使用snownlp对京东购物评论进行情感分析_京东评论 情感分析

使用snownlp对京东购物评论进行情感分析

目的：利用snownlp包，对京东评论进行情感分析
涉及： mysql，snownlp，pandas等工具包
代码结构如下：

注意：

• jd_comments.csv中存储的是从京东爬取的二手苹果收集的评论，共计589条，并已标注了好评和差评
• neg.txt 以及 pos.txt 是利用pandas对jd_comments的数据进行切分过滤得到的好评和差评的训练集
• origin_result.csv 以及 new_result.csv 为使用原始模型以及自己训练的模型的情感分析结果（正负）

1 nlp.py 代码如下：

import snownlp
from snownlp import SnowNLP
import pandas as pd
from snownlp import sentiment
import os
import random

current_path = os.getcwd()
df_train = pd.read_csv('jd_comments.csv', header=None)

if __name__ == '__main__':
    # 1 数据收集 执行一次即可
    # df_train = pd.read_csv('jd_comments.csv', header=None)
    # df_train[df_train[2] == 3].iloc[:, 1].to_csv('pos.txt', sep='\t', index=False)
    # df_train[df_train[2] == 1].iloc[:, 1].to_csv('neg.txt', sep='\t', index=False)

    # 2 训练
    # neg_path = os.path.abspath(os.path.join(os.getcwd(), 'neg.txt'))
    # pos_path = os.path.abspath(os.path.join(os.getcwd(), 'pos.txt'))
    # mod_path = os.path.abspath(os.path.join(os.getcwd(), 'sentiment.marshal'))
    # print(mod_path)
    # sentiment.train(neg_path, pos_path)
    # sentiment.save(mod_path)

    # print(pos_path)

    # 3 随机测试
    # rand = random.randint(0, df_train.shape[0])
    # print(list(df_train.iloc[rand]))
    #
    # df_test_text = df_train.iloc[rand, 1]
    # s = SnowNLP(df_test_text)
    # print(s.sentiments)
    # print(df_train.shape[0])

    # 4 拿训练集集进行测试 识别准确率
    prob_list = []
    for i in range(0, df_train.shape[0]):
        s = SnowNLP(df_train.iloc[i, 1])
        prob = round(s.sentiments)
        prob_list.append(prob)
        # print(type(s.sentiments))
    df_train[df_train.shape[1]] = prob_list
    columns = ['good', 'content', 'eval', 'created_at', 'prob']
    df_train.columns = columns

    df_train_result = df_train.loc[:, ['content', 'eval', 'prob']]
    df_train_result['eval'] = df_train_result['eval'].map({3: 1, 1: 0})
    accurate = df_train_result[df_train_result['eval'] == df_train_result['prob']].shape[0] / df_train_result.shape[0]
    # df_train_result.to_csv('new_result.csv', index=False)
    # print('使用 新 model 识别准确率为：', accurate)

    df_train_result.to_csv('origin_result.csv', index=False)
    print('使用 原始model 识别准确率为：', accurate)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55

注意：

• 如果进行了第2步骤，使用过了自己的数据进行训练。要注意保存生成的模型。
• 进行情感实例分析之前，需要修改模型的加载路径。（改为自己训练好的）

2 识别准确率

新训练的模型和原始模型的识别准确率对比如下：

使用 原始model 识别准确率为： 0.6932203389830508
使用 新model 识别准确率为： 0.9050847457627119
1
2

注意：上面是拿训练集进行测试的，正常情况是要使用新的数据来测试

3 上文中数据的收集

利用Python对京东网站二手苹果手机的评论进行爬取，并存入mysql数据库中（多余了），csv文件中。jd_crawler.py代码如下 ：

import requests
from config import *
import re
import json
import math
from retrying import retry
from sql import jd_sql
import csv
# import pandas

class Jd_Comments(object):
    def __init__(self):
        self.req = Request()
        self.start_url = "https://club.jd.com/comment/productCommentSummaries.action?referenceIds=37245978364&callback=jQuery2257914&_=1567839796282"
        self.base_url = "https://sclub.jd.com/comment/productPageComments.action?callback=fetchJSON_comment98vv1182&productId=37245978364&score=%s&sortType=5&page=%s" \
              "&pageSize=10&isShadowSku=0&fold=1"

        self.headers = {
            'Referer': 'https://item.jd.com/37245978364.html'
        }
        self.sql = jd_sql

    def get_num(self):
        proxies, headers = self.req.proxy(type2=0, headers=self.headers)
        response = requests.get(url=self.start_url, headers=headers, proxies=proxies)
        if response.status_code in [200, 201]:
            data = re.findall('jQuery2257914\((.*?)\);', response.text, re.S)[0]
            data_json = json.loads(data)
            CommentsCount = data_json.get('CommentsCount')[0]
            PoorCount = CommentsCount.get('PoorCount')
            GoodCount = CommentsCount.get('GoodCount')
            CommentCount = CommentsCount.get('CommentCount')
            return GoodCount, PoorCount, CommentCount
        else:
            print("获取好评数以及差评数失败！！")

    def create_urls(self, rank, num):
        urls = []
        for i in range(0, math.ceil(num/10)):
            url = self.base_url%(rank, i)
            print(url)
            urls.append(url)
        return urls

    @retry(stop_max_attempt_number=3, wait_random_min=1000, wait_random_max=2000)
    def spider_one(self, rank, url):
        proxies, headers = self.req.proxy(type2=0, headers=self.headers)
        response = requests.get(url=url, headers=headers, proxies=proxies)
        if response.status_code in [200, 201] and response.text:
            data = re.findall('fetchJSON_comment98vv1182\((.*?)\);', response.text, re.S)[0]
            data_json = json.loads(data)
            comments = data_json.get('comments')
            if comments:
                items = []
                for one in comments:
                    item = {}
                    item['good'] = '二手apple ' + one.get('referenceId')
                    item['content'] = one.get('content').replace('\n', '')
                    item['eval'] = rank
                    item['created_at'] = one.get('creationTime')
                    # 数据转存至mysql数据库中
                    # self.sql.insert_one(item)

                    # 数据保存到本地的csv文件中
                    self.save_to_csv(filename='jd_comments', item=item.values())
                    print(item)
                    if item:
                        items.append(item)
                return items
            else:
                return False
        else:
            print("访问失败！！")
            
    def spider_many(self):
        GoodCount, PoorCount, CommentCount = self.get_num()
        print(GoodCount, PoorCount, CommentCount)
        good_urls = self.create_urls(rank=3, num=GoodCount)
        for g_url in good_urls:
            result = self.spider_one(rank=3, url=g_url)
            if not result:
                break
        # print("111111111111111111111111111111111111111111111111111")
        poor_urls = self.create_urls(rank=1, num=PoorCount)
        page = 0
        for p_url in poor_urls:
            print('page: ', page)
            page += 1
            result = self.spider_one(rank=1, url=p_url)
            if not result:
                break
    def save_to_csv(self, filename=None, item=[]):
        name = '{}.csv'.format(filename)
        with open(name, 'a', encoding='utf-8-sig') as f:
            f_csv = csv.writer(f)
            f_csv.writerow(item)
            # f_csv.writerows(rows)

if __name__ == '__main__':
    jd_comment = Jd_Comments()
    jd_comment.spider_many(）
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101

得到的数据保存在mysql数据库以及jd_comments.csv文件中。如下（mysql）：