Tag Archives: data science

Big Data Forecasting

Forecasting is a foundational activity in analytics and is a fundamental part of everyone’s personal mental calculus. At the simplest level, we live and work constantly using the most basic forecasting assumption – that everything will stay the same. And even though people will throw around aphorisms of the “one constant is change” sort, the assumption that things will stay largely the same is far more often true. The keyword in that sentence, though, is “largely”. Because if things mostly do stay the same, they almost never stay exactly the same. Hence the art and science of forecasting lies in figuring out what will change.

Slide 1 ForecastingBigData
Click here for the 15 minute Video Presentation on Forecasting & Big Data

There are two macro approaches to forecasting: trending and modelling. With trending, we forecast future measurements by projecting trends of past measurements. And because so many trends have significant variation and cyclical behaviors (seasonal, time-of-day, business, geological), trending techniques often incorporate smoothing.

Though trending can often create very reliable forecasts, particularly when smoothed to reduce variation and cycles, there’s one thing it doesn’t do well – it doesn’t handle significant changes to the system dynamics.

When things change, trends can be broken (or accelerated). When you have significant change (or the likelihood of significant change) in a system, then modelling is often a better and more reliable technique for forecasting. Modelling a system is designed to capture an understanding of the true system dynamics.

Suppose our sales have declined for the past 14 months. In a trend, the expectation will be that sales will decline in the 15 month. But if we decide to cut our prices or dramatically increase our marketing budget, that trend may not continue. A model could capture the impact of price or marketing on sales and potentially generate a much better prediction when one of the key system drivers is changed.

This weekend, I added a third video to my series on big data – discussion of the changes to forecasting methodology when using big data.

[I’ve been working this year to build a legitimate YouTube channel on digital analytics. I love doing the videos (webinar’s really since they are just slide-shows with a voice-over), but they are a lot of work. I think they add something that’s different from either a blog or a Powerpoint and I’m definitely hoping to keep knocking them out. So far, I have three video series’ going: one on measuring the digital world, one on digital transformation in the enterprise, and one on big data.]

The new video is a redux of a couple recent speaking gigs – one on big data and predictive analytics and one on big data and forecasting. The video focuses more on the forecasting side of things and it explains how big data concepts impact forecasting – particularly from a modelling perspective.

Like each of my big data videos, it begins with a discussion of what big data is. If you’ve watched (or watch) either of the first two videos in the series (Big Data Beyond the Hype or Big Data and SQL), you don’t need to watch me reprise my definition of big data in the first half of Big Data and Forecasting. Just skip the first eight minutes. If you haven’t, I’d actually encourage you to check out one of those videos first as they provide a deeper dive into the definition of big data and why getting the right definition matters.

In the second half of the video, I walk through how “real” big data impacts forecasting and predictive problems. The video lays out three common big data forecasting scenarios: integrating textual data into prediction and forecasting systems, building forecasts at the individual level and then aggregating those predictions, and pattern-matching IoT and similar types of data sources as a prelude to analysis.

Each of these is interesting in its own right, though I think only the middle case truly adds anything to the discipline of forecasting. Text and IoT type analytics are genuine big data problems that involve significant pattern-matching and that challenge traditional IT and statistical paradigms. But neither really generate new forecasting techniques.

However, building forecasts from individual patterns is a fairly fundamental change in the way forecasts get built. Instead of applying smoothing techniques for building models against aggregated data, big data approaches use individual patterns to generate a forecast for each record (customer/account/etc.). These forecasts can then be added up (or treated probabilistically) to generate macro-forecasts or forecasting ranges.

If you’ve got an interest in big data and forecasting problems, give it a listen. The full video is about 16 minutes split into two pretty equal halves (big data definition, big data forecasting).

The Agile Organization

I’ve been meandering through an extended series on digital transformation: why it’s hard, where things go wrong, and what you need to be able to do to be successful. In this post, I intend to summarize some of that thinking and describe how the large enterprise should organize itself to be good at digital.

Throughout this series, I’ve emphasized the importance of being able to make good decisions in the digital realm. That is, of course, the function of analytics and its my own special concerns when it comes to digital. But there are people who will point out  that decision-making is not the be all and end all of digital excellence. They might suggest that being able to execute is important too.

