Input for Kiva staff on the "Popularity algorithm"

[Guest]

.

[KFLiaison]

The staff at Kiva have been discussing the "Popularity algorithm" question internally for the last couple days.  Gerard has asked me to float these questions at KivaFriends to provide some insight into our community thinking about the issue:

1.  What are we optimizing for?  Increased funding rate?  Equal loan exposure?  etc...

2.  What are the alternate proposals and how will they help achieve (1)?

I know we've touched on issues relative to these questions before, but please offer your thoughts in organized form here, and I'll make sure this input is cycled into the Kiva staff discussion.


--Diane.

[Dagfinn]

Ok, here is my input:

1. I want to aid those needing exposure the most, based on statistics gained (how difficult theay are to fund) and number of days left.  I need to see that Kiva is an equal opportunity provider; that calls for doing something extra for some or they are left out altogehter.

2. Remove popularity sort order entirely and make a new one that highlights those needing it the most instead.  It must be a mix of days left and statistics on how hard it is to get funding.

What a relief  that there is a whole in the wall, Dagfinn A

[Jan & John]

1.  What are we optimizing for?  Increased funding rate?  Equal loan exposure?  etc...

I think moving the loans as fast as possible is the goal here and showcasing on the first page makes sense. 
What I find difficult to believe is the actual word "Popularity".   I always thought that meant that more people had chosen these loans.  Now I find this sort is manipulated in some magical computer fashion.  The fact that newly uploaded loans (0% funded) appear on the first page with no one lending to them yet makes the term 'popularity' pretty much a falsehood if not a bald faced lie.

2.  What are the alternate proposals and how will they help achieve (1)?

I feel that the popular sort should actually show loans with the greatest % of loan funded.  That way both the lenders and the MFI's could tell what has been selling.  New lenders could feel more comfortable and MFI's could actually judge how well the profile markets the loan.

The only other alternate proposal I think would be to change 'Popularity' to 'Random Selection' because this sure seems to be what's up there today.  And of course, more sort options and better breakdowns of sector choices are absolutely necessary but that's another topic

jan

[Jeremy]

I think that the current popularity sort is just fine. I think that the solution may be to make the "Ending Soon" sort the default. I'm sure that there are many lenders who just loan to whoever they see first, so putting the expiring loans as the default would probably fund a lot of those loans.

[Robert]

The meaning of the word "popular" in this context is "what the people likes". As such, any individual value judgment should be stripped from it. No loan should enter the stage with a manipulated start value. If popularity is measured from 0 to 100, each loan should start with 50 and then move the scale up or down according to the users' actions: clicking, fast lending, slow lending, basket hoarding, basket releasing, etc.

However, no-one will be happy with this solution. Everyone wants to change the algorithm to one's own preference:

• The owners of the website want an interesting and ever-changing homepage and prefer new loans.
• The technically minded people believe that videos are more interesting than pictures.
• I can imagine that fans of loans to Americans like to highlight loans with videos, as they are mostly to Americans.
• LLL-lenders want to transform the algorithm to include expiring loans, basically transforming it partially into an "unpopular" index.
• Even the topic of this thread can be read as: "how should we manipulate popularity?"

I am convinced that few lenders are interested in a popularity scale in the true meaning of the word. Similarly I believe that few lenders will be happy with a manipulated popularity index, as it will hardly reflect their own preferences and will as such be of no practical value to them. That is why I suggest to scrap the popularity index.

I want to replace it as follows:

• On the homepage I want 2 other features:
  
  • A random selection of a few loans on the homepage. So it can be said that every loan has the same chance. Furthermore, the homepage will look differently every few minutes. The featured loans would be like the featured lender or featured lending team.
  • A section labeled "Loans in need" which comprises a random selection of red loans. This section should of course only exist when red loans are available. It should come either above or below the random selection of loans. Alternatively, it could replace the section with random loans whenever red loans are available.
• On the lend pages I want the default sort "Old to New". It is the most useful to have a smooth lending activity. It is by the way also the default sort used by MyC4 (they call it "Bidding Ends"). I cannot remember ever having read a complaint or alternative suggestion on their lively discussion forum.

