Defining which reviews have positive or negative sentiment

Loading...

来自 University of Washington 的课程

机器学习基础：案例研究

7675 个评分

University of Washington

机器学习基础：案例研究

7675 个评分

课程 1（共 4 门，Specialization Machine Learning）

你是否好奇数据可以告诉你什么？你是否想在关于机器学习促进商业的核心方式上有深层次的理解？你是否想能同专家们讨论关于回归，分类，深度学习以及推荐系统的一切？在这门课上，你将会通过一系列实际案例学习来获取实践经历。在这门课结束的时候，

从本节课中

Classification: Analyzing Sentiment

How do you guess whether a person felt positively or negatively about an experience, just from a short review they wrote?<p>In our second case study, analyzing sentiment, you will create models that predict a class (positive/negative sentiment) from input features (text of the reviews, user profile information,...).This task is an example of classification, one of the most widely used areas of machine learning, with a broad array of applications, including ad targeting, spam detection, medical diagnosis and image classification.</p>You will analyze the accuracy of your classifier, implement an actual classifier in an iPython notebook, and take a first stab at a core piece of the intelligent application you will build and deploy in your capstone.

Loading & exploring product review data2:53

Creating the word count vector2:05

Exploring the most popular product4:36

Defining which reviews have positive or negative sentiment4:32

Training a sentiment classifier3:17

Evaluating a classifier & the ROC curve4:24

Applying model to find most positive & negative reviews for a product4:43

Exploring the most positive & negative aspects of a product4:41

与讲师见面

Carlos Guestrin
Amazon Professor of Machine Learning
Computer Science and Engineering
Emily Fox
Amazon Professor of Machine Learning
Statistics

[MUSIC]

Build a sentiment classifier.

And I'm gonna put a # here to make it a nice title.

So this is our next task here.

And so, when you build a sentiment classifier,

you're talking about positives and negatives.

So, thumbs down, and thumbs up.

But if you remember, our product ratings, we're not about positive and

negative, they're numerical things.

So, for example, if I take all the products, and

I'll take the rating column and I do a .show, and we did above just for

the giraffe, but do a .show for everything with the view equal

to Categorical, we're now getting a histogram for

all of the views, and if you take a quick look at it, you'll see that most reviews

are positive across the board 107,000 reviews are five stars.

So most people review positively, and just write reviews about products they like.

They don't typically write reviews about products they don't like.

Then the next set of reviews, 33,000 four stars.

Then 3 stars and again, a lot of people write really bad reviews 1 star, 2 star,

why would you give a review product 2 stars?

You might as well just give them 1 star if you really hated it.

And this is what we observe in the histogram.

But again, for sentiment analysis, we have to define what's thumbs up and

what's thumbs down.

And so I'm gonna make an arbitrary choice here.

Let's say that things that 4, 5 stars are things that people liked.

So those are positives.

Things that 1 and 2 stars are negative.

But the things that are 3 stars, those are kind of in the middle.

So, let's just throw those out.

So we're gonna do a little bit of what we'll call data engineering,

just defining what is a positive and negative sentiment.

So let's do that right now.

So in the subsection we're gonna define

what's a positive and a negative sentiment.

And what I'm gonna

do first is ignore all

city star reviews.

The way to do it is by saying,

okay I'm gonna take the product stable the variable products.

And I'm gonna just select everything out

of the products table whose rating was not 3.

So, products[products['rating'] |= 3].

So that was our first step in our little data engineering task.

And now, the next step in this task

is to actually find what stems up and stems down.

So I'm gonna say a positive

sentiment equals 4 star or

5 star reviews.

So let's go ahead and

add a new column to our table that defines the actual sentiment.

So products new column called sentiment.

It's gonna be a binary column, zero one.

And, the way we're gonna define the column, is we're gonna say,

of the products rating, is it greater equal to 4?

If it's greater or equal to 4, it's gonna get a 1, and

if it's less than 4, it's gonna get a 0, and

so now if I take another look at my product stable, so head

You should see that I've added a new column on the right called Sentiment.

And most of the sentiments are positive, as we saw earlier, but

many are negative, zero.

Okay, now we're ready, we're finally

ready to train our sentiment classifier.

[MUSIC]

探索我们的目录

免费加入并获得个性化推荐、更新和优惠。

Coursera 致力于普及全世界最好的教育，它与全球一流大学和机构合作提供在线课程。

© 2018 Coursera Inc. 保留所有权利。

通过 App Store 下载

通过 Google Play 获取