Star's Blog

1
weekday_counts = df.groupby(["month", "weekday"]).size().unstack().fillna(0)
2

3
weekday_order = ["Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "Saturday", "Sunday"]
4
weekday_counts = weekday_counts[weekday_order]
5

6
plt.figure(figsize=(12, 6))
7
sns.heatmap(weekday_counts.T, cmap="YlGnBu", annot=True, fmt=".0f")
8
plt.title("各月中不同星期的聊天记录热力图")
9
plt.xlabel("月份")
10
plt.ylabel("星期")
11
plt.tight_layout()
12
plt.show()

1
# 确保字段准备好
2
df["hour"] = df["time"].dt.hour
3

4
# 分2小时一段
5
bins = list(range(0, 25, 2))  # 0-2, 2-4, ..., 22-24
6
labels = [f"{h:02d}-{h+2:02d}" for h in bins[:-1]]
7
df["time_bin"] = pd.cut(df["hour"], bins=bins, labels=labels, right=False)
8

9
# 统计每月每时间段的聊天数
10
pivot = df.pivot_table(index="month", columns="time_bin", aggfunc="size", fill_value=0, observed=False)
11

12
# 计算百分比
13
percent = pivot.div(pivot.sum(axis=1), axis=0)
14

15
# 按时间段顺序排列列
16
percent = percent[labels]  # 保证顺序一致
17

18
sorted_index = pivot.sum(axis=1).sort_values().index
19
percent = percent.loc[sorted_index]
20

21
# 绘图
22
fig, ax = plt.subplots(figsize=(12, len(percent)*0.6))
23

24
left = [0] * len(percent)  # 初始位置
25
colors = sns.color_palette("Spectral", len(labels))  # 可自定义色板
26

27
for i, col in enumerate(percent.columns):
28
    ax.barh(percent.index, percent[col], left=left, label=col, color=colors[i])
29
    left = [l + p for l, p in zip(left, percent[col])]
30

31
# 标注总聊天数（右侧）
32
totals = pivot.sum(axis=1).loc[sorted_index]
33
for i, (y, total) in enumerate(zip(percent.index, totals)):
34
    ax.text(1.01, i, f"{total}条", va="center", fontsize=10)
35

36
ax.set_title("各月聊天时间段分布（百分比）")
37
ax.set_xlabel("比例")
38
ax.set_xlim(0, 1)
39
ax.set_ylabel("月份")
40
ax.legend(
41
    title="时间段",
42
    loc="upper center",
43
    bbox_to_anchor=(0.5, -0.12),
44
    ncol=6,
45
    frameon=False
46
)
47
plt.tight_layout()
48
plt.show()

1
import jieba
2
from wordcloud import WordCloud, ImageColorGenerator
3
from PIL import Image
4
import numpy as np
5
import matplotlib.pyplot as plt
6
from ipywidgets import interact, widgets
7

8
# 读取停用词
9
with open('stopwords.txt', 'r', encoding='utf-8') as f:
10
    stopwords = set(line.strip() for line in f if line.strip())
11

12
# 词云生成函数
13
def generate_wordcloud(month, user):
14
    # 筛选数据
15
    if month == "全部" and user == "全部":
16
        sub_df = df
17
    elif month == "全部":
18
        user_bool = (user == "用户1")
19
        sub_df = df[df["user"] == user_bool]
20
    elif user == "全部":
21
        sub_df = df[df["month"] == month]
22
    else:
23
        user_bool = (user == "用户1")
24
        sub_df = df[(df["month"] == month) & (df["user"] == user_bool)]
25
    # 拼接所有内容
26
    text = "\n".join(str(x) for x in sub_df["content"] if pd.notnull(x) and str(x).strip())
27
    # 分词并去除停用词
28
    segs = [seg.strip() for seg in jieba.cut(text) if seg.strip() and seg not in stopwords]
29
    if not segs:
30
        print("无内容可生成词云")
31
        return
32
    # 词云参数
33
    coloring = np.array(Image.open("mask.jpeg"))
34
    wc = WordCloud(
35
        font_path='fonts/NotoSansSC-Regular.ttf',
36
        background_color="white",
37
        max_words=400,
38
        mask=coloring,
39
        stopwords=stopwords,
40
        max_font_size=400,
41
        random_state=42
42
    )
43
    wc.generate(' '.join(segs))
44
    image_colors = ImageColorGenerator(coloring)
45
    wc.recolor(color_func=image_colors)
46
    # 展示
47
    plt.figure(figsize=(10, 10))
48
    plt.imshow(wc, interpolation='bilinear')
49
    plt.axis("off")
50
    plt.title(f"{month} - {user} 词云")
51
    plt.show()
52

