Solutions for Assignment 5: Exploring Yelp Reviews in Philadelphia¶

Luming Xu, 2024-12-04¶

In this assignment, we'll explore restaurant review data available through the Yelp Dataset Challenge. The dataset includes Yelp data for user reviews and business information for many metropolitan areas. I've already downloaded this dataset (8 GB total!) and extracted out the data files for reviews and restaurants in Philadelphia. I've placed these data files into the data directory in this repository.

This assignment is broken into two parts:

Part 1: Analyzing correlations between restaurant reviews and census data

We'll explore the relationship between restaurant reviews and the income levels of the restaurant's surrounding area.

Part 2: Exploring the impact of fast food restaurants

We'll run a sentiment analysis on reviews of fast food restaurants and estimate income levels in neighborhoods with fast food restaurants. We'll test how well our sentiment analysis works by comparing the number of stars to the sentiment of reviews.

Background readings

  • Does sentiment analysis work?
  • The Geography of Taste: Using Yelp to Study Urban Culture

1. Correlating restaurant ratings and income levels¶

In this part, we'll use the census API to download household income data and explore how it correlates with restaurant review data.

1.1 Query the Census API¶

Use the cenpy package to download median household income in the past 12 months by census tract from the 2021 ACS 5-year data set for your county of interest.

You have two options to find the correct variable names:

  • Search through: https://api.census.gov/data/2021/acs/acs5/variables.html
  • Initialize an API connection and use the .varslike() function to search for the proper keywords

At the end of this step, you should have a pandas DataFrame holding the income data for all census tracts within the county being analyzed. Feel free to rename your variable from the ACS so it has a more meaningful name!

::: {.callout-caution} Some census tracts won't have any value because there are not enough households in that tract. The census will use a negative number as a default value for those tracts. You can safely remove those tracts from the analysis! :::

In [1]:
import pandas as pd
import cenpy
acs = cenpy.remote.APIConnection("ACSDT5Y2021")
# Search for rows where name contains "Philadelphia"
# counties.loc[ counties[3].str.contains("Philadelphia") ]
variables = ["NAME", "B19013_001E"]
philly_county_code = "101"
pa_state_code = "42"
philly_medhhinc = acs.query(
    cols=variables,
    geo_unit="tract:*",
    geo_filter={"state": pa_state_code, "county": philly_county_code},
)
for variable in variables:
    # Convert all variables EXCEPT for NAME
    if variable != "NAME":
        philly_medhhinc[variable] = pd.to_numeric(philly_medhhinc[variable], errors='coerce')
philly_medhhinc.rename(columns={"B19013_001E": "medhhinc"}, inplace = True)
philly_medhhinc.loc[philly_medhhinc["medhhinc"]<0,"medhhinc"] = 0

philly_medhhinc.head()
Out[1]:
NAME medhhinc state county tract
0 Census Tract 1.01, Philadelphia County, Pennsy... 104052 42 101 000101
1 Census Tract 1.02, Philadelphia County, Pennsy... 91944 42 101 000102
2 Census Tract 2, Philadelphia County, Pennsylvania 91067 42 101 000200
3 Census Tract 3, Philadelphia County, Pennsylvania 86782 42 101 000300
4 Census Tract 4.01, Philadelphia County, Pennsy... 67188 42 101 000401

1.2 Download census tracts from the Census and merge the data from part 1.1¶

  • Download census tracts for the desired geography using the pygris package
  • Merge the downloaded census tracts with the household income DataFrame
In [2]:
#!pip install pygris
import pygris
philly_tracts = pygris.tracts(
    state=pa_state_code, county=philly_county_code, year=2021
)
medhhinc_analysis = philly_medhhinc.merge(
    philly_tracts,
    right_on=["STATEFP", "COUNTYFP", "TRACTCE"],
    left_on=["state", "county", "tract"],
)
medhhinc_analysis = medhhinc_analysis[["NAME_x","medhhinc", "state", "county", "tract", "GEOID", "geometry"]]
medhhinc_analysis.head()
Out[2]:
NAME_x medhhinc state county tract GEOID geometry
0 Census Tract 1.01, Philadelphia County, Pennsy... 104052 42 101 000101 42101000101 POLYGON ((-75.15221 39.94997, -75.15203 39.950...
1 Census Tract 1.02, Philadelphia County, Pennsy... 91944 42 101 000102 42101000102 POLYGON ((-75.15155 39.95544, -75.15148 39.955...
2 Census Tract 2, Philadelphia County, Pennsylvania 91067 42 101 000200 42101000200 POLYGON ((-75.16289 39.95533, -75.16269 39.956...
3 Census Tract 3, Philadelphia County, Pennsylvania 86782 42 101 000300 42101000300 POLYGON ((-75.17994 39.96003, -75.17942 39.959...
4 Census Tract 4.01, Philadelphia County, Pennsy... 67188 42 101 000401 42101000401 POLYGON ((-75.18091 39.95298, -75.18022 39.954...

