Get Your Face Read!

  • //” alt=”Facial Expressions” title=”affectiva-faces” width=”540″ height=”144″ class=”aligncenter size-full wp-image-5024″ />
  • Back in November, I mentioned that //” title=”Affectiva”>Affectiva, a firm in our neuromarketing companies list, was working on doing facial expression analysis using the webcams connected to most computers. Now, if you want to see how this works, you can watch a selection of recent Super Bowl ads with your camera turned on – you’ll see how your expressions stack up to those of other viewers.
  • Here’s what mine looked like for a Chevy ad
  • //×372.jpg” alt=”Affectiva results” title=”affectiva-results” width=”540″ height=”372″ class=”aligncenter size-large wp-image-5021″ />
  • The system reports on a variety of metrics, including “smile” and the more cryptic “valence.” According to Affectiva,
  • Bayesian machine learning processes are used to combine the facial and head movements in order to recognize positive and negative displays of emotion as well as complex states such as interest and confusion.

  • //” alt=”Affectiva Plot” title=”affectiva-plot” width=”540″ height=”170″ class=”aligncenter size-full wp-image-5022″ />
  • The firm strikes a cautious note in describing the limitations of the technique
  • Although some people claim that their face-reading technologies “recognize emotions,” it’s important to note that the state of the technology is such that it recognizes outward expressions, which may or may not correspond to true feelings.
  • So, //” title=”Neuromarketing”>Neuromarketing readers, here’s your assignment:
    1. Go to the //” title=”Face Reading Demo” target=”_blank”>Affectiva face-reading demo and view an ad or two with your webcam on.
    2. Come back here and let everyone know how it worked for you.
  • I’m not sure how I felt about the accuracy of my data. At one point, for example, I found the volume was too low and had to hunt for the controls; I can’t imagine that my expressions at that point had much to do with the ad. Plus, I’m not sure I emote very much when watching a typical ad. On the other hand, if you use a large enough sample, presumably interruptions and irrelevant expressions affecting one subject will be canceled out by the mass of data. Being a web-based solution that uses standard user equipment, scalability should be robust. Let us know what you think!

  1. Monica says

    Really interesting! Here are my results for the VW ad
    Two moments are very accurate: I enjoyed seeing the dog shaking after he came out of the water, and I disliked the “intergalactic superstar”.

  2. Roger Dooley says

    Interesting, Monica. I’m kind of surprised the “confusion” metric didn’t spike when it cut to the Star Wars bar scene. I’ve heard quite a few comments from people who didn’t get the inside joke referencing last year’s popular Super Bowl ad from VW.


  3. Monica says

    Well, I’m a big fan of Volkswagen, I know most their ads and I do know very well last’s year ad from Super Bowl, as it scored first. I will also send the link to my friends to use this tool. Thank you so much for sharing this!

  4. Brad Einarsen says

    Very interesting! My results for the Doritos commercial:

  5. Thanks for trying Affectiva’s Emote Your Vote demo and for sharing your observations. I wanted to comment on your self track — you’re relative lack of expression isn’t a problem at all. In ad testing, even measuring lack of expression provides valuable information. For instance, if a decision-maker believes a scene is hilarious but the data doesn’t support it, it’s better to know it before an ad goes live. Neutral expressions can also convey important info related to attention: one of Affectiva’s key measures. If you watch the entire ad, even with a neutral expression, that’s different information than if you turn away before then end. Finally, your point about scale is spot on. All these patterns become even more meaningful on a quantitative basis , which is why we’re excited to be able to offer this over the web.

    1. Roger Dooley says

      That makes sens, Avril. I must admit, though, I am capable of finding something hilarious while maintaining an exterior poker face. 🙂


  6. Jim Jenks says

    Technology just keeps on amazing me. I wonder how many companies are using this technology when creating their commercials? I feel like some companies keep going more and more downhill with their commercials.

  7. Giancarlo Mirmillo says

    I think it is a methodological error to try to understand if the feelings experienced while watching a video correspond to those detected by the system.
    Feelings and emotions are at two very different levels
    The basic emotions, represented by facial expressions, are not conscious and can not be, this is one of the most important lessons of neuroscience.
    I think this is the real value of the coding systems of facial expressions, detect small changes in emotional expression, so influential on mood, memory, attitudes and opinions regarding (among other things) to brands or products.
    The feelings are consciously perceived as a complex set of appraisal, forecasts, knowledge, importance of goals. Not all of these factors are brought into play in the vision of a advertivement which is normally viewed with low levels of cognitive processing.
    So it’s great a system that allows the feedback of something “unspeakable” but do not try to see if it’s true because the two levels, conscious and unconscious, are not comparable.

    1. Roger Dooley says

      Good point, Giancarlo. The key advantage of facial coding is to try to determine non-conscious reactions. In one of his books, Dan Hill describes a project in which subjects’ faces showed a negative emotion when a new appliance feature was described, even though they said they liked the idea. Hill interpreted this dichotomy as indicating that at some level the consumers were put off by the added complexity of the new feature.


  8. Elena Anne says

    Very interesting Roger!
    I tried the demo and it was funny. My expressions are very intense, and always someone understands how I feel. Family Guy is one of best shows. So, the smile, valence and attention reached the top! Surprise was not very variable, because I knew what to expect from Brian and Stewie 🙂

    1. Roger Dooley says

      Interesting, Elena. Some of us are much more expressive. Yesterday, I saw a young woman reading an email on her phone, and several times she smiled broadly when she read something amusing. She would be a great subject for this kind of study. Others, like me, don’t show much emotion outside of a social setting.


  9. Mikhail Baev says

    It’s very interesting but I think that it would be great to compare Affectiva results with the alternatives — Nviso and Noldus FaceReader. I think that now we have one problem with this technology — the inability to measure difference between reaction on something in general and local reaction. For example I don’t like Chevy but like its advertisement.

    1. Roger Dooley says

      Mikhail, I think that’s the crux of most neuromarketing analysis: even when we can measure emotion, or brain activation, or something else with considerable accuracy, it’s a big leap to predicting actual behavior. An ad that makes me smile may signify I like the product. Or, it could just be a funny ad.


  10. Giancarlo Mirmillo says

    Mikhail this problem is due to a fundamental mistake: neuroscience research are very interesting and essential to understand emotions, perceptions, and other parameters, but the methods used by many nueuromarketing companies are derived from medical research and not from social research.
    Social researchers have for years learned how to build and evaluate experiments to identify the objects of study effectively.
    I think Roger’s approach of using the discoveries of neuroscience but not their techniques is in the right direction.

    1. Mikhail Baev says

      Giancarlo, I agree with you about Roger’s approach. We need some common sense in science data interpretation. It would be more effective not to search for some kind of specific marker (like p300 braine wave) but to see the whole thing.

Leave A Reply

Your email address will not be published.