How do users interact with a Knowledge Graph?
The purpose of a knowledge graph is to answer a user's questions. Some of the questions may be known upfront, while some questions users may never think of themselves. The interaction of a user with the knowledge graph could be real-time, or a batch process might be run to produce certain reports to answer some predefined questions and to produce certain analytics. This results in a matrix of four different modes of interaction interactions with the knowledge graph along the dimensions of whether the interaction is initiated by the user (ie, Pull), or in response to information presented to the user (ie, Push), and whether the questions are known in advance vs questions are not known in advance.
Above modes of interaction are usually supported through a combination of search, query answering, and graphical interfaces. A search interface is like the interface of a search engine where the user may simply type keywords. The query interfaces range from a formal query language to a natural language interface. A graphical interface may be used for composing a query, for viewing the results of a query or for browsing the graph defined by the instances in the knowledge graph.
The actual interface to a knowledge graph will typically use a combination of methods. For example, a query might be composed through a combination of search and structured query interface. Similarly, the results may be partly graphical, and partly textual. It is too common for us to see graphical visualizations of knowledge graphs containing thousands of nodes and edges on a screen. Many times, the graphical visualization is simply a backdrop for the points to be made in contrast to driving and contributing to the insights that help us identify what points to make. Just because we are working with a knowledge graph, it should not automatically imply that a graphical visualization is the best way to interact with it. One should turn to the best principles for visualization design, and choose the most effective medium for presenting the information.
Vinay K. Chaudhri, this captures a fundamental constraint that most knowledge graph implementations miss. The interaction matrix you describe - pull vs push, known vs unknown questions - reveals the real bottleneck: most knowledge graphs are built to answer questions we already know how to ask. The constraint is not the interface design or visualization choice. It is that knowledge graphs become most valuable when they surface questions we never thought to ask, but most systems are optimized for the questions we already have. When you connect mathematical algorithms to the deeper patterns that most people cannot see but can definitely feel, the knowledge graph transforms from a query system into a discovery engine. The interface becomes less important than the system's ability to reveal connections that exist but remain invisible.
This is a deep truth. 👍🏻
Cognitive Software Alliance is looking for projects in 2021- our Ai engineering tools are currently used in several multi million dollar financial systems. We apply knowledge graphs and graph engines together for new capabilities.