ALBA: Adaptive Language-Based Assessments for Mental Health.

Area of Science:

• Psychometrics
• Natural Language Processing
• Mental Health Assessment

Background:

• Mental health conditions present with diverse symptoms, requiring nuanced assessment.
• Current language-based assessments need extensive data for reliable psychological trait scoring.
• Adaptive Language-Based Assessment (ALBA) aims to improve efficiency and accuracy in mental health evaluations.

Purpose of the Study:

• Introduce and evaluate Adaptive Language-Based Assessment (ALBA) methods.
• Develop adaptive testing strategies for scoring latent psychological traits using limited language.
• Compare psychometric approaches including Classical Test Theory and Item Response Theory.

Main Methods:

• Developed adaptive testing methods based on Classical Test Theory and Item Response Theory.
• Introduced two novel methods: a semi-supervised Item Response Theory-based method (ALIRT) and a supervised Actor-Critic model.
• Empirically evaluated ordering and scoring strategies for adaptive language-based assessments.

Main Results:

• Both ALIRT and the Actor-Critic model outperformed non-adaptive baselines in accuracy and efficiency.
• ALIRT demonstrated superior accuracy and scalability, achieving high correlation (Pearson r ≈ 0.93) with only 3 questions.
• Adaptive language-based assessments effectively utilized smaller language samples for depression and anxiety, maintaining validity and low computational cost.

Conclusions:

• Adaptive Language-Based Assessment (ALBA) offers a more efficient approach to psychological trait measurement.
• ALIRT provides a highly accurate and scalable method for language-based mental health assessment.
• This approach reduces the need for extensive language data, making assessments more feasible.