Technology, combined with social networks have the power to stop oppression and improve people’s lives.
We are already seeing the impact of Facebook, Twitter, Google+ and Linkedin in the Middle East, and the media industry, but we believe this change is just the beginning. Other industries like Finance, Education and Healthcare may be nearing their Napster moment.
At my Start-up, Lenddo.com we are using the social graph to improve people’s lives and restructure finance. Members are using their online reputation to prove identity and access credit in regions where there is little access to credit. Every day members are able to get education for loved ones, medicine for family members, and shelter for relatives. Tangible life improvements. We expect to impact millions.
The other day I was asked about the predictive capabilities of algorithms, so I thought some background information might be helpful.
We like to think of ”trustworthiness” algorithms as falling into three broad categories.
Bayesian – This is classic machine learning (pattern matching) to build a probabilistic model, within a acceptable false positive, false negative range. The core to growing these predictors is massive volumes of data. Data could include any information collected or observed. This can include; stated interests, sites visited, biometrics, writing style, age, words uses, images, network composition, machine configuration, geographic data, internet usage, job, demographics, education, hobbies, and financial history. The Adtech industry has built a substantial “behavioral profiling” infrastructure to predict what ads are most appropriate (sell you sugar water), why not leverage and extend this technology to actually empower individuals with access to credit?
Validators – (Heuristic, Taxonomical, or Bayesian) – these algorithms are used to validate or perform a reasonableness test for information volunteered by an applicant or her friend. For example: someone from a big company will generally have friends at that company, or someone from Manila will generally have log-ins from Manila, or someone in medicine will use certain words in their general correspondence. Validators are very important in the Emerging Markets, where standard RAC (Risk Assessment Criteria) rules are necessary but insufficient due to a general shortage of customer data.
Homophily – (“birds of a feather flock together”) builds on the well documented tendency for individuals who interact regularly, or share some social bonding to behave similarly. In science, many of the best Homophily data scientist come out of computational psychology or epidemiology. In business, Online Ad networks (Demand Side Platforms) are a good place to hire as Homophily based algorithms are often the basis of their mathematical underpinnings.
More on Homophily
A good example of Homophily in nature is with smoking, game fraud, sexual promiscuity and obesity. For example there is some great studies and talks by Harvard health researcher Dr. Nicholas Christakis.
His team found in a social network of 12,000 over 32 years, that friends had the greatest impact on becoming obese and that the type of friendship made a difference. For example, a close, mutual friendship resulted in a 171-percent increased risk of obesity while a one-way friendship increased the chances by 57 percent. These findings contrast with a 40-percent increased risk among siblings and 37 percent between spouses. Friendships are so influential, in fact, that the likelihood of obesity even increased when the friend of a friend was the one to gain weight. Same gender relationships also have a stronger influence than those between genders.
Perhaps most surprising was the finding that geographical closeness made no difference. This last fact is interesting to us, because we believe geographic clossness will become even less important as people increasingly use the internet to maintain relationships. We are able to use the social graph to measure “closeness” and “friendship strength” as well as of course “geospatial closeness”.
I Hope this is helpful. Predictive Algorithms and important part of what makes Lenddo so simple for our members.
If you have an interest in using math to makes people’s lives simpler and better, please lets us know. At Lenddo.com we are trying to make a dent in the universe and always looking for exceptional Data Scientists.
Post too long? get the 140 character version https://twitter.com/UrgentSpeed
Image source Forbes