1.3 Load the restaurants data¶

The Yelp dataset includes data for 7,350 restaurants across the city. Load the data from the data/ folder and use the latitude and longitude columns to create a GeoDataFrame after loading the JSON data. Be sure to set the right CRS on when initializing the GeoDataFrame!

Notes

The JSON data is in a "records" format. To load it, you'll need to pass the following keywords:

  • orient='records'
  • lines=True
In [3]:
import geopandas as gpd
restaurants = pd.read_json("data/restaurants_philly.json.gz", orient='records', lines=True)
restaurants_gdf = gpd.GeoDataFrame(
    restaurants, geometry = gpd.points_from_xy(restaurants["longitude"], restaurants["latitude"]),
    crs = 'EPSG:4326'
)
restaurants_gdf.head()
Out[3]:
business_id latitude longitude name review_count stars categories geometry
0 MTSW4McQd7CbVtyjqoe9mw 39.955505 -75.155564 St Honore Pastries 80 4.0 Restaurants, Food, Bubble Tea, Coffee & Tea, B... POINT (-75.15556 39.95551)
1 MUTTqe8uqyMdBl186RmNeA 39.953949 -75.143226 Tuna Bar 245 4.0 Sushi Bars, Restaurants, Japanese POINT (-75.14323 39.95395)
2 ROeacJQwBeh05Rqg7F6TCg 39.943223 -75.162568 BAP 205 4.5 Korean, Restaurants POINT (-75.16257 39.94322)
3 QdN72BWoyFypdGJhhI5r7g 39.939825 -75.157447 Bar One 65 4.0 Cocktail Bars, Bars, Italian, Nightlife, Resta... POINT (-75.15745 39.93982)
4 Mjboz24M9NlBeiOJKLEd_Q 40.022466 -75.218314 DeSandro on Main 41 3.0 Pizza, Restaurants, Salad, Soup POINT (-75.21831 40.02247)

1.4 Add tract info for each restaurant¶

Do a spatial join to identify which census tract each restaurant is within. Make sure each dataframe has the same CRS!

At the end of this step, you should have a new dataframe with a column identifying the tract number for each restaurant.

In [4]:
restaurants_tract = gpd.sjoin(restaurants_gdf, philly_tracts[["geometry","TRACTCE"]], predicate='within', how='left').drop(columns=["index_right"])
restaurants_tract.head()

C:\Users\19397\AppData\Local\Temp\ipykernel_5904\3302763282.py:1: UserWarning: CRS mismatch between the CRS of left geometries and the CRS of right geometries.
Use `to_crs()` to reproject one of the input geometries to match the CRS of the other.

Left CRS: EPSG:4326
Right CRS: EPSG:4269

  restaurants_tract = gpd.sjoin(restaurants_gdf, philly_tracts[["geometry","TRACTCE"]], predicate='within', how='left').drop(columns=["index_right"])
Out[4]:
business_id latitude longitude name review_count stars categories geometry TRACTCE
0 MTSW4McQd7CbVtyjqoe9mw 39.955505 -75.155564 St Honore Pastries 80 4.0 Restaurants, Food, Bubble Tea, Coffee & Tea, B... POINT (-75.15556 39.95551) 000200
1 MUTTqe8uqyMdBl186RmNeA 39.953949 -75.143226 Tuna Bar 245 4.0 Sushi Bars, Restaurants, Japanese POINT (-75.14323 39.95395) 000102
2 ROeacJQwBeh05Rqg7F6TCg 39.943223 -75.162568 BAP 205 4.5 Korean, Restaurants POINT (-75.16257 39.94322) 001500
3 QdN72BWoyFypdGJhhI5r7g 39.939825 -75.157447 Bar One 65 4.0 Cocktail Bars, Bars, Italian, Nightlife, Resta... POINT (-75.15745 39.93982) 001800
4 Mjboz24M9NlBeiOJKLEd_Q 40.022466 -75.218314 DeSandro on Main 41 3.0 Pizza, Restaurants, Salad, Soup POINT (-75.21831 40.02247) 021000

