airbnb北京民宿运营情况分析

airbnb北京民宿运营状况分析

导入数据

import numpy as np
import matplotlib.pyplot as plt
from scipy.special import jn
from IPython.display import display, clear_output
import time
import pandas as pd
import seaborn as sns
from sklearn.preprocessing import StandardScaler
plt.rcParams['font.sans-serif'] = [u'SimHei']
plt.rcParams['axes.unicode_minus'] = False

listing = r'C:\Users\Administrator\Desktop\数据分析\listings.csv'
neighbour = r'C:\Users\Administrator\Desktop\数据分析\neighbourhoods.csv'
review = r'C:\Users\Administrator\Desktop\数据分析\reviews_detail.csv'
neigh_geo = r'C:\Users\Administrator\Desktop\数据分析\neighbourhoods.geojson'
reviews = pd.read_csv(review)
listing = pd.read_csv(listing)
neighbour = pd.read_csv(neighbour)

step1查看数据结构，处理缺失值#查看各个数据结构

#查看各特征丢失缺失情况
#listing.isnull().sum()
#neighbourhood_group 整列缺失，删除
#listing = listing.drop(‘neighbourhood_group’, axis=1)
#删除name为空的记录,只有1条
#listing = listing[-listing[‘name’].isnull()]

print('listing->数据量: %d, 维度: %d' %(listing.shape[0], listing.shape[1]))
print('reviews->数据量: %d, 维度: %d' %(reviews.shape[0], reviews.shape[1]))
print('neighbour->数据量: %d, 维度: %d' %(neighbour.shape[0], neighbour.shape[1]))

listing->数据量: 28452, 维度: 16
reviews->数据量: 202099, 维度: 6
neighbour->数据量: 16, 维度: 2

#listing.info() #查看具体信息, 发现neighbourhood_group字段整列缺失，后续可以删除
#neighbour neighbour表包含北京各区名字
#reviews.info() # reviews包含200k+ 的评论信息， 中英文都有，后续可以考虑做情感分析判断民宿的满意度

listing = listing.drop(['neighbourhood_group'],axis=1)

step2 统计民宿分布情况

res = listing.groupby('neighbourhood') #按区域分组
area_list = neighbour['neighbourhood'].to_list()
neighb_counts = {}
for area in  area_list:
    counts = res.get_group(area).shape[0]
    neighb_counts[area] = counts

#发现东城+海淀+朝阳的民宿数量占比超过60%

plt.figure(figsize=(11,11))
plt.title('各区民宿分布情况')
label = [i[0] for i in neighb_counts.items()]
data = [i[1] for i in neighb_counts.items()]
max_area = sorted(neighb_counts.items(), key=lambda x:x[1],reverse=True)[:3] #选择民宿最多的三个区突出显示
max_area = [i[0] for i in max_area]
explode = [0.1 if i in max_area else 0 for i in label]
color = sns.color_palette("RdBu", len(label))
plt.pie(data, labels = label, explode = explode, autopct='%.2f%%', colors=color)
plt.legend(loc='best', fontsize=6.5)
plt.show()

可以看到，超过60%的房源都集中在朝阳，海淀，东城三个核心区域

step3 去除离群值，做相关性分析

#通过观察数据head 5行发现room_type值有重复， 说明这应该是一个分类字段, 并且是文字信息，后续处理时应该map 到数字编码
listing[“room_type”].value_counts()
#task2 相关性分析， 做相关性分析前可以先检查一下数据分布， 发现离群值,这里重点观察价格

listing['price'].describe()

plt.figure(figsize=(7,1.5))
plt.title('价格分布情况')
sns.boxplot(listing['price'],whis=0.5, palette="Blues")

**加粗

#listing[listing['price']>20000].shape  #price>20000的有21条数据
#listing[listing['price']==0].shape     #price==0的有3条数据 ；  删除这些离群值
listing = listing[listing['price']!=0]
listing = listing[listing['price']<=20000]

copy_listing = listing.copy() #复制原表
zscore = StandardScaler()
scale_features = ['price', 'reviews_per_month', 'number_of_reviews']
copy_listing[scale_features] = zscore.fit_transform(copy_listing[scale_features])

#热力图分析相关性, 重点分析价格和其他变量的相关性，显然房间编号和房主编号没有影响先去除这两个特征
room_type 和 neighbour都是文字信息，应该map 到数字编码
观察热力图发现价格与维度有比较明显的正相关， 说明在北京地区越靠北的民宿价格越高

no_corr = ['id', 'host_id']
columns = [feature for feature in copy_listing.columns if feature not in no_corr]
sub_df = copy_listing[:][columns] 

room_type = sub_df['room_type'].unique()
type_label = {k:v for v,k in enumerate(room_type)}  #0: 整租/公寓, 1:单人房间 2:合租