If you’re a football fan, it’s easy to see the dramatic difference between Peyton Manning – possibly the finest on-field decision-maker in the history of the game – with a good arm and without. It’s one thing to know where to throw the ball on any given play, quite another to be able to get it there accurately. If that wasn’t the case, it’s probably true that many of my readers would be making millions in the NFL!

On the other hand, this divide between decision-making and execution tends to break down if you extend your view to the entire organization. If the GM is doing the job properly, then the decision about which quarterbacks to draft or sign will appropriately balance their physical and decision-making skills. That’s part of what’s involved in good GM decisioning. Meanwhile, the coach has an identical responsibility on a day-to-day basis. A foot injury may limit Peyton to the point where his backup becomes a better option. Then it may heal and the pendulum swings back. The organization makes a series of decisions and if it can make all of those decisions well, then it’s hard to see how execution doesn’t follow along.

If, as an organization, I can make good decisions about the strategy for digital, the technology to run it on, the agencies to build it, the people to optimize it, the way to organize it, and the tactics to drive it, then everything is likely to be pretty good.

Unfortunately, it’s simply not the case that the analytics, organization and capabilities necessary to make good decisions across all these areas are remotely similar. To return to my football analogy, it’s clear that very few organizations are setup to make good decisions in every aspect of their operations. Some organizations excel at particular functions (like game-planning) but are very poor at drafting. Indeed, sometimes success in one-area breeds disaster in another. When a coach like Chip Kelly becomes very successful in his role, there is a tendency for the organization to expand that role so that the coach has increasing control over personnel. This almost always works badly in practice. Even knowing it will work badly doesn’t prevent the problem. Since the coach is so important, it may be that an organization will cede much control over personnel to a successful coach even when everyone (except the coach) believes it’s a bad idea.

If you don’t think similar situations arise constantly in corporate America, you aren’t paying attention.

In my posts in this series, I’ve mapped out the capabilities necessary to give decision-makers the information and capabilities they need to make good decisions about digital experiences. I haven’t touched on (and don’t really intend to touch on) broader themes like deciding who the right people to hire are or what kind of measurement, analysis or knowledge is necessary to make those sorts of meta-decisions.

There are two respects, however, in which I have tried to address at least some of these meta-concerns about execution. First, I’ve described why it is and how it comes to pass that most enterprises don’t use analytics to support strategic decision-making. This seems like a clear miss and a place where thoughtful implementation of good measurement, particularly voice-of-customer measurement of the type I’ve described, should yield high returns.

Second, I took a stab at describing how organizations can think about and work toward building an analytics culture. In these two posts, I argue that most attempts at culture-building approach the problem backwards. The most common culture-building activities in the enterprise are all about “talk”. We talk about diversity. We talk about ethics. We talk about being data-driven in our decision-making. I don’t think this talk adds up to much. I suggest that culture is formed far more through habit than talk; that if an organization wants to build an analytics culture, it needs to find ways to “do” analytics. The word may proceed the deed, but it is only through the force of the deed (good habits) that the word becomes character/culture. This may seem somewhat obvious – no, it is obvious – but people somehow manage to miss the obvious far too often. Those posts don’t just formulate the obvious, they also suggest a set of activities that are particularly efficacious in creating good enterprise habits of decision-making. If you care about enterprise culture and you haven’t already done so, give them a read.

For some folks, however, all these analytics actions miss the key questions. They don’t want to know what the organization should do. They want to know how the organization should work. Who owns digital? Who owns analytics? What lives in a central organization? What lives in a business unit? Is digital a capability or a department?

In the context of the small company, most of these questions aren’t terribly important. In the large enterprise, they mean a lot. But acknowledging that they mean a lot isn’t to suggest that I can answer them – or at least most of them.

I’m skeptical that there is an answer for most of these questions. At least in the abstract, I doubt there is one right organization for digital or one right degree of centralization. I’ve had many conversations with wise folks who recognize that their organizations seem to be in constant motion – swinging like an enormous pendulum between extremes of centralization followed by extremes of decentralization.

