By and large, driver search patterns haven’t changed significantly over the years. Their searches ask the same questions. “What am I looking for?” “Where do I want to go?” “What do I want to find near where?” “When will I be going?” “Why did I get a specific result from the search system?” “How can I get there?” “How long will it take?”
Beyond Spatial-Temporal Reasoning
There’s been an evolution in predicting human mobility patterns. Using machine learning and predictive analytics, technology has evolved for connected cloud and embedded search capabilities. Now several additional factors play a part in how we receive accurate and relevant in-car search results—beyond location and time of day.
Most consumers do not understand the nuances behind local search, so I’m going to explain the anatomy of search—the nuts and bolts if you will. Let me begin with what I call the “5 Ws” of local search and how these “5 Ws” can influence search results for drivers.
The 5 Ws: What, Where, When, Why and Who
What— Again, what we search for while on the go hasn’t changed a whole lot. We’re still looking for a place to eat or directions to the closest gas station etc. (as illustrated in below diagram for a typical user). Of course, the level of relevance we get from the search results today is much higher than in years past.
Where— Your location is also pertinent to the type of search results you get. For example, if you decide to catch a matinee after brunch, top search results will be the closest theaters to your vehicle’s location. Also, there is the convenience factor to consider—traffic, wait-time, reservations, discounts, ratings, etc.
When— These search results are based on the time of day and day of the week, week day or weekend. This is mostly implicitly derived and not an explicit. Let’s say you’re low on gas, and you search for the nearest station. Search results will consider whether it’s during commute hours, and there is heavy traffic. Another example would be if you’re looking for the nearest deli. However, the closest one doesn’t appear because maybe they don’t open for another two hours or they’re closed on Sundays.
Why— This helps to get insights with study of actions taken by the driver either explicitly or implicitly with search results. Understanding this “why” part helps to improve relevance and provide effective result for subsequent searches.
Who— This is based on the driving habits of the driver. Has she been to that location before? How often does she visit that location? Would similar results be appropriate? Personalized or collection of user information has its challenges. What if there are multiple drivers? Also, regulations and privacy issues can sway search results.
Now About the How
Earlier I mentioned that what drivers search for hasn’t changed that much for the last two decades. However, what has changed is “how” people can get the information they need. We’ve gone from multibox and twobox search to onebox search with automatic suggestions and voice with conversational. The technology improved from one generation to the next. Unfortunately, many automakers are still using first generation and very few has moved to second-generation technology.
1st Generation (keyword search)— In automobiles, local search required multibox, meaning one box for each field (street, city, state, country). Also, your input must be an exact keyword match or category click, or you will get an error message. These rule-based programs improved eventually to voice multibox and twobox with NVC (next valid character).
2nd Generation (Semantic search)— Consumer expectations had increased significantly with connected services and smart phone disruption. This generation utilizes a hybrid method of rules defined via knowledge map and models. Machine learning algorithms such as classification and regression techniques has been adopted. Technology evolved from first generation to onebox search (with complete flexibility) and auto suggestions with voice, both in connected and embedded mode.
3rd Generation (Predictive search)— The result you need will come to you rather than your requesting it. The right information will come to you at the right time via the right channel. This could be at the start of engine, when planning for a trip, when you explicitly search via text or as part of conversation via voice or guide you as an assistant while on the go. This generation uses deep-learning techniques such as word embedding, natural language understanding (NLU), and location-embedding learning algorithms—among other things. It has smart assistant with complete conversational voice enabled predictive search as well.
A Grip on Reality
What is the future of local search? I believe that augmented reality will play a key role in local search. This is not far from now as humans are visual by nature. It’s not good enough to read, view pictures and see ratings about a great location. We want to experience it with our own eyes. That’s why search will become image or visual-based with augmented reality. For example, you will be able to look at more than just a restaurant’s menu. You will be able to virtually tour the restaurant itself and check out the ambiance before making a reservation.
Here’s another example. Imagine yourself on a busy street in downtown, filled with mom and pop shops and restaurants. You glance the surroundings, and you get to see information about different business establishments. What a "wow" factor it will create.
At Telenav, we are dedicated to exploring new opportunities to revolutionize local search—in hopes of making our everyday lives more fun and productive while on the go. Telenav technology is currently in third generation as we have gone through generation transition in a short period of time. Our vision? We strive to create a delightful, seamless connected and embedded experience for all drivers.
In future articles, I will describe in detail about how we transitioned from first generation to where we are today and where we’re going from here. Also, I will explain the challenges we encountered and how we overcame them.