1.5 Add income data to your restaurant data¶

Add the income data to your dataframe from the previous step, merging the census data based on the tract that each restaurant is within.

In [5]:
restaurants_income = pd.merge(
    restaurants_tract,
    medhhinc_analysis[["medhhinc","tract"]],
    left_on = "TRACTCE",
    right_on = "tract"
)
restaurants_income.head()
Out[5]:
business_id latitude longitude name review_count stars categories geometry TRACTCE medhhinc tract
0 MTSW4McQd7CbVtyjqoe9mw 39.955505 -75.155564 St Honore Pastries 80 4.0 Restaurants, Food, Bubble Tea, Coffee & Tea, B... POINT (-75.15556 39.95551) 000200 91067 000200
1 L_sXNadtVHjxMw7Yhvkj9Q 39.955454 -75.154900 Naoki Ramen 92 4.0 Ramen, Restaurants, Japanese POINT (-75.15490 39.95545) 000200 91067 000200
2 icp_IKE9zIkAqAucyS1vTA 39.955495 -75.155256 Hakka Beef House 33 4.5 Restaurants, Chinese POINT (-75.15526 39.95549) 000200 91067 000200
3 8cLbpSZOHnmzxzDNSBGQCA 39.955417 -75.155495 About BBQ 14 4.0 Barbeque, Restaurants POINT (-75.15549 39.95542) 000200 91067 000200
4 mCo2uVTTGYrEhRrkQW-CMw 39.953940 -75.156309 Empress Garden 263 4.0 Taiwanese, Chinese, Noodles, Restaurants POINT (-75.15631 39.95394) 000200 91067 000200

1.6 Make a plot of median household income vs. Yelp stars¶

Our dataset has the number of stars for each restaurant, rounded to the nearest 0.5 star. In this step, create a line plot that shows the average income value for each stars category (e.g., all restaurants with 1 star, 1.5 stars, 2 stars, etc.)

While their are multiple ways to do this, the seaborn.lineplot() is a great option. This can show the average value in each category as well as 95% uncertainty intervals. Use this function to plot the stars ("x") vs. average income ("y") for all of our restaurants, using the dataframe from last step. Be sure to format your figure to make it look nice!

Question: Is there a correlation between a restaurant's ratings and the income levels of its surrounding neighborhood?

In [6]:
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
avg_inc_stars = restaurants_income.groupby("stars")["medhhinc"].mean().reset_index()
avg_inc_stars["stars"] = pd.to_numeric(avg_inc_stars["stars"], errors="coerce")
avg_inc_stars["medhhinc"] = pd.to_numeric(avg_inc_stars["medhhinc"], errors="coerce")

fig,ax = plt.subplots(figsize = (10,6))
sns.lineplot(
    data=avg_inc_stars, 
    x="stars", 
    y="medhhinc", 
    marker="o", 
    ci=95, 
    color='#e78ac3'
)

# Format the plot
plt.title("Average Income by Restaurant Star Rating", fontsize=16, fontweight='bold')
plt.suptitle("Philadelphia, 2021", fontsize = 14)
plt.xlabel("Star Rating", fontsize=14)
plt.ylabel("Average Income", fontsize=14)
plt.xticks(fontsize=12)
plt.yticks(fontsize=12)
plt.grid(alpha=0.3)
plt.tight_layout()

# Show the plot
plt.show()

C:\Users\19397\AppData\Local\Temp\ipykernel_5904\66545563.py:9: FutureWarning: 

The `ci` parameter is deprecated. Use `errorbar=('ci', 95)` for the same effect.

  sns.lineplot(
No description has been provided for this image

Annotation: The plot reveals a positive correlation between restaurant ratings and household income for star ratings below 4.0. However, as the star rating increases from 4.0 to 5.0, household income decreases, indicating a negative relationship.

