” had a better chance to grow stronger overtime through getting more visibility in the phone book. From my perspective there is no way of knowing what is relevant for the user before he / she starts typing (profiling aside) and I would rather display random results than giving a few hits an “unfair advantage”. I have no data to back this up though, this is just a theoretical hunch. Maybe you have relevant data that could help seeing this from another perspective?

=> It’s easy to avoid the “phone book” effect by sorting items based on the business popularity. For instance, on the following demo, results are sorted in desc way using the number of followers.

You could easily sort organisations and projects on propoor.org using for example:

  • for organisations: the number of views on an organisation’s page in the last 30 days, or the number of people following that organisation
  • for projects: it could be the number of people following the project, divided by the number of days or weeks since the project got published.

Regarding your idea of having a homepage with a bit more information. It’s a good idea. You could also imagine sliding the search field to the top of the page as soon as the user focusses into it.

What do you think?

Hi Alex,

I somehow got fixated on the idea that the metrics would necessarily go up, in an accelerated fashion thanks to more exposure. But your idea adds an interesting spin on it by depreciating the “currency” over time:

Right now I see two ways of doing this:

  • post retrieval ordering: get a not so healthy amount of results and then order them with some front-end logic. I would characterize this as “sub-optimal”.
  • periodically re-indexing the records for updating the business metric as a function of the elapsed time. Costly.

I am sensing that there is a way to leave the index untouched while performing those computations at query time (figuring out the elapsed time since publication and dividing the absolute metric by it).

Is that what you had in mind?