masterthesis-playground/raft/generate_culture_spirit_interview_prompts.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-

"""
Generate 300–1000+ English interview questions targeted ONLY at culturally/spiritually
interested Bali tourists (Lead Users), covering 5 cognitive destination image dimensions:
- Natural Attractions
- Atmosphere
- Social Environment
- Infrastructure
- Value for Money

Key constraint:
- Every prompt must be meaningful for culture/spirituality-first travelers.
- Avoid party/shopping/hedonistic positioning.
- Include etiquette, authenticity, sacredness, commodification, meaning-making, reflection.

Outputs:
- JSONL: {"dimension": "...", "type": "...", "prompt": "...", "tags": [...]}
- or TXT: one prompt per line
"""

import argparse
import json
import random
import re
from typing import Dict, List, Tuple

DIMENSIONS = [
    "Natural Attractions",
    "Atmosphere",
    "Social Environment",
    "Infrastructure",
    "Value for Money",
]

# -----------------------------
# Segment-specific building blocks
# -----------------------------
# Keep places generic (no need to hallucinate specific proper nouns)
NATURE_FOR_MEANING = [
    "rice terraces that feel lived-in rather than staged",
    "waterfalls approached with a quiet, respectful mood",
    "volcano viewpoints that invite reflection at dawn",
    "jungle walks where you notice offerings and small shrines",
    "lake areas that feel calm and contemplative",
    "coastal paths that feel like a moving meditation",
    "hot springs experienced as restoration rather than spectacle",
]

CULTURE_SPIRIT_SPACES = [
    "temple courtyards and entry paths",
    "a village ceremony you observe respectfully",
    "a traditional market where everyday ritual shows up in small ways",
    "a dance performance where you try to read symbolism",
    "a craft workshop focused on meaning and lineage, not souvenirs",
    "a community space where offerings are prepared",
    "a quiet heritage walk where stories feel layered",
]

RITUAL_ETIQUETTE_TOPICS = [
    "dress codes and modesty",
    "offerings and what not to touch",
    "photography boundaries",
    "when to speak vs stay quiet",
    "how to move through a temple without intruding",
    "how to ask questions without turning sacred life into content",
]

MEANING_MAKING = [
    "a sense of humility",
    "a feeling of gratitude",
    "a moment of awe",
    "a feeling of being a guest",
    "a sense of calm",
    "a quiet emotional reset",
    "a shift in how you see daily life",
    "a stronger respect for local rhythms",
]

AUTHENTICITY_CUES = [
    "how people behave when no one is watching",
    "whether the experience is integrated into local life",
    "how money is handled (transparent vs extractive)",
    "whether rules feel protective or performative",
    "whether the pace allows reflection or pushes consumption",
]

CROWDING_COMMODIFICATION = [
    "overt commercialization around sacred spaces",
    "crowds that change the emotional tone",
    "performative 'authenticity' for tourists",
    "feeling like sacredness is being packaged",
]

CONTEXTS = [
    "early morning before the crowds",
    "late afternoon when light softens and things slow down",
    "during a local ceremony where you are clearly a guest",
    "in rainy season when plans change and patience matters",
    "on a quiet weekday compared to a busy weekend",
    "with a local guide who emphasizes respect and context",
    "solo, when you can be more contemplative",
    "as a repeat visitor, noticing subtler layers",
]

TRAVELER_PROFILE = [
    "a culture-first traveler",
    "a spirituality-curious traveler",
    "a respectful observer who avoids intrusive tourism",
    "a slow traveler seeking depth over volume",
    "a repeat visitor looking for subtler, less packaged experiences",
]

CONSTRAINTS = [
    (
        "time",
        [
            "you only have 6 hours but want depth, not a checklist",
            "you have one full day and want it to feel coherent and meaningful",
            "you have three days and want a gentle pace with time for reflection",
            "you can only travel within a short radius and must choose carefully",
        ],
    ),
    (
        "budget",
        [
            "you have a modest budget but still want cultural depth and fairness",
            "you'll pay more if it supports local communities transparently",
            "you want predictable costs and dislike hidden fees around sacred sites",
            "you prefer smaller, community-rooted experiences over pricey packages",
        ],
    ),
    (
        "crowds",
        [
            "you want to avoid crowds because they dilute atmosphere and respect",
            "you can handle crowds if etiquette and sacredness are preserved",
            "you want a balance: one iconic site, mostly quieter, community-rooted places",
            "you get overwhelmed by busy places and need calmer, respectful alternatives",
        ],
    ),
    (
        "weather",
        [
            "it's rainy season and flexibility is part of respectful travel",
            "it's very hot and you need a pace that still feels mindful",
            "visibility is low and your sunrise plan may fail—how do you adapt meaningfully?",
            "roads feel unsafe, so you prioritize fewer moves and deeper presence",
        ],
    ),
    (
        "mobility",
        [
            "you avoid steep stairs but still want meaningful cultural/spiritual moments",
            "you prefer not to ride a scooter and want low-friction transport options",
            "you want minimal walking but still want authenticity and atmosphere",
            "you need frequent rest and prefer fewer transitions",
        ],
    ),
    (
        "ethics",
        [
            "you want to avoid commodifying sacred life",
            "you prioritize local benefit, consent, and respectful boundaries",
            "you avoid experiences that pressure locals to perform for tourists",
            "you want your presence to feel like 'being a guest' not 'taking'",
        ],
    ),
]