53
# 交互式界面
54
month_options = ["全部"] + sorted(df["month"].unique())
55
user_options = ["全部", "用户1", "用户2"]
56
interact(
57
    generate_wordcloud,
58
    month=widgets.Dropdown(options=month_options, description="月份"),
59
    user=widgets.Dropdown(options=user_options, description="用户")
60
)

1
df = df.sort_values("time")
2
df = df.reset_index(drop=True)
3

4
# 时间差（单位：秒）
5
df["time_diff"] = df["time"].diff().dt.total_seconds().fillna(0)
6

7
# 若时间间隔超过 N 秒，则视为新会话
8
session_threshold = 600  # 单位：秒
9
df["session_id"] = (df["time_diff"] > session_threshold).cumsum()
10

11
# 每个 session 的起止时间与持续时长
12
sessions = df.groupby("session_id").agg(
13
    start_time=("time", "first"),
14
    end_time=("time", "last"),
15
    message_count=("time", "count")
16
)
17

18
sessions["duration_sec"] = (sessions["end_time"] - sessions["start_time"]).dt.total_seconds()
19
sessions["duration_min"] = sessions["duration_sec"] / 60  # 转成分钟
20

21
# 过滤掉只发一条消息的聊天
22
sessions = sessions[sessions["message_count"] > 1]
23

24
# 对数直方图：log1p(duration)
25
log_values = np.log1p(sessions["duration_min"])
26

27
plt.figure(figsize=(10, 6))
28
plt.hist(log_values, bins=40, color="#f48fb1", edgecolor="black")
29

30
# 设置横轴刻度：原始时间（分钟）
31
ticks_raw = [1, 5, 10, 30, 60, 120, 300]  # 你可以按需要改
32
ticks_log = np.log1p(ticks_raw)
33
tick_labels = [str(t) for t in ticks_raw]
34

35
plt.xticks(ticks=ticks_log, labels=tick_labels)
36

37
plt.xlabel("每次聊天持续时间（分钟）")
38
plt.ylabel("聊天轮次数")
39
plt.title("聊天持续时间分布（对数压缩，横轴显示原始分钟）")
40
plt.tight_layout()
41
plt.show()

1
# 标记每个 session 的持续时间（单位：分钟）
2
session_duration = df.groupby("session_id")["time"].agg(["first", "last"])
3
session_duration["duration_min"] = (session_duration["last"] - session_duration["first"]).dt.total_seconds() / 60
4
long_sessions = session_duration[session_duration["duration_min"] > 30].index
5

6
# 仅保留长 session 的数据
7
df_long = df[df["session_id"].isin(long_sessions) & df["content"].notna()].copy()
8
df_long["text_length"] = df_long["content"].astype(str).str.len()
9

10
# 分组统计每个 session 每个用户的文本总长度
11
length_by_user = df_long.groupby(["session_id", "user"])["text_length"].sum().unstack(fill_value=0)
12

13
# 重命名列，避免布尔列名引起混淆
14
length_by_user = length_by_user.rename(columns={True: "用户1", False: "用户2"})
15

16
# 增加总量和主导者列
17
length_by_user["total"] = length_by_user["用户1"] + length_by_user["用户2"]
18
length_by_user["dominant_user"] = length_by_user[["用户1", "用户2"]].idxmax(axis=1)
19

20
length_by_user[["用户1", "用户2"]].plot(kind="bar", stacked=True, figsize=(14, 6), color=["#8ecae6", "#ffb3b3"])
21
plt.title("每轮聊天中双方发言的文本长度（大于30分钟）")
22
plt.xticks([])
23
plt.xlabel("session")
24
plt.ylabel("文本长度")
25
plt.legend()
26
plt.tight_layout()
27
plt.show()

1
conda install snownlp

1
from snownlp import SnowNLP
2

3
def get_sentiment(text):
4
    try:
5
        s = SnowNLP(text)
6
        return s.sentiments # 0 ~ 1 越大越积极
7
    except:
8
        return None
9

10
df["sentiment"] = df["content"].astype(str).apply(get_sentiment)