Spaces:
Sleeping
Sleeping
File size: 3,514 Bytes
e4accbb 4c21555 e4accbb 4c21555 e4accbb 4c21555 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 | """Deterministic reward shaping and grading utilities."""
from __future__ import annotations
import re
from support_queue_env.models import GradingBreakdown, SupportQueueAction, TicketFeedback
from support_queue_env.tasks import TicketSpec
PRIORITY_ORDER = ["P1", "P2", "P3", "P4"]
SCORE_EPSILON = 0.001
def _open_unit_interval(score: float) -> float:
return round(min(max(score, SCORE_EPSILON), 1.0 - SCORE_EPSILON), 4)
def _normalize(text: str) -> str:
return re.sub(r"\s+", " ", text.lower()).strip()
def _contains_keywords(text: str, keywords: list[str]) -> int:
normalized = _normalize(text)
return sum(1 for keyword in keywords if keyword.lower() in normalized)
def _priority_score(expected: str, predicted: str) -> float:
if expected == predicted:
return 0.30
try:
distance = abs(PRIORITY_ORDER.index(expected) - PRIORITY_ORDER.index(predicted))
except ValueError:
return 0.0
if distance == 1:
return 0.15
return 0.0
def _queue_score(ticket: TicketSpec, predicted: str) -> float:
if predicted == ticket.expected_queue:
return 0.25
if predicted in ticket.acceptable_queues:
return 0.15
return 0.0
def _disposition_score(ticket: TicketSpec, predicted: str) -> float:
if predicted == ticket.expected_disposition:
return 0.20
if predicted in ticket.acceptable_dispositions:
return 0.10
return 0.0
def grade_ticket(ticket: TicketSpec, action: SupportQueueAction) -> TicketFeedback:
summary_hits = _contains_keywords(action.summary, ticket.summary_keywords)
response_hits = _contains_keywords(action.response, ticket.response_keywords)
penalty_hits = _contains_keywords(action.response, ticket.disallowed_keywords)
summary_score = 0.15 * (summary_hits / len(ticket.summary_keywords)) if ticket.summary_keywords else 0.15
response_score = 0.10 * (response_hits / len(ticket.response_keywords)) if ticket.response_keywords else 0.10
penalty = -0.10 if penalty_hits else 0.0
breakdown = GradingBreakdown(
priority_score=_priority_score(ticket.expected_priority, action.priority),
queue_score=_queue_score(ticket, action.queue),
disposition_score=_disposition_score(ticket, action.disposition),
summary_score=round(summary_score, 4),
response_score=round(response_score, 4),
penalty=penalty,
)
total = (
breakdown.priority_score
+ breakdown.queue_score
+ breakdown.disposition_score
+ breakdown.summary_score
+ breakdown.response_score
+ breakdown.penalty
)
breakdown.total = _open_unit_interval(total)
matched_summary = summary_hits if ticket.summary_keywords else 0
matched_response = response_hits if ticket.response_keywords else 0
feedback = (
f"priority={action.priority} target={ticket.expected_priority}; "
f"queue={action.queue} target={ticket.expected_queue}; "
f"disposition={action.disposition} target={ticket.expected_disposition}; "
f"summary_keywords={matched_summary}/{len(ticket.summary_keywords)}; "
f"response_keywords={matched_response}/{len(ticket.response_keywords)}"
)
return TicketFeedback(
ticket_id=ticket.ticket_id,
expected_priority=ticket.expected_priority,
expected_queue=ticket.expected_queue,
expected_disposition=ticket.expected_disposition,
breakdown=breakdown,
feedback=feedback,
)
|