呼叫日志FTR7%计算异常：L2列值差异导致数值波动排查

我用Python的pandas和numpy库基于呼叫日志数据集计算首次解决率（First Time Resolution，FTR）百分比。处理calls_logs_cleaned_2025-05-02.csv时，FTR7%约为63.8%；但处理另一个仅填充了L2列有效值（原文件L2列全为空）的CSV文件时，FTR7%骤降至约39.8%。

我的FTR计算逻辑完全没用到l1_intent和l2_intent列，所以对结果差异感到困惑。CSV文件包含列：customer_account_no、agent_id、agent_name、start_time、l1_intent、l2_intent。

计算代码 #

import pandas as pd
import numpy as np

# Loading data
df = pd.read_csv('calls_logs_cleaned_2025-05-02.csv')

# Converting start_time to datetime
df['start_time'] = pd.to_datetime(df['start_time'], format='mixed', dayfirst=True)

# Sorting data by customer and time
df = df.sort_values(by=['customer_account_no', 'start_time']).reset_index(drop=True)

# Function to calculate FTR7 and FTR14
def calculate_ftr(group):
    times = group['start_time'].values.astype('datetime64[ns]')
    n = len(times)
    ftr7, ftr14 = np.ones(n, dtype=int), np.ones(n, dtype=int)
    
    for i in range(1, n):
        current_time = times[i]
        
        # Check if there is a previous call within 7 days
        left_bound_7 = current_time - np.timedelta64(7, 'D')
        idx_7 = np.searchsorted(times[:i], left_bound_7, side='right')
        ftr7[i] = 0 if idx_7 < i else 1
        
        # Check if there is a previous call within 14 days
        left_bound_14 = current_time - np.timedelta64(14, 'D')
        idx_14 = np.searchsorted(times[:i], left_bound_14, side='right')
        ftr14[i] = 0 if idx_14 < i else 1
    
    return pd.DataFrame({'FTR7': ftr7, 'FTR14': ftr14}, index=group.index)

# Applying function per customer group
df[['FTR7', 'FTR14']] = df.groupby('customer_account_no', group_keys=False).apply(calculate_ftr)

# Calculate overall percentages
ftr7_percent = df['FTR7'].mean() * 100
ftr14_percent = df['FTR14'].mean() * 100

print(f"FTR7%: {ftr7_percent:.1f}%")
print(f"FTR14%: {ftr14_percent:.1f}%")

问题 #

1. 为何填充了有效L2值的文件会让FTR7%从_63.8%降到39.8%？
2. 代码里没显式用到的L2列值会不会影响计算结果？
3. 需要检查或调试哪些内容才能弄明白这个差异？

样本数据 #

内容的提问来源于stack exchange，提问作者IAIMT2024