Solving Signal:Noise Through Adversity And Diversity


There’s a persistent signal:noise problem across the Web and women can do a lot to help solve this issue. We can do it because we’re great at contextualizing info and evaluating it in a way that’s different from what currently exists.

By Twain Liu (Founder, Senseus)

They say entrepreneurs are irrational optimists so I must be one. Who else but an irrational optimist is developing technology that might enable them to discern why consumers are buying their products and not just an app that teaches us how to apply lipstick and match our skin tones?

Who else but an irrational optimist would set out to develop technologies that are all about why consumer buy and could replace 5-stars, thumbs-up-thumbs-down, like and +1 – especially when these are the de facto standards used across all the leading platforms (Amazon, Apple, eBay, Facebook, Google, LinkedIn, TripAdvisor, Twitter and Yelp etc)?

In fact, the underlying algorithms of those rating systems are what act as signals and filters and drive recommendations of everything we engage with online: people, content, products, introductions, comments, links and more.

Who else but an irrational optimist would get a meeting with a Google Ventures principal via this exchange:

Google Ventures: So who’s your target audience?
Me: Early adopters.
Google Ventures: 18-30 year old digital natives?
Me: No, 6-year-old me. The me who’s growing up with the iPad and other mobile devices and doesn’t click “like” buttons or +1 because someone innovated alternatives so I don’t have to.

Later, the feedback from the meeting following my demo was: “I think there’s a lot more to your work which you’re not showing me. You’ve obviously thought about it a lot and put a lot of hard work into it. It’s big in scope and for a lot of people.”

Yes, I’m an irrational optimist and a happy one because even the tiniest validation is a step in a good direction.

I chose to do hard work (data analytics) even when Rational Me could have chosen to create a lifestyle business, wherein my friends and I travel the globe, source handicrafts from indigenous populations and then sell them online.

Instead of doing that, I set out with aims to take on what I consider to be outdated rating-recommendation systems, rethink the maths behind those algorithms and build progressively towards a $100+ million proposition. This high-level objective is possibly the by-product of my banking experiences where my team incubated and financed global trading platforms that process billions of dollars a day, so scales of that magnitude are something I’m au fait with.

I believe there’s talent out there I can collaborate with to change ratings-recommendations for the better of the Web.

Worse than my irrational optimism, so passionate was I about my seedlings of a vision I even overlooked the fact I’d need to re-educate myself about coding and would spent countless days being 5 of 100 females in a developer event (and 2 of the 5 were tech bloggers and 1 was a marketer). Ah, and some of the coding libraries I needed didn’t even exist yet! Coding libraries such as Apple’s Mapping ones which didn’t arrive until this month and Apple’s WWDC, by the way.

Why still make the leaps of imagination and put myself through the pains of realisation? Well………

Solving Signal:Noise Through Adversity And Diversity

There’s a persistent signal:noise problem across the Web and women can do a lot to help solve this issue. We can do it because we’re great at contextualizing info and evaluating it in a way that’s different from what currently exists.

What exists today are legacy systems of value ordering (of content, products, relationships, comments, experiences and more) that’s holding innovation as well as these legacy systems being unable to filter out the deluge of noise in our feeds and in their consequent hit&miss recommendations.

These out-dated systems of value ordering are evident in the way that 5-star, thumbs-up-thumbs-down, hot or not, like / dislike and +1 work.

The developer community knows they’re flawed.

Even YouTube arrived at their own conclusion 5-stars doesn’t work back in September 2009. Yet, if we look at the spate of supposedly “innovative” mobile and Web apps launching in 2012, what methods are they still using to value order things to make recommendations for us? Yes, 5-stars!

This despite Google-YouTube with all their engineering knowhow and million-dollar investment in analytics showing us that 5-stars doesn’t work.

Aren’t we supposed to learn from our mistakes, improve and innovate? Especially once we have empirical evidence? Isn’t that how human intelligence has improved over the millennia?

The problem is that innovation isn’t happening because “group think” within the (predominantly male) developer community results in iterations of the same tools over and over. Yes there are small, incremental improvements (such as using “like” in 2012 instead of “hot or not” which was circa 2000) and bigger scale dissemination and embedding but the rating system itself hasn’t improved significantly.

What’s needed are leaps of imagination, diversity of code, complementary skills and the differentiated way in which women can contribute to solving the same problem.

The fact is that when we click on those 5-star, thumbs-up-thumbs-down, like and +1 buttons, the count in the algorithms go towards an accumulation of size – quantity, popularity, frequency of how many times the button was clicked – but they tell us almost nothing about the quality of what was being voted on or importantly WHY people voted the way they did.

Yes, 5-stars et al are a signal but weak ones that end up being drowned out by other noise.

To compensate for the missing quality, data analysts then try to apply Natural Language filters to try to capture whether we were being negative, neutral or positive in the comments that might have accompanied whatever like / +1 signal we gave.

These Natural Language filters are also often wide of the mark.

Why? Well the reference databases against which these Natural Language filters cross-check are themselves not granular and specific enough. They’re based on something called the Osgood scale of semantic differentials, which is a legacy method from 1950s involving bipolar terms (for example: “Adequate-Inadequate”, “Good-Evil” or “Valuable-Worthless”), and the Likert scale created in 1930s.

We often see Likert in questionnaires and surveys of the type involving:

  1. Strongly disagree
  2. Disagree
  3. Neither agree nor disagree
  4. Agree
  5. Strongly agree

Well, this is the 5-star rating system in another disguise, right? And Google-YouTube already showed 5-stars doesn’t work…

So since the anachronistic days of the Coliseum with their thumbs-up-thumbs-down (like/dislike) and the 1930s with its 5-star gradations, our species hasn’t innovated how we rate and value objects.

Meanwhile, we’ve made breakthroughs and raced ahead in space travel, real-time communications and robotics.

Here’s a big opportunity for some more irrational optimists out there who want to take on the challenge of evolving rating-recommendation systems to take us up to another level of Web and data intelligence.

I’d encourage all female technologists (coders and business executives alike) to enroll on Stanford’s courses:

Join me in cracking some more of what’s possible with ratings-recommendations codes.

Yes, my system is different from 5-stars, thumbs-up-thumbs-down, like and +1. Yes, I am coding it with 6 year-old me in mind. After all, we’re all irrational optimists when we’re 6!

Editor’s note: Got a question or answer for our guest blogger? Leave a message in the comments below.

About the guest blogger: Twain Liu is Founder of Senseus, a mobile technologies company that produces applications for consumer surveys and product-place recommendations. Her experiences include technology development in startups; Strategic Investments and corporate strategy, CEO-Chairman’s Office, Tier 1 bank; and taste development labs, IFF. She is a maths graduate, codes and is multi-lingual. She can be contacted here: twain [@]