#关于区域：按各区民宿数量排序，然后进行map

sorted_area = sorted(neighb_counts.items(), key=lambda x:x[1])
sorted_area = [i[1] for i in sorted_area]
neighbour_label = {k:v for v,k in enumerate(sorted_area)}

sub_df['neighbour_label'] = sub_df['neighbourhood'].map(neighbour_label)
sub_df['room_label'] = sub_df['room_type'].map(type_label)

#相关矩阵

corr_matrix = sub_df.corr()
plt.figure(figsize=(10, 5))
sns.set(font_scale=1.5)
sns.heatmap(corr_matrix, vmin=-1, vmax=1, cmap=sns.color_palette('RdBu', n_colors=corr_matrix.size))
plt.title('Correlation heat map')

观察热力图发现， 房屋价格与纬度有正相关关系，说明在北京地区越往北的房子价格越高

step4 查看价格分布情况

#创建区域-价格透视表# 观察发现海淀和朝阳两区性价比最好， 平均价格低而且处于城中心；
下边应该分析房间类型的区域分布

neighbourhood_price=pd.pivot_table(listing,index="neighbourhood",values="price",aggfunc=np.mean)
neighbourhood_price.head()
plt.figure(figsize=(30,10),dpi=80)
plt.bar(neighbourhood_price.index,neighbourhood_price.price,color="b") 
plt.xticks(rotation=45)  # 旋转坐标标签

通过透视图可以发现海淀和朝阳两区性价比最好， 平均价格低而且处于城中心

step5 找出运营状况良好的民宿并分析原因

假设评论数量高的房源为运营状况良好的，所以我们假定评论数量大于整体评论数量分布的Q3分位数的为运营良好的房源

小于整体评论数量分布的Q1分位数的为运营较差的房源

找出运营状态好的民宿的共有特征， 因为上图中我们看到总评论数和每月平均评论数有强正相关，平均每月评论数存在空值
#所以选取总评论数作为评价指标
假设平均总评论数大于4分位值的民宿是运营状态好的民宿

median = listing['number_of_reviews'].quantile(0.75)
median2 = listing['number_of_reviews'].quantile(0.25)
good_house = listing[listing['number_of_reviews']>median]
worse_house = listing[listing['number_of_reviews']<=median2]

print(good_house['price'].mean())
print(worse_house['price'].mean())

446
698

from folium.plugins import HeatMap
import folium
world_map = folium.Map()
lati = 39.9
longti = 116.3
beijing_map = folium.Map(location=[lati, longti], zoom_start=12)

sub_data = good_house.sort_values(by="reviews_per_month" , ascending=False)
heat_data = sub_data.iloc[:200][['latitude', 'longitude']].values.tolist()
HeatMap(heat_data).add_to(beijing_map)
beijing_map

运营状况好的房源平均价格为447元， 运营状况较差的平均价格为698元
运营状况好的民宿主要集中在中心城区

#统计name的描述长度与经营状态的关系,
#判断是否含中文， 如果没有中文则分词， 如果有中文则直接返回字符串长度
#发现运营状态好的房源的平均name描述长度大于运营状态不好的，说明应该认真写房源的名字

import jieba 
import numpy as np
def is_contain_chinese(check_str):
    for ch in check_str:
        if u'\u4e00' <= ch <= u'\u9fff':
            return True
    return False

def count_words(data):
    word_list = data['name'].values.tolist()
    counts = []
    for sent in word_list:
        try:
            if is_contain_chinese(sent)==True:
                counts.append(len(sent))
            else:
                seg = sent.split()
                counts.append(len(seg))
        except Exception as e:
            counts.append(0)
    return np.mean(counts)

print('good_house_words_count: ', count_words(good_house))
print('bad_house_words_count: ', count_words(worse_house))

30.18
22.8

数据中name字段为该民宿的描述信息，运营状况好的民宿平均用30个词来描述自己的房子，运营状况差的平均用23个词

词云分析

from collections import Counter
stop_words = ['着', '了', '过','的', '得', '一']
def words2dict(data):
    words =[]
    res = data['name'].values.tolist()
    for i in res:
        try:
            if is_contain_chinese(i) == True:
                seg_list = jieba.cut(i, cut_all=False, HMM=True)
                for word in seg_list:
                    if is_contain_chinese(word) == True and word not in stop_words:
                        words.append(word)
        except Exception as e:
            pass

    c = Counter(words)
    return c

bad_c = words2dict(worse_house)
c = words2dict(good_house)

