arXiv subjects

Dingjing Shi

Publications and source records attributed to Dingjing Shi.

2 recordsLinked to original sources

arXiv2026

Adaptive Ecological Momentary Assessment with a Hybrid Language Model: Formative Expert Review and Retrospective Evaluation

Ecological momentary assessment (EMA) measures experience in daily life, but fixed questionnaires and schedules collect information of uneven value and can interrupt participants. We present and retrospectively evaluate EMA-E4B, a hybrid framework for question selection and prompt timing. Separate ridge models propose an item set and delay; a supervised Gemma 4 E4B language layer produces the final structured response and explanation. Evaluation distinguishes proxy action performance, output conformity, and formative judgments of response quality. The data contain 4,372 records from 79 participants and yield 3,516 sequential cases under a participant separated split. One involved domain expert preferred the complete hybrid response in 15 of 20 decisive comparisons, with two ties among 22 reviews. On 75 reused development cases, hybrid question utility and timing similarity were 0.832 and 0.818; the head alone reached 0.852 and 0.818. A separate 60 case comparison with untouched E4B under the same head gave action differences of -0.0031 and -0.0105. Thus, the language layer produced structured responses with action scores comparable to or slightly below the reference configurations, while the expert feedback favored the complete hybrid response. These observations establish a concrete, inspectable framework and clarify the distinct roles of action scoring and response review. Repeated adaptive administration and practical effects on measurement and participant burden remain future research.

cs.HC↗

arXiv2026

A Validation-Gated Mechanistic Account of Suicidality Detection in LLMs

Large language models are increasingly proposed for mental-health applications such as detecting suicidal content, raising the question of what they rely on. We study this mechanistically and use it to ask a narrower question: how to make a causal claim about a model's internal features more trustworthy. Our validation-gated framework, with suicidality detection as a case study, interprets a behavior only after the model is shown to perform it: a concept is admitted only once the model ranks it above a simple lexical baseline, and each subsequent property is tested against a matched control. This discipline yields negative as well as positive results. The gate rules out one task at the outset: on DeepSuiMind (Li et al. 2025), Llama-3.1-8B-Instruct cannot separate implicit suicidal intent from ordinary distress, so we do not analyze it. We turn to binary suicide detection, which it does perform. There we find a mid-network feature that appears semantic rather than keyword-based, is causally implicated in the decision (ablating it degrades the judgment; a random direction does not), is low-rank, and recurs across three model families and three suicide datasets. A register-matched control (suicide versus depression) suggests it tracks suicidality more specifically than general distress. Steering raises the model's response, but for unrelated questions too, so we treat it as necessary but not sufficient. The clearest pattern separates encoding from use: smaller models already represent suicidality, yet only larger ones appear to act on it. The positive evidence is English Reddit text, which limits the clinical reading.

cs.CL↗