Why graphical displays work great for quantitative analytics, but words speak louder for Natural Language Understanding (NLU)
Note: this post was originally published on LinkedIn.
Visual expressions are like looking at the Mona Lisa. In a couple sentences, you can describe the beauty of a painting that smiles at you from every angle. Now try to put this into a graph.
Expressing quantitative analytics using heat maps, donut graphs, bar graphs, line graphs are common practice routinely used by quantitative analytics providers. It is almost a natural preference for humans to map numbers with analog expressions. Numbers are designed for comparison. Graphical widgets and charts provide comparative information on similar metrics using scales and coordinates to form a common baseline. For example, if we look at a graphical display showing a comparative bar graph of same store sales over the past 5 years, we can quickly ascertain the increase and decrease in sales by stores over time. If we look at a bar chart by zip code, we can compare sales growth with respect to location. That is the power of graphical display for quantitative analytics. The goal, however, analyzing textual data is to understand context. If we apply the same graphic methods used in quantitative analytics, the context and relevance are typically lost in translation.
…when we look at the broad demographics of an enterprise workforce that stand to benefit from NLU AI-amplified KPIs, we need to use a medium that is naturally compatible with their thinking.
Many commercial Natural Language Processing (NLP) tools turn to the familiar graphical widgets we find in quantitative analytics. Many NLP and NLU vendors routinely convert textual data into quantifiable metrics and then use graphical widgets to depict these metrics (e.g. word counts, sentiment analytics, knowledge graphics, semantic networks, semantic neighbors, spatial relationships between words and context, relationships between signal and noise, grammar, syntax, synonyms, ontologies, etc.).
At SiteFocus, we explored a preferred metaphor for NLP/NLU presentations. Our NLU analytics system, CIF, uses principles of Symbolic AI. Instead of converting words and sentences into quantitative values seen in conventional text analytics, our software analyzes natural language text by symbols, context, semantic neighbors, and Point of View (POV) analytics. Results are presented in graphical widgets, words, and structured textual reports that retain context & relevance of the original text.
However, during the course of presenting our NLU analytics, I have received some interesting feedback. When rendering NLU insights in graphical widgets — our Meta-Vision network quadrant charting medium — most of the feedback was positive, telling us the knowledge network graphics were “cool” and the interactive edge nodes would keep them busy exploring. Yet, a few moments later, I’m often asked the same questions: “What I am looking at?” and “What is the insight?”
We have to respect the human equation in our audience.
People with analytical backgrounds found the Meta-Vision’s graphical network presentation useful. Without the technical background, however, people struggled to distill meaningful insight. To solve this, we developed a new way to present NLU findings in a report similar to how an analyst would present information — a flattened hierarchical knowledge graph of important subjects, relationships, and supporting excerpts. This approach received a much warmer reception from folks with different backgrounds and professional skills.
A picture is worth a thousand words; a thousand visuals won’t describe those words
Visual expressions are like looking at the Mona Lisa. In a couple sentences, you can describe the beauty of a painting that smiles at you from every angle. Now try to put this into a graph. Insights are only as good as the analytic metaphor. Words speak louder in NLU. We have to respect the human equation in our audience. While it is great to show the analog of meta-insights to folks that are familiar with network graphs, others may find it confusing or misleading. Natural language is “natural” because it is the medium people use in their communication. To transcode communication, to seek for a better understanding, it is essential for us to revert back to the natural form. AI has great potential. However, when we look at the broad demographics of an enterprise workforce that stand to benefit from NLU AI-amplified KPIs, we need to use a medium that is naturally compatible with their thinking. In that regard, we concluded that a Natural Language Report that conforms to their expectation is the best way to go.
To see our reports on real-world business applications, visit our website or feel free to connect on LinkedIn or by email.