TRADEOFFS = [
    ("depth of understanding", "convenience"),
    ("sacredness", "accessibility"),
    ("quiet reflection", "seeing iconic places"),
    ("guided cultural context", "self-guided freedom"),
    ("photography", "presence and respect"),
    ("predictable pricing", "spontaneous discovery"),
    ("community benefit", "personal comfort"),
    ("slow pace", "variety of stops"),
]

CONTRASTS = [
    ("a popular temple area", "a quieter village setting"),
    ("a curated tour script", "a guide who shares context and encourages respect"),
    ("a crowded ceremony-adjacent spot", "a calm everyday ritual moment"),
    (
        "a market aisle focused on souvenirs",
        "a market moment that shows daily offerings and rhythm",
    ),
    ("a rushed checklist day", "a slower day with fewer places but deeper presence"),
    ("an 'Instagram moment'", "a moment of quiet meaning that you don't photograph"),
]

INTERVIEW_STYLES = [
    "Tell me about a time when…",
    "Walk me through…",
    "As a culturally/spiritually motivated traveler, how do you…",
    "If you had to advise a tourism marketer focused on respectful cultural travel…",
    "What surprised you about the spiritual or cultural texture of…",
    "What does 'authentic and respectful' look like to you when…",
    "How do you personally decide whether to join, observe, or step back when…",
]

FOLLOWUP_PROBES = [
    "What specifically made it feel respectful or not?",
    "What did you notice first, and what happened next?",
    "How did it change your mood or sense of meaning that day?",
    "What would have improved it without turning it into a spectacle?",
    "What boundary would you not cross again?",
    "What would you tell a marketer to never claim in messaging?",
]

DIM_THEMES: Dict[str, List[str]] = {
    "Natural Attractions": [
        "sense of place and meaning",
        "quiet awe vs spectacle",
        "timing for contemplative experience",
        "routes that support reflection",
        "respectful behavior in nature",
        "access vs sacred calm",
    ],
    "Atmosphere": [
        "sacredness and emotional tone",
        "authenticity cues",
        "commercialization pressure",
        "silence, sound, and pace",
        "crowds and reverence",
        "ritual context shaping ambience",
    ],
    "Social Environment": [
        "being a guest and practicing humility",
        "consent and boundaries",
        "guide trust and cultural context",
        "respectful interaction with locals",
        "tourist behavior that disrupts",
        "learning without extracting",
    ],
    "Infrastructure": [
        "signage for etiquette",
        "visitor flow that protects sacred spaces",
        "frictionless but respectful access",
        "toilets/rest areas without degrading atmosphere",
        "transparent ticketing/donations",
        "accessibility with dignity",
    ],
    "Value for Money": [
        "fairness and transparency",
        "donations vs fees",
        "paying for guides as cultural mediation",
        "avoiding extractive 'spiritual packages'",
        "community benefit",
        "what feels worth paying for (context, respect, time)",
    ],
}


# -----------------------------
# Templates
# -----------------------------
def tmpl_single_dimension(
    d: str, theme: str, style: str, place_hint: str, context: str
) -> str:
    return (
        f"{style} your experience with {place_hint} in Bali during {context}. "
        f"From a {d} perspective, what stands out about {theme}—and why does it matter to you as a culture/spirit-oriented traveler?"
    )


def tmpl_laddering(d: str, theme: str, context: str, meaning: str) -> str:
    return (
        f"Think about a specific moment in Bali during {context} that left you with {meaning}. "
        f"What happened, how did you interpret it, and why did it feel meaningful? "
        f"Frame your answer through {d} (focus on {theme})."
    )


def tmpl_contrast(d: str, a: str, b: str, context: str, cue: str) -> str:
    return (
        f"Compare {a} versus {b} in Bali during {context}. "
        f"In terms of {d}, how do they differ for you as a respectful, culture/spirit-first traveler? "
        f"Use {cue} as a cue in your explanation."
    )