Even this peripatetic motion – which can look so irrational from the inside – may make sense. If we assume that centralization and decentralization have distinct advantages, then not only might it be true that changing circumstances might drive a change in the optimal configuration, but it might even be true that swinging the organization from one pole to the other might help capture the benefits of each.

That seems unlikely, but you never know. There is sometimes more logic in the seemingly irrational movements of the crowd than we might first imagine.

Most questions about digital organization are deeply historical. They depend on what type of company you are, in what of market, with what culture and what strategic imperatives. All of which is, of course, Management 101. Obvious stuff that hardly needs to be stated.

However, there are some aspects of digital about which I am willing to be more directive. First, that some balance between centralization and decentralization is essential in analytics. The imperative for centralization is driven by these factors: the need for comparative metrics of success around digital, the need for consistent data collection, the imperatives of the latest generation of highly-complex IT systems, and the need/desire to address customers across the full spectrum of their engagement with the enterprise. Of these, the first and the last are primary. If you don’t need those two, then you may not care about consistent data collection or centralized data systems (this last is debatable).

On the other hand, there are powerful reasons for decentralization of which the biggest is simply that analytics is best done as close to the decision-making as possible. Before the advent of Hadoop, I would have suggested that the vast majority of analytics resources in the digital space be decentralized. Hadoop makes that much harder. The skills are much rarer, the demands for control and governance much higher, and the need for cross-domain expertise much greater in this new world.

That will change. As the open-source analytics stack matures and the market over-rewards skilled practitioners – drawing in more folks, it will become much easier to decentralize again. This isn’t the first time we’ve been down the IT path that goes from centralization to gradual diffusion as technologies become cheaper, easier, and better supported.

At an even more fundamental level than the question of centralization lives the location and nature of digital. Is digital treated as a thing? Is it part of Marketing? Or Operations? Or does each thing have a digital component?

I know I should have more of an opinion about this, but I’m afraid that the right answers seem to me, once again, to be local and historical. In a digital pure-play, to even speak of digital as a thing seems absurd. It’s the core of the company. In a gas company, on the other hand, digital might best be viewed as a customer service channel. In a manufacturer, digital might be a sub-function of brand marketing or, depending on the nature of the digital investment and its importance to the company, a unit unto-itself.

Obviously, one of the huge disadvantages to thinking of digital as a unit unto-itself is how it can then interact correctly with the non-digital functions that share the same purpose. If you have digital customer servicing and non-digital customer servicing, does it really make sense to have one in a digital department and the other as a customer-service department?

There is a case, however, for incubating digital capabilities within a small compact, standalone entity that can protect and nourish the digital investment with a distinct culture and resourcing model. I get that. Ultimately, though, it seems to me that unless digital OWNS an entire function, separating that function across digital and non-digital lines is arbitrary and likely to be ineffective in an omni-channel world.

But here’s the flip side. If you have a single digital property and it shares marketing and customer support functions, how do you allocate real-estate and who gets to determine key things like site structure? I’ve seen organizations where everything but the homepage is owned by somebody and the home page is like Oliver Twist. “Home page for sale, does anybody want one?”

That’s not optimal.

So the more overlap there needs to be between the functions and your digital properties, the more incentive you have to build a purely digital organization.

No matter what structure you pick, there are some trade-offs you’re going to have to live with. That’s part of why there is no magic answer to the right organization.

But far more important than the precise balance you strike around centralization or even where you put digital is the way you organize the core capabilities that belong to digital. Here, the vast majority of enterprises organize along the same general lines. Digital comprises some rough set of capabilities including:

  • IT
  • Creative
  • Marketing
  • Customer
  • UX
  • Analytics
  • Testing
  • VoC

In almost every company I work with, each of these capabilities is instantiated as a separate team. In most organizations, the IT folks are in a completely different reporting structure all the way up. There is no unification till you hit the C-Suite. Often, Marketing and Creative are unified. In some organizations, all of the research functions are unified (VoC, analytics) – sometimes under Customer, sometimes not. UX and Testing can wind up almost anywhere. They typically live under the Marketing department, but they can also live under a Research or Customer function.

None of this, to me, makes any sense.

