Contribution of Pipeline Components to User Alignment

Determine how conversation segmentation, insightfulness scoring, evidence weighting, and LLM-based value annotation individually contribute to the alignment between Value Faces’ inferred demonstrated-value profiles and participants’ assessments, and whether alternative component designs improve that alignment.

Background

The study evaluates Value Faces end to end rather than independently validating each pipeline component. Although participants’ assessments generally align with the system’s inferences, the separate contributions of segmentation, scoring, weighting, and annotation are not identified. The paper therefore leaves unresolved whether alternative implementations would improve agreement with users.

References

The contribution of each component to these results, and whether alternatives would improve alignment, remains a question for future work.

— Value Faces: Surfacing How Self-Presentation Shifts Across Relationships  (2609.19581 - Koo et al., 17 Sep 2026) in Limitations and Future Work, subsection “Pipeline Validation”

Ultimately, which perspective is relevant to surface depends on the purpose of the system, an area we leave open for future work.

— Value Faces: Surfacing How Self-Presentation Shifts Across Relationships  (2609.19581 - Koo et al., 17 Sep 2026) in Limitations and Future Work, subsection “Subjective Value Interpretation”