The Soleo Discovery Engine
The technology leverages Soleo’s natural language understanding-based knowledge graph, and its complementary response generation capability, to engage a user in a conversation about the services they are seeking. As the application begins to understand the user’s intent, increasingly precise alternatives are offered to the user in conversational form. The system then applies machine learning to make more targeted offers to future users with similar intent.
There are several key technologies applied in the Discovery Engine integrating and conducting the conversational flow. Soleo has employed deep machine learning in several areas to increase our understanding of user intent and to analyze, cleanse, and organize the data related to local businesses:
- Neural networks and machine learning are leveraged to build several language models which are then integrated into Soleo’s knowledge graph related to location, time, business name, and products and services. These models and graph are built from business-related data mined from the Internet and from other public and private sources. The models allow for artificially intelligent understanding of user intent and selection of natural-feeling responses.
- Soleo similarly uses neural networks and other data science techniques to groom business listings data from numerous sources to create a consolidated and comprehensive database of business information.
- In the case of Call Transfer Services, knowledge about the history of phone numbers teaches our models about what future callers to that number are seeking (intent). In other words, we leverage a wealth of existing information, gleaned from years of stored query transactions, to maximize our understanding of the context for any given call before the first exchange with the caller.
- Finally, Soleo applies the knowledge graph in real time to intelligently and automatically select sponsored merchants from dozens of verified content partners.
Let’s take the example of a user interacting via voice on a telephone call. These sophisticated models are useless if the caller hangs up or is confused. Soleo, therefore, must gracefully and succinctly ask the caller “What can I help you find?” We use several real-time A/B tests to maximize the caller engagement with that first prompt. We leverage state-of-the-art Text-To-Speech (TTS) to create a natural sounding conversational prompt to the caller. We then apply artificially intelligent Speech-To-Text (STT) algorithms to maximize our recognition of the caller response.
Now that we have the first utterance from the caller in hand, we apply the knowledge graph. At its core, we are looking for caller intent. We will also recognize search modifiers such as time and location. The knowledge graph generates one or more dialogs representing our first pass at understanding what was said:
- Was the caller seeking a business by name?
- Is it a closed business? Is it a chain store? Is it a well-known business that is simply outside of the caller’s search radius?
- Was the caller speaking a more natural, conversational query, such as “I need to find someone to help me with a termite problem”?
With an understanding of what the caller is seeking, paired with our understanding of what businesses exist within the local area, we generate natural, conversational responses back to the caller. For example –
- “I found the XYZ Business on Main Street for you. Stay on the line to be connected.”
- “The closest XYZ Business is 35 miles away. Would you like to be connected? <brief pause> Otherwise, here are some other similar business locations that I found that are closer to you. Let me know which one you would like to connect to.”
As the caller continues to engage, the Soleo Discovery Engine replies with any natural responses needed to further improve the matching of listings to their intent. In this way, the application emulates the experience a caller would have with a traditional directory assistance operator.
In the end, a few key things happen –
- We connect the caller to a local merchant that meets their needs (possibly with a sponsored listing, if available and precisely applicable) with speed and precision.
- We note the caller affirmation on that phone number for optimization of future calls.
- And, the caller affirmation is fed back into our language models to drive the machine learning that further improves our ability to converse naturally with future callers.
Soleo’s Discovery Engine increases our business intelligence for that local market while also delivering a natural, conversational user experience. Simply put, solutions driven by Soleo’s Discovery Engine make search services better by understanding the user’s intent and delivering better results.