Input for Kiva staff on the "Popularity algorithm"

[Guest]

.

[RichardF]

This is my serious recommendation for the "Popularity algorithm."

• Guiding Principles: The Popularity Algorithm should be...
  
  • Based on Lending behavior
  • Objective
  • Unbiased toward key loan attributes, e.g., region; country; partner, including star rating, default rate, delinquency rate, nonpayment loss liability, currency exchange loss liability; time in Fundraising status; language; video; grouping; gender; sector; activity; amount; term
• Basic Algorithm:
  • For each listing, calculate its Popularity Index in (operationally feasible, e.g., every X minutes) real time as: (Funded Amount) / (Minutes in Fundraising Status)
  • Rank all Fundraising loans by their Popularity Index, with the highest index (the highest funded amount per minute) ranked first

[JoanW]

This is my serious recommendation for the "Popularity algorithm."

• Guiding Principles: The Popularity Algorithm should be...
  
  • Based on Lending behavior
  • Objective
  • Unbiased toward key loan attributes, e.g., region; country; partner, including star rating, default rate, delinquency rate, nonpayment loss liability, currency exchange loss liability; time in Fundraising status; language; video; grouping; gender; sector; activity; amount; term
• Basic Algorithm:
  • For each listing, calculate its Popularity Index in (operationally feasible, e.g., every X minutes) real time as: (Funded Amount) / (Minutes in Fundraising Status)
  • Rank all Fundraising loans by their Popularity Index, with the highest index (the highest funded amount per minute) ranked first

I agree with your Principles in theory - but when I think about how it would actually work in practice, I'm not so sure.

One of the things that concerns me is ending up with a large number of very similar loans clustered together. Some MFIs post a bunch of loans together that are for basically the same loan just a different entrepreneur. Since Kiva's stated purpose in this popularity sort is to entice new people to get interested in making loans, I think variety must be presented. If all the popularity sort shows is men with cow loans and that is not your thing - you are not only not loaning this visit, you are likely not going to return either.

What if we took your formula - (Funded Amount) / (Minutes in Fundraising Status) - and filtered (not the right word) by region.  Basically, index all the loans based on the formula as you said then list the top one from each region, followed by the second most popular from each region, and repeat. If you want more variety, then sector could be used instead of region.

This would build in some variety to the formula, while keeping it about primarily about popularity.

Where it says - Found 1161 Fundraising loans sorted by Popularity. It could be changed to say Found 1161 Fundraising loans sorted by Regional Popularity to clearly show that it is not a 100% pure popularity sort.

[RichardF]

What if we took your formula - (Funded Amount) / (Minutes in Fundraising Status) - and filtered (not the right word) by region.

Joan,

What you called, "filtered," I would call "stratified sampling."

In statistics, stratified sampling is a method of sampling from a population.

When sub-populations vary considerably, it is advantageous to sample each subpopulation (stratum) independently. Stratification is the process of grouping members of the population into relatively homogeneous subgroups before sampling. The strata should be mutually exclusive: every element in the population must be assigned to only one stratum. The strata should also be collectively exhaustive: no population element can be excluded.

The concern I have with "adjustments" to the basic popularity index is the introduction of bias.  I believe that's what Kiva does when it introduces non-lending behavior, e.g., pumping up "new," and video listings.  The key to slicing and dicing in an unbiased way is to ensure subgroupings are exhaustive and mutually exclusive.  The next thing to keep an eye on are those "weightings."  They frequently just turn into handy knobs to game the system.

Personally, I prefer a two-stage approach.  First, offer the most basic implementation of a default sort based on an objective, unbiased lending index.  Second, offer a powerful and flexible set of tools to customize and save a collection of "My Searches" for each lender.

[tomviolence]

One reason I thought the "cloud" idea was worthwhile was that with tags, each loans now has multiple dimensions.

And if you let the tags spin around in the cloud, dominant tags can be small (like "Men" or "Women"), while little bits of population can rotate through and be prominant ("Cobbler")

Instead of "man with cow", you can tag it: men, farmer, livestock, green, single parent, etc.

I realize there is no chance of this even being considered, but I thought it would be a neat, interactive, way to present the variety of loans available.

[RichardF]

tv, I like the idea of tags for browsing on sites like YouTube.  The next question gets to what kind of hits on those tags translate to popularity.  Viewing is quite different than lending, for example.  If Kiva were to use a cloud metaphor, how would you describe the Lend page algorithm?

[tomviolence]

you click on a tag that interests you, which then brings up a sort/search page. Search should be on the main page as well. I think people who know what they want should not need to do 600 clicks to get there.

Ideally, like the present page, having a lender profile is a good idea, having some recent loans is a good idea, and a true example of "popularity, and then having a a bit of an explanation on clouds and tagging. Many people know how tags work (Amazon uses them), but for those who don't, a simple "Click on a word that interests you to see what loans match !"

The second page then lists the loans in an "order" - open to discussion, but I think random is the most honest.

There you could "Click on a loan to read about the entrepreneur or click for more tags" - leading to the loan or multi level search.

I think you need to keep it simple for those who are happy to fund what ever is there, or powerful enough to let people get more involved, and make choices relevant to their world view.

Thus, from main page, you could loan (based on featured loan or most recent), you can make simple choices (cloud click "green"), or deep search.

 

[Limesarah]

I think my 'main' comment regarding the 'popularity' choice is...

I hope Kiva begins to see that 'clicks' or 'viewing activity' is a negative not a positive number.

When I 'click' and then move on, it should mean I have voted against that loan.

thanks, jan

That's not necessarily true, for me.  I might click on a loan while browsing, then come back and lend.  Or I might be out of credits, but really interested in the loan and trying to decide if I should add more money to my account, or watching to make sure it gets funded. 

[karlynwagner]

Sorry, I don't have time to find right thread.  I am at Kadet office.  Delivered calendars.  All is well.  Karen

[cjp1973]

Sorry, I don't have time to find right thread.  I am at Kadet office.  Delivered calendars.  All is well.  Karen

So glad to hear it Karen.  Thanks for checking in!
Charmaine

[alan]

This is my serious recommendation for the "Popularity algorithm."

• Guiding Principles: The Popularity Algorithm should be...
  
  • Based on Lending behavior
  • Objective
  • Unbiased toward key loan attributes, e.g., region; country; partner, including star rating, default rate, delinquency rate, nonpayment loss liability, currency exchange loss liability; time in Fundraising status; language; video; grouping; gender; sector; activity; amount; term
• Basic Algorithm:
  • For each listing, calculate its Popularity Index in (operationally feasible, e.g., every X minutes) real time as: (Funded Amount) / (Minutes in Fundraising Status)
  • Rank all Fundraising loans by their Popularity Index, with the highest index (the highest funded amount per minute) ranked first

How would this differ substantially from the "Amount Left" sort? As I have suggested above, would it not begin to resemble a slightly buggy version of Amount Left?

I agree with your Principles in theory - but when I think about how it would actually work in practice, I'm not so sure.

I'm with Joan.

"In theory there's no difference between theory and practice; in practice there is." - Jan van de Snepscheut