Let's look at the shape of our dataset:

hotel_reviews.shape

(35912, 19)

This is showing us that we are working with 35,912 rows and 19 columns. Eventually, we will be concerned only with the column that contains the text data, but for now, let's see what the first few rows look like to get a better sense of what is included in our data:

hotel_reviews.head()

This gives us the following table:

Let's only include reviews from the United States in order to try and include only English reviews. First, let's plot our data, like so:

# plot the lats and longs of reviews
hotel_reviews.plot.scatter(x='longitude', y='latitude')

The output looks something like this:

For the purpose of making our dataset a bit easier to work with, let's use pandas to subset the reviews and only include those that came from the United States:

# Filter to only include reviews within the US
hotel_reviews = hotel_reviews[((hotel_reviews['latitude']<=50.0) & (hotel_reviews['latitude']>=24.0)) & ((hotel_reviews['longitude']<=-65.0) & (hotel_reviews['longitude']>=-122.0))]

# Plot the lats and longs again
hotel_reviews.plot.scatter(x='longitude', y='latitude')
# Only looking at reviews that are coming from the US

The output is as follows:

It looks like a map of the U.S.! Let's shape our filtered dataset now:

hotel_reviews.shape

We have 30,692 rows and 19 columns. When we write reviews for hotels, we usually write about different things in the same review. For this reason, we will attempt to assign topics to single sentences rather than to the entire review.

To do so, let's grab the text column from our data, like so:

texts = hotel_reviews['reviews.text']