People Matching

Whether you are looking for a ”partner in crime”, a new friend, a business partner, or just someone who lives in your neighborhood there is a lot of work to be done in matching people in interesting ways. But excitingly, there is a lot of change happening in this space recently.

We have seen the success of dating site businesses such as E-Harmony, OK Cupid, etc. and they appear to be getting somewhat more sophisticated. I recently saw they are now looking to hire in the field of machine learning. There is plenty of work to be done in modeling these matching problems.

And of course the now famous Ok Cupid Hack-tacular.

With this hack the public is now more aware of what kinds of things could be done, well sort of. At least there is more talking about matching algorithms outside of the few of us who are mildly obsessed with the ideas.

Running models and simulations, and having a say in hand crafted personal algorithms, I can imagine a real set of services for dating counseling based on analytics, optimization, statistical evidence, etc. Basically we could tailor make prescriptive models as a premium service within dating site data etc.

In my spare time I have informally offered these types of counseling to friends, family, and colleagues in the past. But have recently sworn off the practice for any work colleagues(lose lose proposition I’ve found—you get them what they wanted(asked for) then they blame you when it gets hard. You don’t find what they want (because it is mathematically impossible) then they think it is you, eg, you can not do it).

But for family and friends whom you can discuss a bit more openly some of the moral/ethical/psychological/motivational aspects of their choices then hmmm, so far it has been ok.  But just one major incident and I will swear that off too, unfortunately.

Methods:

  1. Narrow the known search space: First I usually just walk the seeker through an estimating exercise to get a scope of what the search area is and what the numbers are of what they are seeking. Ie, we take 6 or 7 attributes of individuals that are more fundamental to the search(They usually pick, gender, age, and other predictable indicators) and we cut the planes of all the possible people with those attributes. Ie, start with a pot of 7 billion people, if they will only date women cut in half(rough approximations are fine for this exercise).
    I find this exercise alone adds insight if carefully applied. Usually people know that they are doing this but actually visualizing it on a white board or having to declare it helps them to see which cuts are most limiting or which cuts are most important etc.
  2. Apply the search space to pre-matched Matching:
    Actually gather intelligence from online profiles and reapply the exercise above. Such that the options of matches presented are only matches that fit the prescribed attributes of interest.
  3. Match on More than what 1 party can describe as interesting:
    The above can sometimes solve the matching enough that they find what they are looking for. However, it does not address the je ne sais quoi factors at all. For those it gets a bit more interesting. If one does not know what one wants, it can be argued that there are still ways to find it. Think going to a buffet, when you cannot decide on a single type of food for dinner, or more powerfully, think Amazon predicting that you will order the next John Gresham novel and presenting it to you for purchase the day you finish your current escape reading choice.When I first read e-harmony’s patent many years ago, it was evident that they were not using math(s) the way that I expected. But articles written about E-Harmony’s failings rarely address what we should be doing instead. Because the science of compatibility has not yet been scientifically combined with the science of matching among other issues in this puzzle. Machine Learning, optimization, and psychology are all intricately needed to understand what the probabilities of success are in intimate relationships. Not to mention all the other types of relationships that also need to be successfully managed.

    To illustrate a psychological component of concern; we do not yet have a comprehensive scientifically provable model of personality. The closest that I am aware of is The Big 5. However, the main difference between The Big 5 and its predecessors is that The Big 5 tries not to claim to be more than it is.

    Each of the 5 factors has a subset of traits that correlate with that factor and are distinct from other factors.  That is all it claims.

    Valuable and as huge a step forward as The Big 5 is, it is not alone enough to base all assumptions of personality on in order to identify and prescribe relationship matches.

    Why do we need a theory of personality to find a match of a friend or business partner you ask? Good question.  If you only want to find the person who matches the ideas and traits that you are aware of wanting or think that you need and you are able to list those traits then perhaps you don’t! You can just follow an optimization search method similar to the above 1 and 2, and call it good. And hope that your skills at being in a relationship and your partner’s skills are compatible, and that you both bring to bear what it takes to be successful.. and that you both see that success as successful, and success worthy enough that you both continue to find value in the relationship for as long as it is valuable to you.

