Scraping iQiyi Bullet Comments and Generating a Word Cloud with Python

Recently a popular mystery drama "The Hidden Corner" (Douban rating 9.0) was selected for analysis; the author crawled the bullet comments of its first episode from iQiyi and created a word cloud to visualize audience feedback.

The article is divided into two parts: (1) crawling the bullet comments from iQiyi, and (2) processing the comments and generating a word cloud.

iQiyi bullet files are harder to crawl because the downloaded files appear as garbled binary data. The author explains how to open the browser’s Network panel, search for "bullet", locate the binary files, and note that each episode loads a new bullet file every 5 minutes.

The URL pattern for bullet files is:

https://cmts.iqiyi.com/bullet/{tvid_first_two}/{tvid_last_two}/{tvid}_300_{x}.z

where x is the ceiling of total duration divided by 300 seconds (5‑minute intervals). For the first episode (77 minutes) this results in 16 files.

Scraping code (Python):

import zlib
import requests
for x in range(16):
    x += 1
    url = 'https://cmts.iqiyi.com/bullet/92/00/9000000005439200_300_' + str(x) + '.z'
    bulletold = requests.get(url).content  # garbled binary
    bulletnew = bytearray(bulletold)        # re‑encode binary
    xml = zlib.decompress(bulletnew, 15+32).decode('utf-8')
    with open('./iqiyi' + str(x) + '.xml', 'a+', encoding='utf-8') as f:
        f.write(xml)
    f.close()

The resulting XML files contain content fields that hold the actual comments. To extract these, the following code is used:

from xml.dom.minidom import parse
import xml.dom.minidom
for x in range(16):
    x += 1
    DOMTree = xml.dom.minidom.parse(r"C:\Users\dmj\PycharmProjects\test\iqiyi" + str(x) + ".xml")
    collection = DOMTree.documentElement
    entrys = collection.getElementsByTagName("entry")
    for entry in entrys:
        content = entry.getElementsByTagName('content')[0]
        i = content.childNodes[0].data
        with open("dan_mu.txt", mode="a+", encoding="utf-8") as f:
            f.write(i)
            f.write("
")

The extracted dan_mu.txt file contains all bullet comments, which are then processed for word‑cloud generation.

Word‑cloud creation uses the wordcloud and jieba libraries. The code performs Chinese word segmentation, removes stop words, and generates the cloud:

from wordcloud import WordCloud
import jieba
import matplotlib.pyplot as plt
with open('./dan_mu.txt', encoding='utf-8', mode='r') as f:
    myText = f.read()
myText = " ".join(jieba.cut(myText))
words = myText.split(" ")
# remove unwanted tokens
for i in range(len(words)-1, -1, -1):
    if len(words[i]) == 1 or words[i] in ["这个", "不是", "这么", "怎么", "这是", "还是"]:
        words.pop(i)
myText = " ".join(words)
wordcloud = WordCloud(background_color="white", font_path="simsun.ttf", height=300, width=400).generate(myText)
plt.imshow(wordcloud)
plt.axis("off")
plt.show()
wordcloud.to_file("wordCloudMo.png")

The author notes that installing wordcloud may produce various errors; a linked CSDN article provides troubleshooting steps.

The final word cloud highlights frequent terms such as "真实" (real), "孩子" (child), "演技" (acting), indicating positive audience sentiment toward the drama.