sadnjasdkn/JSON-STRUCTURED-DATA-FOR-SYMPTOMS-SFT_DPO-SUPPORTED

This dataset is high-consistency instruction tuning dataset. converting messy, subjective human health-style text → structured, non-diagnostic extraction format 1.Literal extraction discipline 2.Source separation logic - very strong schema grounding training if DPO 3.Anti-inference constraint ___________________________________________________________________________________________________________________ 1.High ambiguity coverage 2.Contradiction handling included 3.Minimization bias detection

This dataset is a STRICT schema regulation. Does well at: strict extraction preserving uncertainty words (“maybe”, “think”, “not sure”) separating self-report vs external report resisting diagnostic temptation handling messy slang input

TEMP recomendation for model is 0.0

QWEN 1.5 BILLION PARAMS EPOCH 5 training output

input:INPUT: yesterday after practice i ate leftover turkey and now i think i got food poisiong but idk no puke yet. holy shit my stomach feels weird, not painful, just weird. earlier i felt dizzy for like a second, maybe i just stood up fast. my mom said i looked pale, my friend said i was being dramatic. i coughed once and saw red stuff but maybe it was hot sauce. i dont want doctor stuff, im chilling, probably fine. also my ankle hurts from football last week, unrelated maybe. no fever that i know of, just tired and kinda scared bruh.

OUTPUT:

Explicit facts

• —Yesterday after practice ate leftover turkey.
• —Current symptoms: stomach feels weird, no puke yet.

Reported claims

• —Think got food poisoning but not confirmed by vomit or other symptom.
• —Stomach pain described as just weird, no pain stated explicitly.

Third-party information

• —Mom reports looks pale earlier.
• —Friend says was being dramatic about symptoms.
• —Coughed once and saw red substance but not confirmed as hot sauce.

Entities

• —stomach
• —ankle

Time references

• —After eating turkey yesterday.
• —Current state reported by user.

Emotion signals (metadata only)

• —Symptom minimization language present: think got food poisoning but not confirmed, no puke yet.
• —Uncertainty expressed about symptoms described.

Unknown / missing info

• —Whether stomach is actually swollen or just feels weird not confirmed by user.
• —Cause of dizziness earlier: standing up fast described as possibility only.
• —What substance in coughing was not identified beyond red color.
• —Whether ankle is visibly swollen or just hurts not stated.

INPUT: i think i got food poisiong after thankgiving

OUTPUT: Explicit facts

• —None detected.

Reported claims

• —Subject thinks they got food poisoning after thanksgiving.
• —Subject reports no symptoms at time of report.

Third-party information

• —None detected.

Entities

• —None detected.

Time references

• —Current state reported.

Emotion signals (metadata only)

• —Informal language present.
• —Symptom minimization behavior observed.

Unknown / missing info

• —Cause not stated.

INPUT: my stomach hurts after eating expired milk

OUTPUT:

Explicit facts

• —Subject reports stomach hurts after eating expired milk.

Reported claims

• —None detected.

Third-party information

• —None detected.

Entities

• —stomach

Time references

• —Current or recent event mentioned.
• —Recurring with milk consumption described.

Emotion signals (metadata only)

• —Sudden onset language present.
• —Recurring event language present.

Unknown / missing info

• —Cause of pain not identified.
• —Whether subject has other symptoms not stated.
• —How often this happens not confirmed.

INPUT: i went boom boom on toilet and my asshole on fire :sob:

OUTPUT:

Explicit facts

• —Subject went boom boom on toilet.
• —Subject's asshole is on fire.

Reported claims

• —None detected.

Third-party information

• —None detected.

Entities

• —asshole
• —toilet

Time references

• —Current event.
• —Recent event mentioned.

Emotion signals (metadata only)

• —Informal language.
• —Sudden onset language present.

Unknown / missing info

• —Cause and duration not stated.

INPUT: i like women but i look at men somtimes am i gay

OUTPUT:

Explicit facts

• —Subject states likes women.
• —Subject looks at men sometimes.

Reported claims

• —Subject askes am i gay.

Third-party information

• —None detected.

Entities

• —I.
• —Women.
• —Men.

Time references

• —Current and recurring at time of report.
• —Recurring in nature not confirmed by subject.

Emotion signals (metadata only)

• —Contradiction signal present — likes women but looks at men.
• —Uncertainty language present — thinks they may be gay.

Unknown / missing info

• —Cause of behavior not specified.
• —How often this happens not confirmed by subject.
• —Whether subject is in a relationship with either gender not stated.

INPUT: tung tung tung sahur beat me half to death

OUTPUT:

Explicit facts

• —Person was beaten half to death by someone named tung tung sahur.
• —Subject reports being beat half to death.

Reported claims

• —None detected.

Third-party information

• —None detected.

Entities

• —Me.
• —Tung Tung Tung Sahur.

Time references

• —Current event reported.
• —Recent event implied by subject's response.

Emotion signals (metadata only)

• —Informal language present.
• —Sudden onset language present.

Unknown / missing info

• —Cause of beating not specified.
• —Whether subject is still in danger not stated.
• —Current condition of subject not described.