python 爬取京东指定商品评论并进行情感分析_Python

项目地址

https://github.com/DA1YAYUAN/JD-comments-sentiment-analysis

爬取京东商城中指定商品下的用户评论，对数据预处理后基于SnowNLP的sentiment模块对文本进行情感分析。

运行环境

Mac OS X
Python3.7 requirements.txt
Pycharm

运行方法

数据爬取（jd.comment.py)

启动jd_comment.py，建议修改jd_comment.py中变量user-agent为自己浏览器用户代理
输入京东商品完整URL
得到京东评论词云，存放于jd_ciyun.jpg（词云轮廓形状存放于jdicon.jpg)
得到京东评论数据，存放于jd_comment.csv

import os
import time
import json
import random
import csv
import re

import jieba
import requests
import numpy as np
from PIL import Image
import matplotlib.pyplot as plt
from wordcloud import WordCloud

# 词云形状图片
WC_MASK_IMG = "jdicon.jpg"
# 评论数据保存文件
COMMENT_FILE_PATH = "jd_comment.txt"
# 词云字体
WC_FONT_PATH = "/Library/Fonts/Songti.ttc"


def spider_comment(page=0, key=0):
    """
    爬取京东指定页的评价数据
    :param page: 爬取第几，默认值为0
    """

    url = "https://club.jd.com/comment/productPageComments.action?callback=fetchJSON_comment98vv4646&productId=" + key + "" 
          "&score=0&sortType=5&page=%s&pageSize=10&isShadowSku=0&fold=1" % page
    kv = {"user-agent": "Mozilla/5.0", "Referer": "https://item.jd.com/"+ key + ".html"}#原本key不输入值，默认为《三体》

    try:
        r = requests.get(url, headers=kv)
        r.raise_for_status()
    except:
        print("爬取失败")
    # 截取json数据字符串
    r_json_str = r.text[26:-2]
    # 字符串转json对象
    r_json_obj = json.loads(r_json_str)
    # 获取评价列表数据
    r_json_comments = r_json_obj["comments"]
    # 遍历评论对象列表
    for r_json_comment in r_json_comments:
        # 以追加模式换行写入每条评价
        with open(COMMENT_FILE_PATH, "a+") as file:
            file.write(r_json_comment["content"] + "
")
        # 打印评论对象中的评论内容
        print(r_json_comment["content"])


def batch_spider_comment():
    """
        批量爬取某东评价
        """
    # 写入数据前先清空之前的数据
    if os.path.exists(COMMENT_FILE_PATH):
        os.remove(COMMENT_FILE_PATH)
    key = input("Please enter the address:")
    key = re.sub("D","",key)
    #通过range来设定爬取的页面数
    for i in range(10):
        spider_comment(i,key)
        # 模拟用户浏览，设置一个爬虫间隔，防止ip被封
        time.sleep(random.random() * 5)


def cut_word():
    """
    对数据分词
    :return: 分词后的数据
    """
    with open(COMMENT_FILE_PATH) as file:
        comment_txt = file.read()
        wordlist = jieba.cut(comment_txt, cut_all=False)#精确模式
        wl = " ".join(wordlist)
        print(wl)
        return wl


def create_word_cloud():
    """44144127306
    生成词云
    :return:
    """
    # 设置词云形状图片
    wc_mask = np.array(Image.open(WC_MASK_IMG))
    # 设置词云的一些配置，如：字体，背景色，词云形状，大小
    wc = WordCloud(background_color="white", max_words=2000, mask=wc_mask, scale=4,
                   max_font_size=50, random_state=42, font_path=WC_FONT_PATH)
    # 生成词云
    wc.generate(cut_word())
    # 在只设置mask的情况下,你将会得到一个拥有图片形状的词云
    plt.imshow(wc, interpolation="bilinear")
    plt.axis("off")
    plt.figure()
    plt.show()
    wc.to_file("jd_ciyun.jpg")


def txt_change_to_csv():
    with open("jd_comment.csv", "w+", encoding="utf8", newline="")as c:
        writer_csv = csv.writer(c, dialect="excel")
        with open("jd_comment.txt", "r", encoding="utf8")as f:
            # print(f.readlines())
            for line in f.readlines():
                # 去掉str左右端的空格并以空格分割成list
                line_list = line.strip("
").split(",")
                print(line_list)
                writer_csv.writerow(line_list)

if __name__ == "__main__":
    # 爬取数据
    batch_spider_comment()

    #转换数据
    txt_change_to_csv()

    # 生成词云
    create_word_cloud()

模型训练（train.py）

准备正负语料集online_shopping_10_cats.csv，分别存入negative.txt和positive.txt
启动train.py，新建文件sentiment.marshal，存入训练后的模型
找到外部库中snownlp中sentiment模块，将训练得到的sentiment.marshal.3文件覆盖sentiment模块中自带的sentiment.marshal.3