def tmpl_tradeoff(d1: str, d2: str, x: str, y: str, constraint: str) -> str:
    return (
        f"Under this constraint: {constraint}. "
        f"How do you trade off {x} versus {y} when choosing cultural/spiritual experiences in Bali? "
        f"Answer with examples touching {d1} and {d2}."
    )


def tmpl_marketer_advice(d: str, theme: str, constraint: str, dont_claim: str) -> str:
    return (
        f"If you had to advise a tourism marketer for culturally/spiritually interested travelers: under the constraint '{constraint}', "
        f"what should they understand about {d} (especially {theme})? "
        f"Also: what is one thing they should NOT claim in messaging because it would feel misleading or disrespectful—e.g., {dont_claim}?"
    )


def tmpl_etiquette_scenario(d: str, topic: str, context: str) -> str:
    return (
        f"Walk me through an etiquette situation related to {topic} in Bali during {context}. "
        f"What did you do, what did you avoid, and what would you want a marketer to communicate to travelers upfront? "
        f"Connect it to {d}."
    )


def tmpl_route_design(
    d: str, nature_hint: str, culture_hint: str, constraint: str
) -> str:
    return (
        f"Design a mini day-route that combines {nature_hint} and {culture_hint} under this constraint: {constraint}. "
        f"How would you protect atmosphere and respect while still making it accessible to culture/spirit-first travelers? Link your reasoning to {d}."
    )


def tmpl_probe_followup(base_q: str, probe: str) -> str:
    return f"{base_q} {probe}"


def pick_constraint(rng: random.Random) -> Tuple[str, str]:
    key, vals = rng.choice(CONSTRAINTS)
    return key, rng.choice(vals)


def pick_place_hint_for_dim(d: str, rng: random.Random) -> str:
    if d == "Natural Attractions":
        return rng.choice(NATURE_FOR_MEANING)
    return rng.choice(CULTURE_SPIRIT_SPACES)


# -----------------------------
# Generation
# -----------------------------
def generate_prompts(
    n: int,
    seed: int = 42,
    add_followups_ratio: float = 0.35,
    ensure_balance: bool = True,
) -> List[Dict]:
    rng = random.Random(seed)

    # Mix of question archetypes, all segment-targeted
    types = [
        ("single", 0.24),
        ("laddering", 0.18),
        ("contrast", 0.16),
        ("tradeoff", 0.18),
        ("marketer", 0.12),
        ("etiquette", 0.08),
        ("route", 0.04),
    ]
    type_names = [t for t, _ in types]
    type_weights = [w for _, w in types]

    prompts: List[Dict] = []
    seen = set()

    # Balanced dimension coverage
    dim_cycle = []
    if ensure_balance:
        per_dim = max(1, n // len(DIMENSIONS))
        for d in DIMENSIONS:
            dim_cycle.extend([d] * per_dim)
        while len(dim_cycle) < n:
            dim_cycle.append(rng.choice(DIMENSIONS))
        rng.shuffle(dim_cycle)

    # A small set of "don't claim" examples to anchor respectful marketing constraints
    DONT_CLAIM = [
        "guaranteed 'authentic spirituality' on demand",
        "a ceremony 'for tourists' as the main attraction",
        "access to sacred spaces without emphasizing etiquette and consent",
        "a 'hidden local ritual' framed as a product",
        "permission to photograph everything",
    ]

    def add_prompt(obj: Dict) -> bool:
        key = re.sub(r"\s+", " ", obj["prompt"].strip().lower())
        if key in seen:
            return False
        # hard filter: must include at least one segment anchor term
        anchors = [
            "respect",
            "sacred",
            "etiquette",
            "meaning",
            "authentic",
            "ceremony",
            "guest",
            "context",
            "spirit",
        ]
        if not any(a in key for a in anchors):
            return False
        seen.add(key)
        prompts.append(obj)
        return True

    max_attempts = n * 25
    attempts = 0

    while len(prompts) < n and attempts < max_attempts:
        attempts += 1

        d = (
            dim_cycle[len(prompts)]
            if ensure_balance and len(dim_cycle) > len(prompts)
            else rng.choice(DIMENSIONS)
        )
        theme = rng.choice(DIM_THEMES[d])
        style = rng.choice(INTERVIEW_STYLES)
        context = rng.choice(CONTEXTS)
        place_hint = pick_place_hint_for_dim(d, rng)
        c_key, c_val = pick_constraint(rng)

        t = rng.choices(type_names, weights=type_weights, k=1)[0]

