Building Behavioral Ontology v0.1

Every science begins with classification. Before chemistry could advance, it needed the periodic table. Before biology could mature, it needed Linnaean taxonomy. The study of digital human behavior has no equivalent — no shared vocabulary, no structured framework for describing what happens when people interact in online communities. We set out to build one.

This is not because no one has tried to describe online behavior. Sentiment analysis offers a binary lens: positive or negative. Engagement metrics count actions: messages sent, reactions given, time spent on platform. Moderation taxonomies classify content by harm category: harassment, spam, misinformation. But none of these frameworks describe behavior itself. They describe outcomes, or volume, or content — never the dynamics of how people actually function within a social system. A person who posts a single thoughtful message per week and a person who posts fifty reactive messages per day are treated identically by activity metrics, despite performing fundamentally different social functions.

Behavioral ontology is our attempt to build the missing layer. It is a structured classification system for digital interaction patterns — not what people say, but what their behavior does within the social environment they participate in. The distinction matters. Two people can say nearly identical things and produce completely different effects on a conversation, depending on their history, their position in the social structure, and the state of the discussion they are entering.

The ontology operates across five core dimensions, each capturing a distinct aspect of how an interaction functions. Relational Dynamic captures how a person positions themselves relative to others in the conversation: deferring, collaborating, competing, or asserting dominance. This is not about personality — the same person might defer in one context and assert in another. The dimension tracks the behavior, not the person. Epistemic Behavior tracks how they handle ideas and information: absorbing and building on others’ contributions, selectively filtering what they engage with, or actively rejecting perspectives that challenge their position. Emotional Register measures the degree of regulation in their expression — not whether they are emotional, but whether their emotional expression is calibrated to the context or reactive to it. Influence Vector identifies what the interaction does to the trajectory of a conversation: does it advance, redirect, stall, or reverse the direction of the discussion? Trust Impact captures whether the interaction builds or erodes the group’s willingness to engage constructively with each other.

Each dimension is measured on a continuous scale rather than assigned a categorical label. This is a deliberate design choice born from early failures. In our first prototypes, we used labels — “dominant,” “collaborative,” “passive” — and quickly discovered that labels collapse nuance in ways that produce misleading results. A person who is “dominating” a conversation might be bulldozing productive discussion, or they might be anchoring a chaotic thread that was spiraling into noise. The behavioral signal is similar, but the social function is entirely different. Continuous scales preserve the information that labels discard, and they allow the system to represent the gradients and edge cases that are where the most interesting behavioral patterns tend to live.

The ontology also introduces the concept of social function — the role an interaction plays in the group dynamic, independent of its content. An interaction might bridge two otherwise separate conversational threads, connecting people who were not previously engaging with each other. It might anchor a discussion that was drifting by reintroducing evidence or refocusing the question. It might escalate tension, deflect a difficult topic, repair damage from a previous conflict, or suppress legitimate dissent through appeals to harmony. These functions are visible only in context, which is why any analysis system that examines messages in isolation will miss them entirely. A message that reads as aggressive in isolation might be performing a repair function in context — directly naming a problem that passive avoidance was making worse.

We got several things wrong in early versions. The initial ontology had seven dimensions, not five. Two of them — “cognitive complexity” and “narrative positioning” — proved impossible to classify reliably from text alone. They required inference about internal mental states that the available data could not support with sufficient confidence. We removed them rather than produce unreliable signals. We also initially treated social function as a property of the message rather than of the interaction-in-context. This produced systematically wrong classifications because the same message performs different functions depending on what precedes and follows it.

Version 0.1 is a starting point, not a conclusion. We expect it to evolve significantly as we analyze more data across more community types. Some dimensions may need to be split as we encounter behavioral patterns that the current resolution cannot distinguish. New social functions will surface. The goal is not to build a permanent taxonomy on the first attempt — it is to build a framework that is precise enough to be useful and flexible enough to accommodate what we do not yet know.

The deeper motivation behind this work is simple: you cannot study what you cannot name. Without a structured vocabulary for digital behavior, every insight about community dynamics remains anecdotal — one person’s subjective impression, impossible to verify, compare, or build on. With a shared ontology, behavioral intelligence becomes possible. Patterns can be identified across communities. Trajectories can be measured over time. Predictions can be tested against outcomes. That is what we are building toward.