Robert

[Alaska Pack]

Here is an e-mail we received today.

Thanks so much for your email. At the moment we don't have any plans to change the default sort order for loans. Our hope is that, if expirations do happen, it will reinforce to our partners that they should post more compelling loan requests or otherwise change the pace of their posting.

Thanks again for your feedback!

Best wishes,
Gerard Niemira
Customer Service Manager

                                                                                                      

[Bastet]

Sorting from old to new makes a lot of sense, in my mind, although it does come with the issue that, unless the loans get funded, the first page on the 'lend' tab might be slow to change, and might not grab people's interest.

[wthepoo]

The meaning of the word "popular" in this context is "what the people likes". As such, any individual value judgment should be stripped from it. No loan should enter the stage with a manipulated start value. If popularity is measured from 0 to 100, each loan should start with 50 and then move the scale up or down according to the users' actions: clicking, fast lending, slow lending, basket hoarding, basket releasing, etc.

However, no-one will be happy with this solution. Everyone wants to change the algorithm to one's own preference:

• The owners of the website want an interesting and ever-changing homepage and prefer new loans.
• The technically minded people believe that videos are more interesting than pictures.
• I can imagine that fans of loans to Americans like to highlight loans with videos, as they are mostly to Americans.
• LLL-lenders want to transform the algorithm to include expiring loans, basically transforming it partially into an "unpopular" index.
• Even the topic of this thread can be read as: "how should we manipulate popularity?"

I am convinced that few lenders are interested in a popularity scale in the true meaning of the word. Similarly I believe that few lenders will be happy with a manipulated popularity index, as it will hardly reflect their own preferences and will as such be of no practical value to them. That is why I suggest to scrap the popularity index.

I want to replace it as follows:

• On the homepage I want 2 other features:
  
  • A random selection of a few loans on the homepage. So it can be said that every loan has the same chance. Furthermore, the homepage will look differently every few minutes. The featured loans would be like the featured lender or featured lending team.
  • A section labeled "Loans in need" which comprises a random selection of red loans. This section should of course only exist when red loans are available. It should come either above or below the random selection of loans. Alternatively, it could replace the section with random loans whenever red loans are available.
• On the lend pages I want the default sort "Old to New". It is the most useful to have a smooth lending activity. It is by the way also the default sort used by MyC4 (they call it "Bidding Ends"). I cannot remember ever having read a complaint or alternative suggestion on their lively discussion forum.

Robert

Hear, hear! Excellent comments and suggestions.

I still want to summarize my own opinion to basically say the same thing with different words - sorry about that, but we now know that numbers do count for Kiva (500 e-mails?):

* The current "Popularity" sort has little to do with actual but rather with assumed popularity (new loans start with a bonus, loans with videos start with a huge bonus - regardless of how popular they prove to be).

* Few lenders (outside KF) will actually take the time to alter the sort, to browse through 30 or 40 pages of loans or even realize when they are looking at a true "gem" of a loan - most lenders with "average portfolios" (i.e. less than 10 loans) will find the majority of Kiva loan requests almost equally appealing and touching simply because they lack comparison. Thus, a "Popularity" sort (assumed or actual) as default acts as a self-fulfilling prophecy (assumed popularity => actual popularity) or at least an accelerant (or "turbo") for the funding of these "popular" loans on the one hand, and (in times of plenty) as a restrainer for the funding of the "unpopular" loans. - Thus, the "Popularity" sort default makes "unpopular" loans automatically even less popular! And loans very easily and quickly become "unpopular" when posted around the 1st of the month.