# -*-coding:utf-8-*-

def train():
    from snownlp import sentiment
    print("开始训练数据集...")
    sentiment.train("negative.txt", "positive.txt")#自己准备数据集
    sentiment.save("sentiment.marshal")#保存训练模型
    #python2保存的是sentiment.marshal；python3保存的是sentiment.marshal.3
    "训练完成后，将训练完的模型，替换sentiment中的模型"

def main():
    train()  # 训练正负向商品评论数据集
    print("数据集训练完成！")

if __name__ == "__main__":
    main()

情感分析（sentiment.analysis.py）

启动sentiment.analysis.py
开始对jd_comment.csv中评论进行数据处理，处理后文件存入processed_comment_data.csv
sentiment模块根据sentiment.marshal.3对评论进行情感评分，评分结果存入result.csv
评分结果可视化，生成文件fig.png

from snownlp import sentiment
import pandas as pd
import snownlp
import matplotlib.pyplot as plt
from matplotlib.font_manager import FontProperties

#from word_cloud import word_cloud_creation, word_cloud_implementation, word_cloud_settings

def read_csv():
    """读取商品评论数据文件"""
    comment_data = pd.read_csv("jd_comment.csv", encoding="utf-8",
                               sep="
", index_col=None)
    #返回评论作为参数
    return comment_data


def clean_data(data):
    """数据清洗"""
    df = data.dropna()  # 消除缺失数据 NaN为缺失数据
    df = pd.DataFrame(df.iloc[:, 0].unique())  # 数据去重
    return df
    # print("数据清洗后：", len(df))


def clean_repeat_word(raw_str, reverse=False):
    """去除评论中的重复使用的词汇"""
    if reverse:
        raw_str = raw_str[::-1]
    res_str = ""
    for i in raw_str:
        if i not in res_str:
            res_str += i
    if reverse:
        res_str = res_str[::-1]
    return res_str


def processed_data(filename):
    """清洗完毕的数据，并保存"""
    df = clean_data(read_csv())#数据清洗
    ser1 = df.iloc[:, 0].apply(clean_repeat_word)#去除重复词汇
    df2 = pd.DataFrame(ser1.apply(clean_repeat_word, reverse=True))
    df2.to_csv(f"{filename}.csv", encoding="utf-8", index_label=None, index=None)


def train():
    """训练正向和负向情感数据集，并保存训练模型"""
    sentiment.train("negative.txt", "positive.txt")
    sentiment.save("seg.marshal")#python2保存的是sentiment.marshal；python3保存的是sentiment.marshal.3


sentiment_list = []

res_list = []


def test(filename, to_filename):
    """商品评论-情感分析-测试"""
    with open(f"{filename}.csv", "r", encoding="utf-8") as fr:
        for line in fr.readlines():
            s = snownlp.SnowNLP(line)
            #调用snownlp中情感评分s.sentiments
            if s.sentiments > 0.6:
                res = "喜欢"
                res_list.append(1)
            elif s.sentiments < 0.4:
                res = "不喜欢"
                res_list.append(-1)
            else:
                res = "一般"
                res_list.append(0)
            sent_dict = {
                "情感分析结果": s.sentiments,
                "评价倾向": res,
                "商品评论": line.replace("
", "")
            }
            sentiment_list.append(sent_dict)
            print(sent_dict)
        df = pd.DataFrame(sentiment_list)
        df.to_csv(f"{to_filename}.csv", index=None, encoding="utf-8",
                  index_label=None, mode="w")


def data_virtualization():
    """分析结果可视化，以条形图为测试样例"""
    font = FontProperties(fname="/System/Library/Fonts/Supplemental/Songti.ttc", size=14)
    likes = len([i for i in res_list if i == 1])
    common = len([i for i in res_list if i == 0])
    unlikes = len([i for i in res_list if i == -1])

    plt.bar([1], [likes], label="喜欢")#（坐标，评论长度，名称）
    plt.bar([2], [common], label="一般")
    plt.bar([3], [unlikes], label="不喜欢")

    x=[1,2,3]
    label=["喜欢","一般","不喜欢"]
    plt.xticks(x, label)

    plt.legend()#插入图例
    plt.xlabel("评价种类")
    plt.ylabel("评价数目")
    plt.title(u"商品评论情感分析结果-条形图", FontProperties=font)
    plt.savefig("fig.png")
    plt.show()
"""
def word_cloud_show():
    #将商品评论转为高频词汇的词云
    wl = word_cloud_creation("jd_comment.csv")
    wc = word_cloud_settings()
    word_cloud_implementation(wl, wc)
"""

def main():
     processed_data("processed_comment_data")#数据清洗
     #train()  # 训练正负向商品评论数据集

     test("jd_comment", "result")

     print("数据可视化中...")
     data_virtualization()  # 数据可视化

     print("python程序运行结束。")

if __name__ == "__main__":
    main()