二、亮相工具库 

2.1. 工具库TextBlob介绍：

图像。TextBlob: Simplified Text Processing — TextBlob 0.16.0 documentation

TextBlob 是一个 python 库，可用于多个自然语言处理 （NLP） 任务，例如：

• 名词短语提取
• 词性标记
• 情绪分析
• 分类
• 标记化
• 单词和短语频率
• 解析
• n 元语法
• 词形变化（复数和单数化）和词形还原
• 拼写更正

#VADER 图像

2.2. 工具库VADER介绍：

VADER（Valence Aware Dictionary and sEntiment Reasoner）是一种基于词典和规则的情感分析工具，（Lexicon 意味着 NLP 系统的组件，其中包含有关每个单词或单词字符串的语义或语法等信息。例如。“无击球手”，“前进跑”和“巴尔的摩排骨”等是棒球词典的一部分，“边界”，“死亡结束”，“杜斯拉”等是板球词典的一部分。这是根据社交媒体的表达进行训练的。

词典情绪分析输出从 -1 到 1 的极性分数，其中 -1 表示真正的负面情绪，1 表示真正积极的情绪。接近 0 的值表示中性情绪。

三、TextBlob对比VADER:

        TextBlob和VADER之间的关键区别在于VADER专注于社交媒体。因此，VADER 投入了大量精力来识别通常出现在社交媒体上的内容的情绪，例如表情符号、重复单词和标点符号。

现在，我们将在情感分析中比较 VADER 和 TextBlob，因为 VADER 是仅用于情感分析的库。

3.1 步骤：

让我们首先安装这两个库。以下代码片段将帮助我们进行安装。

!pip install textblob
!pip install vaderSentiment

现在，让我们导入库。

#Importing libraries
from textblob import TextBlob
from vaderSentiment.vaderSentiment import SentimentIntensityAnalyzer

现在，让我们创建函数：

#Define funcitons for VADER and textblobdef vader_score(text):#After using VADER we will get  4 values: pos, compound, neu and neg.#pos:positive, neu:neutral, neg:negative#Here we are only collecting the compound. Why?#Because compound score is computed by summing the valence scores of each word in the lexicon, #adjusted according to the rules, and then normalized to be between -1 (most extreme negative) and +1 (most extreme positive).vader_sentiment = SentimentIntensityAnalyzer()score = vader_sentiment.polarity_scores(text) return score['compound']

def textblob_score(text):#textblob_sentiment.sentiment will give us 2 values: polarity and subjectivity#The polarity score is a float within the range [-1.0, 1.0]. #The subjectivity is a float within the range [0.0, 1.0] where 0.0 is very objective and 1.0 is very subjective.# Here we are interested in polarity, so we are using polarotytextblob_sentiment = TextBlob(text)score = textblob_sentiment.sentiment.polarityreturn score

        让我们看一下每种情绪（积极、消极和中性）的示例。

        让我们创建一个包含 3 个句子的列表，每个句子有一个肯定的、一个否定的和一个中性的句子。所以我们可以看到，VADER和TextBlob在它们上的表现。

#List of sentences with different emotions.
#First sentence is neutral.
#second is positive and 
#third is negative.
text_list = ["This is my first ever post on the internet.", "I am very excited to write this post.", "It's not good to work late hours."]

        现在让我们创建一个 for 循环，它将遍历每个句子，并为我们提供之前创建的 “vader_score” 和 “textblob_score” 函数的输出。

#Here it will iterate through every sentence from the text_list list
#and will output the sentence first and on the next line it will
#print the vader score and in the next line it will print the
#textblob score.
for text in text_list:print(f'sentence: {text} \n VADER sentiment score: {vader_score(text)} \n TextBlob score: {textblob_score(text)}')print("=" * 30)

我们将得到这样的输出。

sentence: This is my first ever post on the internet. VADER sentiment score: 0.0 TextBlob score: 0.25
==============================
sentence: I am very excited to write this post. VADER sentiment score: 0.4005 TextBlob score: 0.48750000000000004
==============================
sentence: It's not good to work late hours. VADER sentiment score: -0.3412 TextBlob score: -0.32499999999999996
==============================