2. Fast food trends in Philadelphia¶

At the end of part 1, you should have seen a strong trend where higher income tracts generally had restaurants with better reviews. In this section, we'll explore the impact of fast food restaurants and how they might be impacting this trend.

Hypothesis

  1. Fast food restaurants are predominantly located in areas with lower median income levels.
  2. Fast food restaurants have worse reviews compared to typical restaurants.

If true, these two hypotheses could help to explain the trend we found in part 1. Let's dive in and test our hypotheses!

2.1 Identify fast food restaurants¶

The "categories" column in our dataset contains multiple classifications for each restaurant. One such category is "Fast Food". In this step, add a new column called "is_fast_food" that is True if the "categories" column contains the term "Fast Food" and False otherwise

In [7]:
restaurants_income['is_fast_food']= restaurants_income["categories"].str.contains("Fast Food", case = False, na = False)
restaurants_income.head()
Out[7]:
business_id latitude longitude name review_count stars categories geometry TRACTCE medhhinc tract is_fast_food
0 MTSW4McQd7CbVtyjqoe9mw 39.955505 -75.155564 St Honore Pastries 80 4.0 Restaurants, Food, Bubble Tea, Coffee & Tea, B... POINT (-75.15556 39.95551) 000200 91067 000200 False
1 L_sXNadtVHjxMw7Yhvkj9Q 39.955454 -75.154900 Naoki Ramen 92 4.0 Ramen, Restaurants, Japanese POINT (-75.15490 39.95545) 000200 91067 000200 False
2 icp_IKE9zIkAqAucyS1vTA 39.955495 -75.155256 Hakka Beef House 33 4.5 Restaurants, Chinese POINT (-75.15526 39.95549) 000200 91067 000200 False
3 8cLbpSZOHnmzxzDNSBGQCA 39.955417 -75.155495 About BBQ 14 4.0 Barbeque, Restaurants POINT (-75.15549 39.95542) 000200 91067 000200 False
4 mCo2uVTTGYrEhRrkQW-CMw 39.953940 -75.156309 Empress Garden 263 4.0 Taiwanese, Chinese, Noodles, Restaurants POINT (-75.15631 39.95394) 000200 91067 000200 False

2.2 Calculate the median income for fast food and otherwise¶

Group by the "is_fast_food" column and calculate the median income for restaurants that are and are not fast food. You should find that income levels are lower in tracts with fast food.

Note: this is just an estimate, since we are calculating a median of median income values.

In [8]:
restaurants_income.groupby("is_fast_food")["medhhinc"].mean()
Out[8]:
is_fast_food
False    75153.311165
True     58079.269755
Name: medhhinc, dtype: float64

2.3 Load fast food review data¶

In the rest of part 2, we're going to run a sentiment analysis on the reviews for fast food restaurants. The review data for all fast food restaurants identified in part 2.1 is already stored in the data/ folder. The data is stored as a JSON file and you can use pandas.read_json to load it.

Notes

The JSON data is in a "records" format. To load it, you'll need to pass the following keywords:

  • orient='records'
  • lines=True
In [9]:
reviews = pd.read_json("data/reviews_philly_fast_food.json.gz", orient='records', lines=True)
reviews.head()
Out[9]:
business_id review_id stars text
0 kgMEBZG6rjkGeFzPaIM4MQ E-yGr1OhsUBxNeUVLDVouA 1 I know I shouldn't expect much but everything ...
1 FKrP06TDAKtxNG1vrRQcQQ 0IpFZoaY_RRNjha8Q_Wz6w 2 Perfect place to go if you like waiting 20 mi...
2 w9hS5x1F52Id-G1KTrAOZg 0KlwfaHZyvao41_3S47dyg 2 Was not a fan of their cheesesteak. Their wiz ...
3 fr2qDm_mY1afIGMvqsKUCg oKSUOq7pCQzyypFDSa1HoA 3 Ok this is an aberration from my city foodie r...
4 fr2qDm_mY1afIGMvqsKUCg 6SMUmb7Npwnq6AusxqOXzQ 5 My family has been customers of George's for y...

