ECサイトのゴールデンルートを発見する方法 – Google Analytics 4とPythonを活用して（第2回）

pip install markovclick

from markovclick.models import MarkovClickstream
from markovclick.utils import ModelDataFormatter

# ユーザーごとのページ遷移シーケンスを抽出
sequences = preprocessed_data.groupby('user_id')['page_path'].apply(list).tolist()

# Markov Chainを構築
mcf = MarkovClickstream(sequences)

# 遷移確率行列を取得
transition_matrix = mcf.calculate_transition_probabilities()

# 遷移確率行列を表示
print(transition_matrix)

from sklearn.cluster import KMeans

# クラスタ数を設定
num_clusters = 3

# k-meansのインスタンスを作成
kmeans = KMeans(n_clusters=num_clusters, random_state=0)

# クラスタリングを実行
kmeans.fit(feature_vectors)

# 各ユーザーにクラスタラベルを割り当て
cluster_labels = kmeans.labels_

# クラスタごとの遷移パターンを分析
for i in range(num_clusters):
    cluster_sequences = [seq for idx, seq in enumerate(sequences) if cluster_labels[idx] == i]
    cluster_mcf = MarkovClickstream(cluster_sequences)
    cluster_transition_matrix = cluster_mcf.calculate_transition_probabilities()
    
    # クラスタごとの遷移確率行列を表示
    print(f"Cluster {i+1} Transition Matrix:")
    print(cluster_transition_matrix)
    print("\n")

def calculate_conversion_rate(group_data, purchase_page):
    total_users = group_data['user_id'].nunique()
    total_purchases = group_data[group_data['page_path'] == purchase_page]['user_id'].nunique()
    conversion_rate = total_purchases / total_users
    return conversion_rate

purchase_page = "/purchase_complete"
for i in range(num_clusters):
    group_data = preprocessed_data[preprocessed_data['cluster_label'] == i]
    conversion_rate = calculate_conversion_rate(group_data, purchase_page)
    print(f"Conversion Rate for Cluster {i+1}: {conversion_rate:.2%}")

def calculate_group_revenue(group_data, purchase_page):
    purchase_data = group_data[group_data['page_path'] == purchase_page]
    total_revenue = purchase_data['revenue'].sum()
    return total_revenue

for i in range(num_clusters):
    group_data = preprocessed_data[preprocessed_data['cluster_label'] == i]
    group_revenue = calculate_group_revenue(group_data, purchase_page)
    print(f"Total Revenue for Cluster {i+1}: ${group_revenue:.2f}")

import pandas as pd
import numpy as np

# 仮想データのロード
ab_test_data = pd.read_csv("ab_test_data.csv")

# コンバージョン率を計算する関数
def calculate_conversion_rate(data, group_col, user_col, purchase_page):
    total_users = data[user_col].nunique()
    total_purchases = data[data['page_path'] == purchase_page][user_col].nunique()
    conversion_rate = total_purchases / total_users
    return conversion_rate

# A/Bテストの結果を評価する関数
def evaluate_ab_test_results(data, group_col, user_col, purchase_page):
    unique_groups = data[group_col].unique()
    for group in unique_groups:
        group_data = data[data[group_col] == group]
        conversion_rate = calculate_conversion_rate(group_data, group_col, user_col, purchase_page)
        print(f"Conversion Rate for {group}: {conversion_rate:.2%}")

# A/Bテストの結果を評価
group_col = "group"
user_col = "user_id"
purchase_page = "/purchase_complete"
evaluate_ab_test_results(ab_test_data, group_col, user_col, purchase_page)

import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from scipy.stats import ttest_ind

# 仮想データのロード
ab_test_results = pd.read_csv("ab_test_results.csv")

# A/Bテストの各グループの売上データを抽出
group_a_data = ab_test_results[ab_test_results['group'] == 'A']['revenue']
group_b_data = ab_test_results[ab_test_results['group'] == 'B']['revenue']

# 売上データの分布をプロット
sns.histplot(group_a_data, kde=True, label="Group A")
sns.histplot(group_b_data, kde=True, color='r', label="Group B")
plt.legend()
plt.xlabel("Revenue")
plt.ylabel("Frequency")
plt.title("Revenue Distribution for A/B Test Groups")
plt.show()

# t検定を実行して、グループ間の売上の違いを検証
t_stat, p_value = ttest_ind(group_a_data, group_b_data)
print(f"t-statistic: {t_stat}, p-value: {p_value}")

# 有意水準を設定（例: 0.05）
alpha = 0.05

# p値をもとに結果を解釈
if p_value < alpha:
    print("The difference between Group A and Group B is statistically significant.")
else:
    print("The difference between Group A and Group B is not statistically significant.")

import pandas as pd

# 遷移パターンと売上データの読み込み
transition_patterns = pd.read_csv('transition_patterns.csv')
sales_data = pd.read_csv('sales_data.csv')

# 遷移パターンごとの売り上げ効果を計算
pattern_sales = transition_patterns.merge(sales_data, on='user_id', how='left')
pattern_sales['revenue'] = pattern_sales['revenue'].fillna(0)

pattern_effect = pattern_sales.groupby('pattern_id').agg({'revenue': 'sum', 'user_id': 'count'})
pattern_effect['avg_revenue'] = pattern_effect['revenue'] / pattern_effect['user_id']
pattern_effect['conversion_rate'] = pattern_effect['revenue'].gt(0).sum() / pattern_effect['user_id'].sum()

top_10_percent = int(0.1 * len(pattern_effect))
golden_routes = pattern_effect.nlargest(top_10_percent, 'avg_revenue')

import pandas as pd
import scipy.stats as stats

# A/Bテストデータの読み込み
ab_test_data = pd.read_csv('ab_test_data.csv')

# コンバージョン数をカウント
conversion_count = ab_test_data.groupby('variant')['conversion'].sum()

# 各バリアントのユーザー数をカウント
user_count = ab_test_data['variant'].value_counts()

# 二項検定を実施
test_result = stats.binom_test(conversion_count, user_count)

print(test_result)