What does an effective content strategy look like, when specifically geared to take advantage of semantic search? The following is a presentation I gave for the fine folks at the DC, VA & MD Search Engine Marketing Group on January 16, 2014. Since slides only tell half the story, I’ll try to replicate as much of the talk as I can in this post. Also a big thanks to Haiku Deck for having such an awesome product.
There are 5 main buckets of semantic technology, each specific to a type of content. As a result, there are 5 new hats which digital marketers should try on. Let’s dive in.
Become a Linguist
Natural Language Processing & Speech Analysis
I stumbled onto a company called Veveo, who was recently awarded patents for speech-based interfaces. One of their studies found 3 of 4 people are dissatisfied with their TV content discovery experience. The top reasons were: the inability to search using keywords, not knowing how to spell what they’re looking for, it takes too long to scroll through the electronic guide, and the finding the recommendations irrelevant. As a result, Veveo set out to develop conversational interfaces which can be used with phones, smart TVs, set-top boxes and other devices. This technology aloows you to simply talk to your TV or device (making it no longer something only crazy people do). Here is a demo of Veveo:
Google already has speech-based interfaces on their mobile and desktop search applications, so this is something to take advantage of today. But why mention small technology companies? Because if a small company can do it, so can Google. If patent restrictions are a barrier, Google can just buy the company outright. The Knowledge Graph sprang from their acquisition of Freebase (Metaweb), and just this week Google bought $3B worth of thermostats. So, there’s that.
To be found through speech and conversational interfaces, don’t rely on the technology to get you there. Make it work for you by researching how your prospects speak and incorporate this language in your content. Keep in mind this isn’t always the same as keyword research. Get creative with how you collect this data.
Become A Photographer
Semantic Image Processing
When using images on your site, be aware of everything that goes into these photos. Composition, quality, uniqueness and metadata all play a role in how well the meaning can be extracted from them. For one, make sure there is a clear subject, ideally with an action or another object showing a relationship with the subject. Make sure there is enough contrast in the image to facilitate background detection – no washed out Instagram filters.
Uniqueness is especially important. Although Matt Cutts said stock photography won’t hurt you, I think it’s a terrible idea to use them both for search and for UX. Images are another opportunity to be unique, and uniqueness is always an indicator of quality. We already know Google can find duplicate images across the web, so why wouldn’t they be using it to judge your content? (Here’s my favorite example – 231,000 results for just 1 image). Google and Bing have both vowed to rid the world of child pornography, likely with this same semantic image processing technology.
Success in image search means more traffic, because it can help your other content become visible. A great way to facilitate successful image semantics is with schema.org structured data for images. With this you can provide a description of the image, URL of where it came from (to suggest ownership/originality) and even an author of the image.
Become a Movie Director
Semantic Video Processing
Although it used thumbnail stills from videos, Google’s X lab accomplished object recognition involving artificial intelligence and machine learning. The computer was given 10 million random images from YouTube videos and, without first showing it was to look for, could identify cats with a 75% success rate.
Thinkglue LTD is another small company with big technology. Their video analysis software can handle sentiment analysis, video summarization, object detection, and even searching within videos. In fact they’re so good at it, they just invented a wifi router than can block inappropriate videos based on the content alone. So rather than by blocking access to specific websites (and the constant lag in updating lists of these domains) the router simply “watches” the videos and determines if they’re acceptable for your family.
Now that videos also serve as a medium for relevance, we also need to step up and get involved with their production. Be a marketer in the director’s chair by satisfying need states, speaking to sales cycle stages, and using the language of your audience. Make a point in your videos- no more shots of the CEO talking at a desk the whole time. Use visual indications of what your company actually does and the value it provides.
In December, Schema.org released new markup specifically for TV and Radio shows. Classes include Series, Season, Episode, Clip, etc. Although this is for programs, I’m sure it is only the beginning of structured data for video.
Prediction (aka I’m making this up)
Considering semantic video technology combined with schema markup for TV shows, it’s a huge opportunity for media producers to get their shows found more easily. But for marketers, it’s not so much as what’s in the shows so much as what’s between the shows. It’s not hard to imagine this semantic processing of episodes to allow for “smart” commercials, specifically matched to the topic or subject of the show during which they’re played. In an environment where Google Fiber delivers both TV and Internet service, advertisers would easily be able to target their commercials with greater accuracy. Using both web search history and TV viewing history, messages can be delivered to the right audience. Like the Google Display Network, I can easily foresee a video ad network in which you could upload your video commercial and Google would not only automatically determine the content of your commercial (think video quality score) but also serve it up in relevant contexts. Chromecast or Commercialcast?
Become a Fortune Teller
The ability to foresee information needs is at the center of Google Now, enabling you to be there before your prospects are even looking for you. This type of experience is part of branding, as is any interaction with a company be it predictive or otherwise.
On the technical side, predictive search can be facilitated with structured data. What many people don’t know is that structured data can be used in more than just web pages. For example if you make a reservation with OpenTable the confirmation email is complete with structured data. This data can then be used by apps to help you get to your reservation on time. Zillow uses Google Now cards to suggest new houses on the market which might be suitable for you, based on your search history. Structured data can also be used in social media, which all have the potential to be used in predictive search.
On the content side, the trick is to not only serve up what your prospects are looking for, but what they’ll look for next. Use topic mapping to visualize the entire information-seeking experience, using semantic relationships to help determine subsequent searches. The goal is to anticipate search activity, and provide excellent content for each of these steps. Use a tool like ConceptNet 5 or other linked data networks to discover these semantic relationships.
Become a Psychologist
Finally, the only thing robots can’t have (yet) is feelings. Emotions are an important aspect of the search process, since they can have a profound effect on the type of satisfactory search results. Sentiment analysis has been part of social media monitoring for a while, but emotional states also provide context in regard to search intent.
One company doing this well is Semantria, who makes sentiment analysis easy as pie. Not only does the APi work with social media channels, but also Gmail, Google Docs, Evernote, WordPress and Zendesk (customer service). Plus, if you have any other data source their Excel plugin conditionally formats cells based on sentiment.
Emotion Markup Language 1.0 (EmotionML) was formally proposed by the Wc3 this past April. EML is proposed for use in manual annotation of data (think structured data), automatic recognition of emotional states from user behavior (think customer service interaction), and the generation of emotion-related system behavior. This last one is interesting, as it suggests different computer outputs based on human emotional input. If emotion markup language becomes prevalent in content discovery, it will be necessary to consider emotional states for effective content strategies. Even if search engines do not adopt sentiment analysis, your content will be much more effective if it is taken into consideration just like we do for device types or locations.
Wear More Hats
For a successful semantic content strategy, consider the technology and media formats which help deliver your message. Become a linguist, photographer, movie director, fortune teller, and psychologist.