        从上面的单元格中，我们可以得出结论，VADER完美地将第一句话识别为中性句子，而TextBlob离它并不远。然后对于第二句话，VADER给出了正分数，但TextBlob给了我们一个更积极的分数。对于最后一句话，VADER给出的负分比TextBlob更高。

        现在，我们可以说VADER和TextBlob都给出了相似的分数。

        现在，由于它看起来很棒，我们需要对句子进行一些更改以进一步深化我们的研究。例如。我们可以添加表情符号，标点符号，大写，重复等。然后我们将看到哪个表现更好。

        检查标点符号的影响：

#Adding punctuations
text_list = ["This is my first ever post on the internet!", "I am very excited to write this post!", "It's not good to work late hours!"]

sentence: This is my first ever post on the internet! VADER sentiment score: 0.0 TextBlob score: 0.3125
==============================
sentence: I am very excited to write this post! VADER sentiment score: 0.4561 TextBlob score: 0.609375
==============================
sentence: It's not good to work late hours! VADER sentiment score: -0.4015 TextBlob score: -0.3625
==============================

        现在，从上面的单元格中，我们可以说，感叹号确实提高了我们在所有句子中的分数。但是对于我们的中性句子（句子 1），TextBlob 走得更远了。

        检查大写的影响：

#Capitalizing words
text_list = ["This is my FIRST EVER post on the internet!", "I am very EXCITED to write this post!", "It's NOT GOOD to work late hours!"]

sentence: This is my FIRST EVER post on the internet! VADER sentiment score: 0.0 TextBlob score: 0.3125
==============================
sentence: I am very EXCITED to write this post! VADER sentiment score: 0.5744 TextBlob score: 0.609375
==============================
sentence: It's NOT GOOD to work late hours! VADER sentiment score: -0.5007 TextBlob score: -0.3625
==============================

        现在，我们可以说，我们的VADER分数提高了，但TextBlob分数保持不变。为什么会这样？好吧，原因是VADER认为大写版本具有更强的情绪并增加了情绪得分。同时，TextBlob 没有区分单词的大写和小写版本之间的情绪。

        检查重复单词的影响：

text_list = ["This is my VERY VERY FIRST EVER post on the internet!", "I am very very EXCITED to write this post!", "It's NOT NOT NOT GOOD to work late hours!"]

sentence: This is my VERY VERY FIRST EVER post on the internet! VADER sentiment score: 0.0 TextBlob score: 0.40625
==============================
sentence: I am very very EXCITED to write this post! VADER sentiment score: 0.6119 TextBlob score: 0.609375
==============================
sentence: It's NOT NOT NOT GOOD to work late hours! VADER sentiment score: -0.3311 TextBlob score: -0.3625
==============================

        通过重复的单词，我们可以看到VADER分数的明显变化，但这不会影响TextBlob分数。这是为什么呢？简单地说，这是因为VADER认为重复的单词具有更强的情感，而TextBlob没有解释重复的单词。

检查表情符号的影响：

text_list = ["This is my VERY VERY FIRST EVER post on the internet🫡🤔!","I am very very EXCITED to write this post😍😇!", "It's NOT NOT NOT GOOD to work late hours☹️😳!"]

sentence: This is my VERY VERY FIRST EVER post on the internet🫡🤔! VADER sentiment score: 0.0 TextBlob score: 0.40625
==============================
sentence: I am very very EXCITED to write this post😍😇! VADER sentiment score: 0.8749 TextBlob score: 0.609375
==============================
sentence: It's NOT NOT NOT GOOD to work late hours☹️😳! VADER sentiment score: -0.5802 TextBlob score: -0.3625
==============================

我们可以清楚地看到，维德分数正在提高。但是 TextBlob 分数根本没有变化。

四、结论：

根据我们所做的实验，这是否意味着VADER库比TextBlob更好？答案是否定的。但是为什么？因为：

TextBlob 可以做的不仅仅是情绪分类。

VADER-情感分析主要针对基于社交媒体数据的情感分析进行训练。TextBlob 的情况并非如此。

因此，我们不能得出VADER更好或TextBlob更好的结论。每个都有自己的用例。您可以根据问题陈述选择要使用的一个。

哈沙德·帕蒂尔