2.4 Trim to the most popular fast food restaurants¶

There's too many reviews to run a sentiment analysis on all of them in a reasonable time. Let's trim our reviews dataset to the most popular fast food restaurants, using the list provided below.

You will need to get the "business_id" values for each of these restaurants from the restaurants data loaded in part 1.3. Then, trim the reviews data to include reviews only for those business IDs.

In [10]:
popular_fast_food = [
    "McDonald's",
    "Wendy's",
    "Subway",
    "Popeyes Louisiana Kitchen",
    "Taco Bell",
    "KFC",
    "Burger King",
]
In [11]:
popular_restaurants = restaurants_gdf[restaurants_gdf["name"].isin(popular_fast_food)]
popular_reviews = reviews[reviews["business_id"].isin(popular_restaurants["business_id"])]
popular_reviews.head()
Out[11]:
business_id review_id stars text
0 kgMEBZG6rjkGeFzPaIM4MQ E-yGr1OhsUBxNeUVLDVouA 1 I know I shouldn't expect much but everything ...
8 PjknD8uD_0tisZQbomiYoQ 6TqKBa-HDiq2_W_ip2AItA 5 I am only giving 5 stars because the Shamrock ...
13 kgMEBZG6rjkGeFzPaIM4MQ NGaXI03qbtBLshjfJV4pbQ 3 Dirty bathrooms and very slow service, but I w...
17 LACylKxImNI29DKUQpWuHw HHy9yIjW07VHUE6nXVbsVA 3 Burger King is an okay alternative to Mcdonald...
21 gq4zw-ru_rkZ2UBIanaZFQ yMZTK5B_0SAdUXSrIkXrmA 1 ive tried going here four times with no succes...

2.5 Run the emotions classifier on fast food reviews¶

Run a sentiment analysis on the reviews data from the previous step. Use the DistilBERT model that can predict emotion labels (anger, fear, sadness, joy, love, and surprise). Transform the result from the classifier into a DataFrame so that you have a column for each of the emotion labels.

In [12]:
from transformers import pipeline
sentiment_analysis = popular_reviews.dropna(subset=["text"])
sentiment_analysis_final = sentiment_analysis.loc[sentiment_analysis["text"] != ""]
# Strip out spaces and convert to a list
text = sentiment_analysis_final["text"].str.strip().tolist()

# The model
model = "bhadresh-savani/distilbert-base-uncased-emotion"
# Initialize our sentiment analyzer
emotion_classifier = pipeline(
    task="text-classification",  # The task we are doing
    model=model,  # The specific model name
    top_k=None,  # Predict all labels, not just top ones
    tokenizer=model,  # Tokenize inputs using model tokenizer
    truncation=True,  # Truncate text if we need to
)

# %%time 
emotion_scores = emotion_classifier(text)

emotion = pd.DataFrame(
    [{d["label"]: d["score"] for d in dd} for dd in emotion_scores]
).assign(text=text)
emotion.head()

c:\Users\19397\.conda\envs\musa-550-fall-2023\lib\site-packages\huggingface_hub\file_download.py:1150: FutureWarning: `resume_download` is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use `force_download=True`.
  warnings.warn(
Xformers is not installed correctly. If you want to use memory_efficient_attention to accelerate training use the following command to install Xformers
pip install xformers.
Out[12]:
sadness fear anger joy surprise love text
0 0.733869 0.250677 0.011039 0.002758 0.001015 0.000643 I know I shouldn't expect much but everything ...
1 0.000230 0.000126 0.000165 0.998759 0.000246 0.000475 I am only giving 5 stars because the Shamrock ...
2 0.000216 0.000088 0.000153 0.998563 0.000161 0.000819 Dirty bathrooms and very slow service, but I w...
3 0.000838 0.000403 0.000811 0.996928 0.000140 0.000880 Burger King is an okay alternative to Mcdonald...
4 0.005284 0.000753 0.006195 0.985421 0.001620 0.000726 ive tried going here four times with no succes...