        if t == "single":
            q = tmpl_single_dimension(d, theme, style, place_hint, context)
            obj = {
                "dimension": d,
                "type": "single",
                "prompt": q,
                "tags": [d, theme, context, "segment:culture-spirit"],
            }
            ok = add_prompt(obj)

        elif t == "laddering":
            meaning = rng.choice(MEANING_MAKING)
            q = tmpl_laddering(d, theme, context, meaning)
            obj = {
                "dimension": d,
                "type": "laddering",
                "prompt": q,
                "tags": [d, theme, context, "laddering", "segment:culture-spirit"],
            }
            ok = add_prompt(obj)

        elif t == "contrast":
            a, b = rng.choice(CONTRASTS)
            cue = rng.choice(AUTHENTICITY_CUES + CROWDING_COMMODIFICATION)
            q = tmpl_contrast(d, a, b, context, cue)
            obj = {
                "dimension": d,
                "type": "contrast",
                "prompt": q,
                "tags": [d, "contrast", context, "segment:culture-spirit"],
            }
            ok = add_prompt(obj)

        elif t == "tradeoff":
            d2 = rng.choice([x for x in DIMENSIONS if x != d])
            x, y = rng.choice(TRADEOFFS)
            q = tmpl_tradeoff(d, d2, x, y, c_val)
            obj = {
                "dimension": f"{d} + {d2}",
                "type": "tradeoff",
                "prompt": q,
                "tags": [d, d2, "tradeoff", c_key, "segment:culture-spirit"],
            }
            ok = add_prompt(obj)

        elif t == "marketer":
            dont_claim = rng.choice(DONT_CLAIM)
            q = tmpl_marketer_advice(d, theme, c_val, dont_claim)
            obj = {
                "dimension": d,
                "type": "marketer_advice",
                "prompt": q,
                "tags": [d, theme, "marketer", c_key, "segment:culture-spirit"],
            }
            ok = add_prompt(obj)

        elif t == "etiquette":
            topic = rng.choice(RITUAL_ETIQUETTE_TOPICS)
            q = tmpl_etiquette_scenario(d, topic, context)
            obj = {
                "dimension": d,
                "type": "etiquette",
                "prompt": q,
                "tags": [d, "etiquette", topic, context, "segment:culture-spirit"],
            }
            ok = add_prompt(obj)

        elif t == "route":
            nature_hint = rng.choice(NATURE_FOR_MEANING)
            culture_hint = rng.choice(CULTURE_SPIRIT_SPACES)
            q = tmpl_route_design(d, nature_hint, culture_hint, c_val)
            obj = {
                "dimension": d,
                "type": "route_design",
                "prompt": q,
                "tags": [d, "route", c_key, "segment:culture-spirit"],
            }
            ok = add_prompt(obj)

        else:
            ok = False

        # follow-up probe variant
        if ok and rng.random() < add_followups_ratio and len(prompts) < n:
            probe = rng.choice(FOLLOWUP_PROBES)
            q2 = tmpl_probe_followup(prompts[-1]["prompt"], probe)
            obj2 = {
                "dimension": prompts[-1]["dimension"],
                "type": prompts[-1]["type"] + "+probe",
                "prompt": q2,
                "tags": prompts[-1]["tags"] + ["probe"],
            }
            add_prompt(obj2)

    if len(prompts) < n:
        print(f"Warning: only generated {len(prompts)} unique prompts (requested {n}).")

    return prompts[:n]


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument(
        "--n",
        type=int,
        default=600,
        help="Number of prompts to generate (300–1000 recommended).",
    )
    ap.add_argument("--seed", type=int, default=42)
    ap.add_argument("--out", default="culture_spirit_interview_prompts.jsonl")
    ap.add_argument("--format", choices=["jsonl", "txt"], default="jsonl")
    ap.add_argument(
        "--no_balance",
        action="store_true",
        help="Disable balanced coverage across dimensions.",
    )
    ap.add_argument("--followups_ratio", type=float, default=0.35)
    args = ap.parse_args()

    prompts = generate_prompts(
        n=args.n,
        seed=args.seed,
        add_followups_ratio=args.followups_ratio,
        ensure_balance=not args.no_balance,
    )

    if args.format == "jsonl":
        with open(args.out, "w", encoding="utf-8") as f:
            for p in prompts:
                f.write(json.dumps(p, ensure_ascii=False) + "\n")
    else:
        with open(args.out, "w", encoding="utf-8") as f:
            for p in prompts:
                f.write(p["prompt"].strip() + "\n")

    print(f"Saved {len(prompts)} prompts to: {args.out} ({args.format})")


if __name__ == "__main__":
    main()