    As a systems engineer I would say, that we need a theory of personality in order to inform our models of predictors of behavior and therefore as probabilities of outcomes and events given a system of assumed interactions. Ie, the question is not so much is there a way for me to predict whether or not my business partner will run off with the cash or be able to follow through on projects? The question is more a question of, If, in the scientifically observed history of business relationships were the relationships like mine and this proposed business partner successful and if not shouldn’t I then pick a different partner?  So if the goal is to get people to start a businesses together we may prescribe a different match than if the goal is for people to successful start, grow, and sell off a multi-million $ enterprise.

    The system can begin to be described as the relationships of the actors in a scenario and the outcomes of the goals implied by the scenario.

     

    So depending on the system and the goals and definitions of successfully meeting the objectives implied by the scenario definition then different personalities and different personality mixes may be more useful to generate more optimal or successful outcomes.

    Then identifying a strategy given a single user’s personality assets toward the single user’s goal, likely includes finding complementary assets in another partner’s personality and then gaming. Ie, finding first if that person’s personality is likely to desire or be interested in the goal as defined by the first user. If not, a different set of techniques should be deployed(perhaps looking for a different match that is not as likely to succeed but is more interested in the goal, or choosing a strategy that enrolls the second user in the actions necessary to attain the first user’s goal?)

    The idea of personality describing the way a person is likely to think, value, feel, or act is then very useful in gaming probabilities of successful outcomes based on a goal of another actor in a system and their ‘personality traits’ and corresponding probability matrices.

    But if the theory of personality is fundamentally flawed or our application of it is flawed then likely all of this number 3. can be thrown out.

    As my father said to me once when I wanted to commit to a relationship and asked for his advice,

    “You do not know who you are. You do not know what you want. You do not know who they are. You do not know what they want. You do not know who you will become, nor what you will want. You do not know who they will become, nor what they will want. But if you know this, and are still willing to commit to who they are and who they will become then you have already decided, and I respect your decision”

    This made me smile, because it was true. And it stated his disapproval and support, and the truth of commitment under uncertainty, and implied the responsibility of the actor within the system without acknowledging any specific advantages and disadvantages other than intent and choice, which it so adeptly illustrates as the only thing truly considered and therefore the only thing already decided and thus all that mattered.

  4. Utilize states to describe willingness to engage with good matches. Eg, If the person is attracting their ideal matches for marriage success but are interested in one night stands, or they are attracting their ideal business partner to be able to attract investors, but they really want a drinking budy then they are not appropriately matched for their state. And suggestions should be presented.
  5. Help people meet their goals:
    Reward goal creating behaviors. User experience can affect the outcomes of the goals of the system. Create experiential spaces where we can reward the behaviors that in turn improve the likelihood of successfully reaching the goal or goal state. Eg, There is a lot to be said for the old idea of being introduced by someone you trust, and in online or matching services we often disregard the truth of whether or not this is being done successfully, and we even more rarely apply common sense and group knowledge of best practices let alone rigorous analysis of what works and what does not. Have you ever seen millionaire match maker and watched how if the ‘millionaire’ disrespects the ‘match-maker’ or vice verse how that can effect the plans and type of experience created for the millionaire and their match? Or how the effect from being paraded in front of the millionaires affects the feelings of success and failure of the matches and what they are likely to do in reaction to those feelings in this given scenario.  Ie, if we know that certain scenarios, experiences, introductions, or methods will be experienced by all rational players in a way that negatively impacts the goal seeking behaviors then those patterns should be eliminated from the processes which we create to enable more positive outcomes.

    This goes without saying in most of the situations where we are in control of the systems and understand them well enough and are cable of affecting them. But when in an online application unless you are more of a Chris McKinlay type then you are not likely to understand the systems well enough to manipulate the majority of the presented situations toward more successful outcomes, nor is that even necessarily the most preferred method.

    There are some people in between as evidenced by the Top 4 most attractive on OKCupid.

    Most players likely manipulate the systems in some ways.  This may in fact be successful goal seeking behavior, but in order for it to work the goals need to be aligned with the design of the manipulation and the outcomes which are likely a bit occluded from the user. The OKCupid 4 most attractive showed us how this system manipulation can result in some locally optimal outcomes. Eg, if the goal is defined as attracting more individuals then the OKCupid most desired list is also the OKCupid most successful list. But it is not. Not because the 4  are not very successful, but because more than one of them has stated that just being sought after was not in fact enough to meet their continuing overall goals in matching and they have now adjusted their strategy.