To do digital well requires a deep integration of these capabilities. What’s more, it requires that these teams work together on a consistent basis. That’s not the way it’s mostly done.

Almost every enterprise I see not only siloes these capabilities, but puts in place budgetary processes that fund each digital asset as a one-time investment and which requires pass-offs between teams.

That’s probably not entirely clear so let me give some concrete examples.

You want to launch a new website. You hire an agency to design the Website. Then your internal IT team builds it. Now the agency goes away. The folks who designed the website no longer have anything to do with it. What’s more, the folks who built it get rotated onto the next project. Sometimes, that’s all that happens. The website just sits there – unimproved. Sometimes the measurement team will now pick it up. Keep in mind that the measurement team almost never had anything to do with the design of the site in the first place. They are just there to report on it. Still, they measure it and if they find some problem, who do they give it to?

Well, maybe they pass it on to the UX team or the testing team. Those teams, neither of which have ever worked with the website or had anything to do with its design are now responsible for implementing changes on it. And, of course, they will be working with developers who had nothing to do with building it.

Meanwhile, on an entirely separate track, the customer team may be designing a broader experience that involves that website. They enlist the VoC team to survey the site’s users and find out what they don’t like about it. Neither team (of course) had anything to do with designing or building the site.

If they come to some conclusion about what they want the site to do, they work with another(!) team of developers to implement their changes. That these changes may be at cross-purposes to the UX team’s changes or the original design intent is neither here nor there.

Does any of this make sense?

If you take continuous improvement to heart (and you should because it is the key to digital excellence), you need to realize that almost everything about the way your digital organization functions is wrong. You budget wrong and you organize wrong.

[Check out my relatively short (20 min) video on digital transformation and analytics organization – it’s the perfect medium for distributing this message through your enterprise!]

Here’s my simple rule about building digital assets. If it’s worth doing, it’s worth improving. Nothing you build will ever be right the first time. Accept that. Embrace it. That means you budget digital teams to build AND improve something. Those teams don’t go away. They don’t rotate. And they include ALL of the capabilities you need to successfully deliver digital experiences. Your developers don’t rotate off, your designers don’t go away, your VoC folks aren’t living in a parallel universe.

When you do things this way, you embody a commitment to continuous improvement deeply into your core organizational processes. It almost forces you to do it right. All those folks in IT and creative will demand analytics and tests to run or they won’t have anything to do.

That’s a good thing.

This type of vertical integration of digital capabilities is far, far more important than the balance around centralization or even the home for digital. Yet it gets far less attention in most enterprise strategic discussions.

The existence or lack of this vertical integration is the single most important factor in driving analytics into digital. Do it right, and you’ll do it well. Do what everyone else does and…well…it won’t be so good.

Measuring the Digital World – The Movie!

I’ve put together a short 20 minute video that’s a companion piece to Measuring the Digital World. It’s a guided tour through the core principles of digital analytics and a really nice introduction to the book and the field:

Measuring the Digital World : Introduction

Measuring the Digital World

An Introduction to Digital Analytics

The video introduces the unique challenges of measuring the digital world. It’s a world where none of our traditional measurement categories and concepts apply. And it doesn’t help that our tools mostly point us in the wrong direction – introducing measurement categories that are unhelpful or misleading. To measure the digital world, we need to understand customer experiences not Websites. That isn’t easy when all you know is what web pages people looked at!

But it’s precisely that leap – from consumption to intent – that underlies all digital measurement. The video borrows an example from the book (Conan the Librarian) to show how this works and why it can be powerful. This leads directly to the concepts of 2-Tiered segmentation that are central to MTDW and are the foundation of good digital measurement.

Of course, it’s not that easy. Not only is making the inference from consumption to intent hard, it’s constantly undermined by the nature of digital properties. Their limited real-estate and strong structural elements – designed to force visitors in particular directions – make it risky to assume that people viewed what they were most interested in.

This essential contradiction between the two most fundamental principles of digital analytics is what makes our discipline so hard and (also) so interesting.

Finally, the video introduces the big data story and the ways that digital data – and making the leap from consumption to intent – challenges many of our traditional IT paradigms (not to mention our supposedly purpose-built digital analytics toolkit).