* Thus, either scrap the "Popularity" sort at all, nobody really needs it, or at least don't make it the default (FIFO - first in, first out, i.e. an old-to-new default [don't call it "Expiring Soon" please, as all loans are listed, regardless of whether they are expiring soon], sounds great to me as a default; but a completely "random" default sort would be fine, too, might just be problematic to implement in a way to make sure that while browsing through the pages you will see most loans but only few loans twice [though that also is a current problem with Popularity and indeed with new-to-old and even (!) old-to-new]). Just tweaking the "Popularity" algorithm wouldn't address the problem mentioned before.

* Regardless of the default sort order, Kiva can still highlight (or as Jan called it: showcase) a couple of popular, particularly touching, well presented [video?] etc. loans on the homepage to explain to new lenders why Kiva is so great (these don't even have to be currently fundraising loans, I'd rather choose a couple of success stories with recent journal updates, but well...). Also, the homepage could be used to present loans in acute danger of expiring (when there are any) and - as Robert suggested - completely random loans, changing every 5, 10 or 15 minutes.


Best wishes,
Wolfgang.

[RichardF]

1.  What are we optimizing for?  Increased funding rate?  Equal loan exposure?  etc...

• Minimize the time to fully fund each listing.

2.  What are the alternate proposals and how will they help achieve (1)?

• Change the default Lend page sort order to "Old to New."
• Add a "My Searches" feature to individual lender portfolios.  Advanced filter and sort capabilities should be added to this feature.  This amounts to the same (more powerful) engine needed for the matching and auto-lending features.
• Allow the default on the Lend page to be changed to "My Search" as described above.
• Disburse the loans collected for partially funded listings.

This will help minimize the time to fully fund each listing first by implementing a default, classic first-in, first-out inventory management queue structure for most (the casual) lenders.  Second, it will allow Kivaddicts uh power lenders um connected people or lenders-who-know-what's-Popular!-with-them the opportunity to more efficiently focus in on the real meaning of Popular listings!  Finally, disbursing partially funded listings after the listing period expires precludes the need for lenders to adopt what-if-it-doesn't-get-fully-funded tactics for listings they already have determined to be "popular enough" for them.  Refunding partially funded amounts will unnecessarily disrupt the lending cycle and could lead to a downward spiral of ever-increasing "unpopular" listings.  As far as I'm concerned, if even one lender finds a listing to be popular enough to make a loan commitment, then that's good enough for me!
 

[alan]

I agree with Wolfgang  and Richard 

Just to add my own tuppence:

The Popularity algorithm, regardless of how it is devised, is intrinsically arbitrary. I would want to ask Kiva two questions:

1) What is the purpose of the Popularity sort?

2) Is Kiva unalterably committed to retaining the Popularity sort as the default?

In the absence of answers to those two questions, I suggest the following:

1) Ideally, scrap the popularity sort as default, and as already suggested numerous times, use an Old-to-New, FIFO sort as the default.

2) If Kiva is unalterably committed to retaining a Popularity sort as default, then remove the most egregiously arbitrary elements from it completely, namely the tweaking for video and the practice of listing new loans as the most popular.

I would add that we must disabuse Gerard of the fallacy that danger of expiry indicates that a loan is so unpopular that it simply can't attract funding. The activity of the past few days reveals how fallacious this notion is. Scores of loans on the brink of expiry have been able to attract funding due to two factors: individuals and teams such as the LLL team have made a concerted effort to support these expiring loans; and others who are perhaps completely unaware of the issue have happened to stop by the site long enough to notice one or other of the expiring loans on the front page and add their support. And a good number of similar new loans have been funded in the past several days. There is, in fact, nothing obviously wrong with any of the loans that have otherwise languished, and the fact that to date the funding rate is 100% - a record in grave danger right now - indicates that to date no loan request has been so defective or unpopular that it will not attract funding from the Kiva lenders.

I acknowledge that there is plenty of credit to share for this effort and it is obvious that several teams have joined it. I also note that the constant stream of new loans coming in at first place has made it more difficult to keep loans on page 1, and this indicates all the more how much of an effort has gone into this campaign.