Glossary·methodology

Sentiment trajectory

How feature-level sentiment changes over time, computed per product × property × period. The single most actionable perception metric — because trajectory matters more than level.

What it is

For every product × canonical property × time period, Theia computes:

  • mention count
  • average sentiment score
  • direction (improving / declining / stable)
  • statistical significance of the change

The output is a trajectory — not a snapshot.

Why trajectory beats level

A product at sentiment 0.69 tells you almost nothing.

A product at sentiment 0.69, improving from 0.41 over 12 months, with autofocus sentiment going from 0.59 → 0.82, tells you the market is actively re-evaluating it.

That second product is a buy signal — for the brand (double down on what's working), for the retailer (push it to feature placement), for the PE firm watching the category.

This is the Canon EOS R8 pattern. Mid-pack on level. Fastest-improving in the segment on trajectory. Marketing strategy entirely different.

How it's computed

For each (product, canonical_property, period):

  1. Pull all matching snippets from rag_snippets
  2. Filter to a single snippet_type (feature / benefit / use_case)
  3. Compute avg_sentiment and mention_count
  4. Compare to prior periods to assign direction:
    • Improving: trend > +0.1 over 3 periods, significant
    • Declining: trend < −0.1 over 3 periods, significant
    • Stable: otherwise

Persisted in the sentiment_trajectories table. ~53K rows for the Canon EU deployment alone.

Why monthly granularity matters

Quarterly is too slow. By the time a sentiment problem shows up in a quarterly tracker, it's already affected three months of sales. Monthly trajectory tracking gives a 6-10 week early warning vs sales data — long enough to react.

Source publication dates are used as the time dimension, not pipeline execution dates. A review published 6 months ago is processed today but dated 6 months ago. This means trajectories are real, not artefacts of when collection happened.

Use cases

Consumer brands: "Which features are improving for our product? Which are declining? Where should the next content drop focus?"

PE diligence: "Has perception of this portco's flagship product turned the corner, or is it still trending down?"

Research firms: "Quarterly client report needs a trajectory chart for the top 5 features — not just current levels."

Industrial B2B: "Engineer-forum sentiment on GenICam compatibility — is the new SDK release fixing the complaints?"

Strategic implication

The single biggest gap in legacy market research is the assumption that a quarterly tracker is sufficient. Sentiment trajectories at monthly granularity, refreshed weekly, are now table stakes for continuous brand monitoring.

The firms that built trajectory infrastructure are the ones who can give a CMO a "sentiment-as-a-leading-indicator" dashboard. The firms still on quarterly waves can't.