Give it a look. Even if you’re an experience practitioner I think you’ll find parts of it illuminating. And if you’re new to the field or a consumer of digital reporting and analytics, I don’t think you could spend a more productive 20 minutes.

Afterward (when you want to order the book), here’s the link to it on Amazon!

Matching (and Scoring) Content to Culture and Predicting the Academy Awards

Thoughts and Reflections on the Process

We’ve spent our spare time in the last six weeks participating in the 538 Academy Awards Prediction Challenge. On Sunday, we’ll find out how we did. But even though we expect to crash and burn on the acting awards and are probably no better than 1-3 in a very close movie race, we ended up quite satisfied with our unique process and the model that emerged. You can get full and deep description of our culture matching model with it’s combination of linguistic analysis and machine learning in this previous post.

What I love about projects like this is that they give people a glimpse into how analytics actually works. Analysis doesn’t get made at all the way people think and in most cases there is far more human intuition and direction than people realize or that anyone reading screeds on big data and predictive analytics would believe. Our culture-matching analysis pushes the envelope more than most we do in the for-pay world, so it’s probably an exaggerated case. But think about the places where this analysis relied on human judgment:

  1. Deciding on the overall approach: Obviously, the approach was pretty much created whole-cloth. What’s more, we lacked any data to show that culture matching might be an effective technique for predicting the Oscars. We may have used some machine learning, but this approach didn’t and wouldn’t have come from throwing a lot of data into a machine learning system.
  2. Choosing potentially relevant corpora for Hollywood and each movie: This process was wholly subjective in the initial selection of possible corpora, was partly driven by practical concerns (ease of access to archival stories), and was largely subjective in the analyst review stage. In addition to selecting our sources, we further rejected categories like “local”, “crime” and “sports”. Might we have chosen otherwise? Certainly. In some cases, we tuned the corpora by running the full analysis and judging whether the themes were interesting. That may be circular, but it’s not wrong. Nearly every complex analysis has elements of circularity.
  3. Tuning themes: Our corpora had both obvious and subtle biases. To get crisp themes, we had to eliminate words we thought were too common or were used in different senses. I’m pretty confident we missed lots of these. I hope we caught most. Maybe we eliminated something important. Likely, we’ll never know.
  4. Choosing our model: If you only do 1 model, you don’t have this issue. But when you have multiple models it’s not always easy to tell which one is better. With more time and more data, we could try each approach against past years. But lots of analytic techniques don’t even generate predictions (clustering, for example). The analyst has to decide which clustering scheme looks better, and the answer isn’t always obvious. Even within a single approach (text analytics/linguistics), we generated two predictions based on which direction we used to match themes. Which one was better? That was a topic of considerable internal debate with no “right” answer except to test against the real-world (which in this case will be a very long test).
  5. Deciding on Black-Box Validity: This one is surprisingly hard. When you have a black-box system, you generally rely on being able to measure it’s predictions against a set of fairly well known decisions before you apply it to the real-world. We didn’t have that and it was HARD to decide how and whether our brute force machine-learning system was working at all. But even in cases where external measurement comparisons exist, it’s the unexpected predictions that cause political problems with analytics adoption. If you’ve ever tried to convince a skeptical organization that a black-box result is right, you know how hard this.
  6. Explaining the model: There’s an old saying in philosophy (from James) that a difference that makes no difference is no difference. If a model has an interesting result but nobody believes it, does it matter? A big part of how interesting, important and valid we think a model is comes from how well it’s explained.

This long litany is why, in the end, the quality of your analysis is always about the quality of your people. We had access to some great tools (Sysomos, Boilerpipe, Java, SPSS, R and Crimson Hexagon), but interesting approaches and interesting results don’t come from tools.

That being said, I can’t resist special call-outs to Boilerpipe which did a really nice job of text extraction and SPSS Text Analytics which did a great job facilitating our thematic analysis and matching.

 

Thoughts on the Method and Results

So is culture matching a good way to predict the Oscars?

It might be a useful variable but I’m sure it’s not a complete prediction system. That’s really no different that we hoped going into this exercise. And we’ll learn a little (but not much) more on Awards night. It would be better if we got the full vote to see how close our rank ordering was.