从词云可以看出， 运营状态好的房源着重突出天安门， 三里屯，故宫，国贸等地标性建筑或者商圈

from PIL import Image
from wordcloud import WordCloud
font = r'C:\Windows\Fonts\simfang.ttf'
png = 'twitt.png'  # <- 词云要填充的图案，我这里用推特小鸟
fig = plt.gcf()
fig.set_size_inches(16, 14)

fig1 = plt.subplot(121)  # 1 行 3列 第 1张图
mask = np.array(Image.open(png))                        # img -> array, 词云的填充对象
wc1 = WordCloud(background_color='white', mask=mask, font_path=font).generate_from_frequencies(c)
plt.imshow(wc1)
plt.title('good house chinese words')
plt.axis('off')            #关闭边框线

fig1 = plt.subplot(122)  # 1 行 3列 第 1张图
mask = np.array(Image.open(png))                        # img -> array, 词云的填充对象
wc1 = WordCloud(background_color='white', mask=mask, font_path=font).generate_from_frequencies(bad_c)
plt.imshow(wc1)
plt.title('bad house chinese words')
plt.axis('off')            #关闭边框线

通过词云发现， 运营状况好的民宿更喜欢突出国贸，天安门，故宫，三里屯等地理指标性名词

运营状况最好的民宿

median_best = good_house['number_of_reviews'].quantile(0.75)
best_house = good_house[good_house['number_of_reviews']>median_best]
best_house = best_house.sort_values(by='number_of_reviews',ascending=False)  # 按评论数排序

counts = {}
room_typ_good = best_house.groupby('neighbourhood')
for typ in area_list:
    try:
        res = room_typ_good.get_group(typ)
        counts[typ] = res.shape[0]
    except Exception as e:
        counts[typ] = 0
        
best_house['price'].describe()

count 1669.000000
mean 410.194727
std 396.428221
min 60.000000
25% 208.000000
50% 349.000000
75% 497.000000
max 9998.000000
Name: price, dtype: float64

分布图

plt.figure(figsize=(11,11))
plt.title('运营状况最好的民宿分布图')
label = [i[0] for i in counts.items()]
data = [i[1] for i in counts.items()]
max_area = sorted(counts.items(), key=lambda x:x[1],reverse=True)[:3] #选择民宿最多的三个区突出显示
max_area = [i[0] for i in max_area]
explode = [0.1 if i in max_area else 0 for i in label]
color = sns.color_palette("RdBu", len(label))
plt.pie(data, labels = label, explode = explode, autopct='%.2f%%', colors=color)
plt.legend(loc='best', fontsize=6.5)
plt.show()

收入情况

best_house['income'] = best_house['price'] * best_house['reviews_per_month']
best_house['income'].describe()

count 1669.000000
mean 1286.337837
std 1274.554455
min 68.580000
25% 517.000000
50% 916.500000
75% 1683.400000
max 22995.400000
Name: income, dtype: float64

一个简单的svm分类器， 预测房屋的运营状况, 标签列：good为1， worse为0

number_of_reviews和reviews_per_month都是与分类标准有高度相关的特征， 不可以使用

#做一个简单的svm分类器， 预测什么样的房子会经营成功

are_map = {key:value for value,key in enumerate(area_list)}
features_no = ['id', 'name', 'host_id', 'host_name', 'last_review', 'number_of_reviews','reviews_per_month']
good_house_sub = good_house.drop(columns=features_no)
worse_house_sub = worse_house.drop(columns=features_no)

#添加tag列， 1为good， map room_type, neighbourhood为数字

good_house_sub['tag']=1
good_house_sub['room_type'] = good_house_sub['room_type'].map(type_label)
good_house_sub['neighbourhood'] = good_house_sub['neighbourhood'].map(are_map)

worse_house_sub['tag']=0
worse_house_sub['room_type'] = worse_house['room_type'].map(type_label)
worse_house_sub['neighbourhood'] = worse_house['neighbourhood'].map(are_map)
worse_house_sub.fillna(0, inplace=True)

df_data = pd.concat([good_house_sub, worse_house_sub], axis=0)

import numpy as np
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.utils import shuffle
from sklearn.svm import SVC
from sklearn.model_selection import train_test_split
df_data = shuffle(df_data)

y = df_data['tag']
X = df_data.drop(columns=['tag'])
X_train,X_test,y_train,y_test = train_test_split(X,y,test_size=0.3)
clf = make_pipeline(StandardScaler(), SVC(gamma='auto'))
clf.fit(X_train, y_train)
clf.score(X_test, y_test)  

0.7152971865101546

准确率是0.715

总结

在本次airbnb民宿运营情况分析中， 通过区域分布分析， 发现民宿集中在朝阳东城海淀三区

用评论数量来衡量民宿运营情况 运营状态最好的民宿集中在朝阳东城两区
运营状态好的民宿的房源描述普遍比较详细

北京民宿的价格与纬度有正相关关系，通常越靠北的房子房价越贵
运营状态最好的民宿价格集中在208-497元之间

运营状况好的民宿的房源描述中多有天安门，国贸，三里屯等北京地标性地理名词

把价格 * 平均月评论数看作收入的话， 发现运营情况最好的民宿的月收入集中在 517-1683元之间

用经纬度，价格，房间类型等特征训练了一个svm分类器,可以预测民宿的运营情况

阿里数据集
[1]: https://tianchi.aliyun.com/competition/gameList/coupleList