What interaction history signals indicate what a participant finds relevant?
This explores how an AI can tell what a user actually cares about from the traces they leave behind — the structure, timing, and selection of their interactions — rather than from what they explicitly state.
This question asks what in a person's interaction history reveals relevance — and the corpus suggests the signal lives less in *what* people say than in the shape, selection, and timing of how they engage. A striking thread is that conversation structure alone predicts whether an interaction is working. A structure-only model tracking the geometry of a dialogue's trajectory hit 68% accuracy on satisfaction, nearly matching a full-text analysis at 70%, and combining the two reached 80% Can conversation shape predict whether it will work? Can conversation structure predict dialogue success better than content?. Relevance, in other words, leaves a geometric fingerprint. The same instinct drives "conversational DNA," which treats dialogue as a living system and tracks linguistic complexity, emotional trajectory, topic coherence, and relevance as parallel temporal streams Can tracking dialogue dimensions simultaneously reveal hidden conversation patterns?.
A second thread says: not all history counts equally, and indiscriminate recall actively hurts. Selectively retrieving the relevant past turns beats dumping in the full conversation, because topic switches inject noise that drowns the signal Does including all conversation history actually help retrieval?. Pushed further, abstract summaries of preference outperform replaying specific past interactions — semantic memory beats episodic memory, and recency-based recall beats similarity-based retrieval Does abstract preference knowledge outperform specific interaction recall?. The lesson is that relevance signals must be distilled, not accumulated; the raw transcript is mostly distraction.
Third, relevance signals come through more than one channel. Conversational recommenders that lean only on the live session miss the user-level and item-level patterns that traditional systems exploit — you need current session, historical dialogues, and look-alike users together, conditioned on present intent, to reconstruct who someone is Can conversational recommenders recover lost preference signals from history?. And some signals don't live in any single person's history at all: aggregating clicks across users builds a global graph that exposes article relationships invisible in any one sparse individual trail Can cross-user behavior reveal news relations that individual histories miss?. What you find relevant is partly legible only against what people like you found relevant.
Where this gets genuinely unexpected is the body. Beyond text and clicks, behavioral cues — gaze, hesitation, typing speed — function as a continuous readout of cognitive state, letting a system infer engagement and load without ever interrupting to ask Can AI systems read cognitive state from interaction patterns alone?. There's even a relational version: in therapy transcripts, working alliance can be inferred turn by turn Can we measure therapist-patient alliance from dialogue turns in real time?, and linguistic coordination — measured as the embedding distance between two speakers' words — tracks empathy and rapport, rising as relationships improve Can we measure empathy and rapport through word embedding distances?. The thing you didn't know you wanted to know: the strongest relevance signal may not be the content a person selects but the rhythm of their hesitation and how closely their language drifts toward the system's — relevance as a property of the dance, not the words.
Sources 10 notes
A structure-only model analyzing conversation trajectory achieved 68% accuracy predicting satisfaction, nearly matching full-text LLM analysis at 70%. Combined structural and textual features reached 80%, showing that how conversations unfold geometrically captures interaction quality text-based classifiers miss.
TRACE achieved 68% accuracy predicting dialogue success from structural features alone, matching a 70% content-based baseline. A hybrid combining both reached 80%, suggesting how agents communicate rivals what they say.
Conversational DNA encodes four simultaneous dimensions—linguistic complexity, emotional trajectories, topic coherence, and conversational relevance—as temporal streams. The reverse Turing test finding showed expert assessments of AI diverged sharply, suggesting conversational structure shapes interpretation as much as content.
Research shows that automatically selecting relevant previous turns improves retrieval effectiveness more than including all context. Topic switches inject irrelevant information; joint optimization of selection and retrieval beats both full-context baselines and human annotation.
PRIME framework shows semantic memory (preference summaries, parametric encodings) consistently beats episodic memory (retrieved past interactions) across models. Recency-based recall outperforms similarity-based retrieval, and task fine-tuning exceeds preference tuning methods.
Current CRS systems only use the active dialogue session to infer preferences, losing item-CF and user-CF signals proven valuable in traditional recommenders. Integrating current session, historical dialogues, and look-alike users—conditioned on current intent—recovers essential user representation structure.
GLORY constructs a global news graph from aggregated user clicks to discover article relationships invisible in any single user's sparse history. This population-level behavioral structure enables recommendations even when direct textual or per-user similarity fails.
Research shows AI systems can instrument multimodal behavioral signals (gaze, hesitation, speed) to read cognitive state during interaction, preserving flow by avoiding disruptive explicit probes. However, the same substrate enables both helpful timing and manipulative profiling.
COMPASS maps dialogue turns onto WAI embeddings to produce 36-dimensional alliance scores per turn. Anxiety and depression show convergence in alliance metrics over time, while suicidality shows persistent misalignment between patient and therapist.
Word Mover's Distance captures lexical, syntactic, and semantic coordination simultaneously and correlates with therapist empathy in MI and affective behaviors in couples therapy. Couples showing relationship improvement exhibit increasing coordination over the therapy course.