Either way, the culture-matching approach is promising as a technique. Looking through the results, I’m confident that it passes the analyst sniff test – there’s something real here. There are a number of extensions to the system we haven’t (and probably won’t) try – at least for this little challenge. We’d like to incorporate sentiment around themes, not just matching. We generated a number of analyst-driven cultural dimensions for machine training that we haven’t used. We’d like to try some different machine-learning techniques that might be better suited to our source material. There is a great deal of taxonomic tuning around themes that might drive better results. It’s rare that an ambitious analytics project is every really finished, though the world often says otherwise.

In this case, I was pleased with the themes we were able to extract by movie. A little less with the themes in our Hollywood corpus. Why? I suspect because long-form movie reviews are unusually rich in elaborating the types of cultural themes we were interested in. In addition, a lot of the themes that we pulled out of the culture corpus are topical. It’s (kind of) interesting to know that terrorism or the presidential campaign were hot topics this last year, but that isn’t the type of theme we’re looking for. I’m particularly interested in whether and how successful we can be in deepening themes beyond the obvious one. Themes around race, inequality and wealth are fairly easy to pick out. But if the Martian scores poorly because Hollywood isn’t much about engineering and science (and I’m pretty sure that’s true), what about its human themes around exploration, courage and loneliness? Those topics emerged as key themes from the movie reviews, but they are hard to discover in the Hollywood corpus. That might be because they aren’t very important in the culture – that’s certainly plausible – but it also seems possible that our analysis wasn’t rich enough to find their implicit representations.

Regardless, I’m happy with the outcome. It seems clear to me that this type of culture matching can be successful and brings analytic rigor to a topic that is otherwise mostly hot-air. What’s more it can be successful in a reasonable timeframe and for a reasonable amount of money (which is critical for non-academic use-cases). From start to finish, we spent about four weeks on this problem – and while we had a large team, it was all part-timers.

This was definitely a problem to fall in love with and we’d kill to do more, expand the method, and prove it out on more substantial and testable data. If you have a potential use for culture matching, give us a call. We probably can’t do it for free, but we will do if for less than cost. And, of course, if you just need an incredible team of analysts who can dream up a creative solution to a hard, real-world problem, pull data from almost anything, bring to bear world-class tools across traditional stats, machine-learning and text analytics, and deliver interesting and useful results…well, that’s fine too.

 

Torture is Bad – Don’t Waterboard your Models even when you know they are Wrong

Predicting the Best Actor and Actress Categories

My Analytics Counseling Family here at EY has been participating in the 538 Academy Award Challenge. Our project involved creating a culture-matching engine – a way to look at pieces of content (in this case, obviously, movies) and determine how well they match a specific community’s worldview. The hypothesis is that the more a movie matches the current Hollywood zeitgeist, the more likely it I to win. In my last post, I described in some detail the way we did that and our results for predicting the Best Movie (The Big Short). We were pretty happy with the way the model worked and the intuitive fit between the movies and our culture-matching engine. Of course, nothing in what we’ve done proves that culture matching is a great way to predict the Oscars (and even if we’re right it won’t prove much in a single year), but that wasn’t really the point. Culture-matching is a general technique with interesting analytics method and if the results are promising in terms of our ability to make a match, we think that’s pretty great.

The second part of our task, however, was to predict the Best Actor and Actress awards. Our method for doing this was similar to our method for predicting the best movie award but there were a few wrinkles. First, we extracted language specific to each character in the nominated movie. This is important to understand. We aren’t looking at how Hollywood talks about DiCaprio or Cranston or Lawrence as people and actors. We aren’t looking at how they are reviewed. We’re entirely focused on how their character is described.

This is the closest analogue we could think of to culture matching movies. However, this was a point of considerable debate internal to our team. To me, it seems intuitively less likely that people will prefer an actor or actress because their character matches our worldview than when discussing a movie as a whole. We all understood that and agreed that our approach was less compelling when it came to ANY of the secondary awards. However, our goal was to focus on culture-matching more than it was to find the best method for predicting acting awards. We could have predicted screenplay, I suppose, but there’s no reason to think the analysis would deviate in the slightest from our prediction around movie.

