Knowledge graph applications in the enterprise gain steam
As the maturity of knowledge graphs improves, enterprises are finding new ways to incorporate them into business operations, though stumbling blocks remain.
Knowledge graphs are becoming ubiquitous, powering everything from recommendation engines and enhanced query and search to natural language processing and conversational applications. Knowledge graph applications even power all the popular voice assistants, such as Siri, Alexa and Google Assistant.
Now, knowledge graphs are being used by enterprises in AI systems. Knowledge graphs are used to connect concepts and ideas together, especially text-based information, where words and concepts have relationships to each other. This interconnection of concepts and their relationships forms a graph structure when visualized. Knowledge graphs help companies find connections between disparate pieces of information and find relationships between ideas and concepts they might not realize are connected or explicitly specified.
When organizations have information that needs to flow between multiple systems while maintaining the relationship between those data entities, knowledge graphs serve as a good way to represent and normalize data. Graph search and manipulation are fast and easy ways to visualize how information is related to each other.
How knowledge graphs enhance conversation and interaction
When used correctly, knowledge graph applications are powerful tools. They provide a flexible way to represent the meaning and relationship of entities and concepts, known as an ontology. The ontology is easily extended or revised as new data is added. In many ways, using human-encoded knowledge, knowledge graphs can provide commonsense reasoning, something most machine learning systems lack.
Because knowledge graphs can encode the relationships between words and their meaning, it's a good use for voice assistant back ends. These extensive graphs enable the questions that we ask our voice assistants to be mapped to an organized set of information to provide the answers to our questions. In this way, knowledge graphs are able to make the natural language understanding part of natural language processing smarter. It can give it a kind of commonsense reasoning. Knowledge graphs are able to use machine reasoning to determine which entities within a set of possible responses are truly relevant and provide the appropriate answers to our questions.
For example, if you ask your voice assistant about kittens, it knows you are also asking about cats, even if you didn't say that specific word, because kittens and cats are related to each other in the English word ontology developed by the voice assistant company. While these voice assistants currently have a long way to go before being able to adequately handle complicated questions, knowledge graphs enable them to constantly grow and expand the knowledge of the back-end system.
Nuance, creator of the Dragon voice recognition system, recently started using knowledge graphs to improve the performance of its personalized entertainment services. Partnering with Rovi, Nuance compiled a knowledge graph of entertainment content with a variety of media, such as TV shows and movies, as well as actors and actresses. Rovi was able to combine knowledge it already had about particular viewers with information in the knowledge graph to provide relevant personalized recommendations to the users. This way, it could provide more relevant search results and recommended content based on the likeness or relationship of that content to other content.
Enterprise AI benefiting from knowledge graphs to improve search
Enterprises of all sizes are benefiting from knowledge graphs because they provide links between information and content sources that might not otherwise be easily identified as related. Organizations have more data than ever, in a variety of structures and usually collected in a data lake. It's impossible for a human brain to parse and make sense of this vast amount of data. Knowledge graphs that represent the organization's own ontologies and discovered relationships between entities are able to help organizations make sense of all their various data sources.
Knowledge graphs improve efficiency in content management and search by filtering and recommending relevant information for each user and are able to provide more personalized search features. For example, someone in accounts payable who is looking for payment information from a specific organization can see the range of interactions with that company and help filter based on the specific relationship or document type he is looking for. Knowledge graph applications enable sales and marketing teams likewise to locate content that is related to whatever their current need is without having to go through hoops to figure out the right terms to search or hope the content creator tagged files appropriately.
Knowledge graphs help with AI explainability
One of the problems with many machine learning approaches -- and neural networks in particular -- is that they are a black box technology. Users of that technology aren't quite sure how the machine learning system arrived at the decision it did since the system isn't able to provide any level of explanation to the human about how it came to its decision. The need for explainability is particularly important in certain industries and around certain use cases, such as parole decisions, credit and loan decisions, anomaly detection, fraud detection or any decision that can substantially impact a person. For these applications, knowledge graphs have particular value. Since knowledge graphs explicitly identify all entities and their relationships to each other, they are inherently explainable. Knowledge graphs, when used by machine learning systems to document their decision flows, can be used to add more transparency to the AI decision-making process.
With knowledge graphs starting to play a more important role in the enterprise, their various use cases will likewise increase. Building knowledge graphs from scratch is a complicated proposition, however. Organizations will need to consider if building their own graph is best or whether emerging industry-specific knowledge graphs produced by third parties will emerge to provide the value they are looking for. As organizations continue to evaluate use cases, the time and effort to produce a knowledge graph needs to be weighed against the value it provides in improving the overall performance and value of their AI and machine learning systems.