Resources:
Not Sure? Would love to update with any suggested resources…
TED Talk on using an algorithmic type approach to finding love.

Ideas:

Algorithms:
Have you applied dating algorithms we can share a link to here?

If you know of someone who is doing something interesting in this area please leave a comment and I will add them to the list:

 

Giving Voice to the Voiceless -In Math

If the patterns in Motzart’s music could have been detected and understood well enough that we could predict what he would write or close then we could possibly back engineer a platform for  him to write in….If he had had that platform what music would he create? And more importantly what would a platform that could predict and understand his structures be used for by others?

I am working on this idea, so I know this is not clear writing, but I am struggling to find the words to describe this thing that I am wishing to describe.

I really feel that just as machine language may be too difficult for most of use to use in order to write a novel, that the current mathematical languages and programs are out of date in how unwieldy they are to use. I am worried that we are keeping them as a hair shirt or as a misunderstanding of the applications and demand for creating something much easier to interface with.

SO, I am using this post as a place to store my ideas, so as I get better words to describe how, what, why, and who is creating this new platform that we need. I can add this hear.

 

Systems Art, Visualization, and Simulations Topics

This is a work in progress page, and a place to store notes as I build and organize some links and thoughts related to Systems Visualization.

I was unsatisfied with the definitions of Systems Visualization  I found in my short search online.

So for my purposes I define it:
Systems visualization: The process and product of forming mental, representational, and physical constructs of systems

Systems Visualization & Systems Visualizations: http://systemsvisualization.com/page/2

Systems Visualization and Machine Learning: http://www.cc.gatech.edu/~lebanon/pub/tutorialPart1.pdf

 

Systems Engineering and Visualization:

Design of Visuals of Systems:

http://visual.ly/search/node?keys=systems

https://en.wikipedia.org/wiki/Visual_communication

 

Home Analytics

If you work all day on something super smart and get frustrated when you get home that what you implemented at work is not available for you to utilize in your personal life then perhaps you also have an interest in Home Analytics. But if you search the web currently on this subject it takes a bit to find anything of interest. So I hope to compile a list of useful sites and resources on this subject such that we can promote interesting optimization projects in this emerging and interesting field that could use some encouragement from like minded individuals such as us.

Here is my first link of what I hope one day will be a rich resource of Home Analytics sources:

http://www.strategyanalytics.com/default.aspx?mod=saservice&a0=311#0

http://blogs.strategyanalytics.com/SHS/post/2014/01/21/The-Smart-Home-Steals-the-Show-at-the-2014-CES!.aspx

Here is another interesting perspective to add to the list:

http://www.nyi.net/blog/2012/09/personal-analytics-and-the-quantified-life/

http://www.wired.co.uk/news/archive/2011-11/28/martin-blinder-personal-analytics

The Internet of Things:

http://en.wikipedia.org/wiki/Internet_of_Things

Event-driven architecture (EDA) & Service-oriented architecture (SOA):

 

Interesting Resources for Optimization

Open Source:
Utilizing R for Optimization:
CRAN Task View: Optimization and Mathematical Programming

NEOS: NEOS (Network-Enabled Optimization System) Server is a free Internet-based service for solving optimization problems.

NEOS publishes an online optimization guide which is quite thorough and extensive. It is a great resource for students trying to understand context of what type of optimization they are learning and where it fits in relating to other types they may have done before in another course or discipline of computer science, math, and/or engineering:

http://www.neos-guide.org/Optimization-Guide

MOS Open publication resource for the optimization community

Understanding Optimization: A Taxonomy of Optimization Algorithms 

Constraint Programming:

:a problem library for constraints

http://www.csplib.org/#!

 

OR Model Resource:

*Extremely Structured Resource*
http://www.me.utexas.edu/~jensen/ORMM/models/

Article: ‘Begining Prescriptive Analytics with Optimization Modeling’ by Jen Underwood
Simple setup using Excel Addin for a solver
http://www.b-eye-network.com/view/17152

OR Toolkit

 

Software:

 

University Course Pages and Links:

*Topics in Applied Math: Optimmization*
http://www.math.utah.edu/~cherk/teach/opt/course.html
Links:

Institutions and Societies:

Wisconsin Institute for Discover
Hosts of the NEOS server also have interesting events for Optimization: