"""Simplified MRTKT/DKVMN-style knowledge tracing placeholder.
中文：MRTKT/DKVMN 风格的占位推理逻辑，输出下一题正确率与掌握度向量。
"""
from __future__ import annotations
from typing import Dict, List
from ...pydantic_models import KTResponse, StudentInteraction
def _time_decay_adjustment(timestamp_seq: List[float]) -> float:
if len(timestamp_seq) < 2:
return 1.0
gaps = [timestamp_seq[i] - timestamp_seq[i - 1] for i in range(1, len(timestamp_seq))]
avg_gap = sum(max(g, 0.0) for g in gaps) / len(gaps)
# 将时间间隔映射到 [0.7, 1.0] 的衰减系数
return max(0.7, min(1.0, 1.0 - avg_gap * 0.01))
def run_inference_kt(interactions: List[StudentInteraction]) -> KTResponse:
"""输入学生交互序列，输出下一题正确概率和知识掌握状态。"""
if not interactions:
return KTResponse(next_question_correct_prob=0.5, mastery={"K1": 0.5})
correct_seq = [it.correct for it in interactions]
time_seq = [it.timestamp for it in interactions]
accuracy = sum(correct_seq) / len(correct_seq)
time_decay = _time_decay_adjustment(time_seq)
base_prob = accuracy * 0.8 + 0.1
prob_with_decay = max(0.05, min(0.95, base_prob * time_decay))
# 生成一个简化的 3 维知识掌握字典
mastery_vec = {
"K1": round(prob_with_decay * 0.95, 3),
"K2": round(prob_with_decay * 1.05 if accuracy > 0.6 else prob_with_decay * 0.9, 3),
"K3": round(prob_with_decay, 3),
}
next_prob = max(0.05, min(0.95, prob_with_decay + 0.05 * (correct_seq[-1] - 0.5)))
return KTResponse(
next_question_correct_prob=round(next_prob, 3),
mastery=mastery_vec,
)