Once we had key themes around each nominated role, we matched those themes to our Hollywood corpus. In our first go round, we matched to the entire corpus matching actor themes to broad cultural themes. This didn’t work well. It turned out that we were conflating themes about people with themes about other things in ways that didn’t make much sense. So for our second pass, we tightened the themes in the Hollywood corpus to only those which were associated with people.

In essence, we’re saying which roles best correspond to the way Hollywood talks about people and picking the actor/actress who played that role.

So here’s how it came out:

RankActor
1Bryan Cranston
2Michael Fassbender
3Leonardo DiCaprio
4Eddie Redmayne
5Matt Damon

And

RankActress
1Jennifer Lawrence
2Brie Larson
3Cate Blanchett
4Saoirse Ronan
5Charlotte Rampling

 

Do I think we’re going to be right? Not a chance.

But that doesn’t mean the method isn’t working pretty well. In fact, I think it worked about as well as we could have hoped. Here, for example, are the themes we extracted for some of the key actors and actresses (by which I mean their nominated roles):

For Matt Damon in the Martian: Humor, Optimism, Engineer, Scientist, and leadership.

For Leonardo DiCaprio in the Revenant: Survival, Endurance, Tragedy, Individual, Unrelenting, Warrior, Physicality

For Bryan Cranston in Trumbo: Idealist, humanity, drinking, liberal, civil rights

If you’ve seen these movies, I think you can agree that the thematic pulls are reasonable. And is it any surprise, as you read the list, that Cranston is our predicted winner? I think not. To me, this says more about whether our method is applicable to this kind of prediction – and the answer is probably not – than whether the method itself is working well. Take away what we know about the actors and the process, and I think you’d probably agree that the model has done the best possible job of culture matching to Hollywood.

I was a bit concerned about the Jennifer Lawrence prediction. I saw the logic of Cranston’s character immediately, but Joy didn’t immediately strike me as an obvious fit to Hollywood’s view of people. When I studied the themes that emerged around her character, though, I thought it made reasonable sense:

Lawrence in Joy: Forceful, personality, imagination, friendship, heroine

WDYT? There are other themes I might have expected to emerge that didn’t, but these seem like a fairly decent set and you can see where something like forceful, in particular, might match well (it did).

In the end, it didn’t make me think the model was broken.

We tried tuning these models, but while different predictions can be forced from the model, nothing we did convinced us that, when it came to culture matching, we’d really improved our result. When you start torturing your model to get the conclusions you think are right, it’s probably time to stop.

It’s all about understanding two critical items: what your model is for and whether or not you think the prediction could be better. In this case, we never expected our model to be able to predict the Academy Awards exactly. If we understand why our prediction isn’t aligned to likely outcomes, that may well be good enough. And, of course, even the best model won’t predict most events with anything like 100% accuracy. If you try too hard to fit your model to the data or – even worse – to your expectations, you remove the value of having a model in the first place.

Just like in the real world, with enough pain you can make your model say anything. That doesn’t make it reliable.

So we’re going down with this particular ship!

 

Machine Learning

We’ve been experimenting with a second method that focuses on machine learning. Essentially, we’re training a machine learning system with reviews about each movie and then categorizing the Hollywood corpus and seeing which movie gets the most hits. Unfortunately, real work has gotten in the way of some our brute-force machine learning work and we haven’t progressed as much on this as we hoped.

To date, it hasn’t done a great job. Well, that’s being kind. Really it kind of sucks. Our results look pretty random and where we’ve been able to understand the non-random results, they haven’t captured real themes but only passing similarities (like a tendency to mention New York). With all due respect to Ted Cruz, we don’t think that’s a good enough cultural theme to hang our hat on.

As of right now, our best conclusion is that the method doesn’t work well.

We probably won’t have time to push this work further, but right now I’d say that if I was doing this work again I’d concentrate on the linguistic approach. I think our documents were too long and complex and our themes too abstract to work well with the machine learning systems we were using.

In my next post, I have some reflections on the process and it what it tells us about how analytics works.