2.6 Identify the predicted emotion for each text¶

Use the pandas idxmax() to identify the predicted emotion for each review, and add this value to a new column called "prediction"

The predicted emotion has the highest confidence score across all emotion labels for a particular label.

In [13]:
emotion_labels = ["anger", "fear", "sadness", "joy", "surprise", "love"]
emotion_pred = emotion
emotion_pred['prediction'] = emotion[emotion_labels].idxmax(axis=1)
emotion_pred.head()
Out[13]:
sadness fear anger joy surprise love text prediction
0 0.733869 0.250677 0.011039 0.002758 0.001015 0.000643 I know I shouldn't expect much but everything ... sadness
1 0.000230 0.000126 0.000165 0.998759 0.000246 0.000475 I am only giving 5 stars because the Shamrock ... joy
2 0.000216 0.000088 0.000153 0.998563 0.000161 0.000819 Dirty bathrooms and very slow service, but I w... joy
3 0.000838 0.000403 0.000811 0.996928 0.000140 0.000880 Burger King is an okay alternative to Mcdonald... joy
4 0.005284 0.000753 0.006195 0.985421 0.001620 0.000726 ive tried going here four times with no succes... joy

2.7 Combine the ratings and sentiment data¶

Combine the data from part 2.4 (reviews data) and part 2.6 (emotion data). Use the pd.concat() function and combine along the column axis.

Note: You'll need to reset the index of your reviews data frame so it matches the emotion data index (it should run from 0 to the length of the data - 1).

In [14]:
emotion_tidy = emotion_pred
emotion_tidy["score"] = emotion_pred.apply(
    lambda row: row[row["prediction"]], axis=1)
emotion_tidy = emotion_tidy[["text", "prediction", "score"]]

popular_reviews_reset = popular_reviews.reset_index(drop=True)
emotion_restaurants = pd.concat([popular_reviews_reset, emotion_tidy], axis = 1)
emotion_restaurants.rename(columns = {"prediction":"Emotion"}, inplace=True)
emotion_restaurants.head()
Out[14]:
business_id review_id stars text text Emotion score
0 kgMEBZG6rjkGeFzPaIM4MQ E-yGr1OhsUBxNeUVLDVouA 1 I know I shouldn't expect much but everything ... I know I shouldn't expect much but everything ... sadness 0.733869
1 PjknD8uD_0tisZQbomiYoQ 6TqKBa-HDiq2_W_ip2AItA 5 I am only giving 5 stars because the Shamrock ... I am only giving 5 stars because the Shamrock ... joy 0.998759
2 kgMEBZG6rjkGeFzPaIM4MQ NGaXI03qbtBLshjfJV4pbQ 3 Dirty bathrooms and very slow service, but I w... Dirty bathrooms and very slow service, but I w... joy 0.998563
3 LACylKxImNI29DKUQpWuHw HHy9yIjW07VHUE6nXVbsVA 3 Burger King is an okay alternative to Mcdonald... Burger King is an okay alternative to Mcdonald... joy 0.996928
4 gq4zw-ru_rkZ2UBIanaZFQ yMZTK5B_0SAdUXSrIkXrmA 1 ive tried going here four times with no succes... ive tried going here four times with no succes... joy 0.985421

2.8 Plot sentiment vs. stars¶

We now have a dataframe with the predicted primary emotion for each review and the associated number of stars for each review. Let's explore two questions:

  1. Does sentiment analysis work? Do reviews with fewer stars have negative emotions?
  2. For our fast food restaurants, are reviews generally positive or negative?

Use seaborn's histplot() to make a stacked bar chart showing the breakdown of each emotion for each stars category (1 star, 2 stars, etc.). A few notes:

  • To stack multiple emotion labels in one bar, use the multiple="stack" keyword
  • The discrete=True can be helpful to tell seaborn our stars values are discrete categories
In [15]:
# Define the custom palette
custom_palette = ['#fbb4ae','#b3cde3','#ccebc5','#decbe4','#fed9a6','#ffffcc']
ax,fig = plt.subplots(figsize = (10,6))
sns.histplot(
    emotion_restaurants,
    x = "stars",
    hue = "Emotion",
    multiple="stack",
    discrete=True,
    shrink=0.8,  # Adjust width of bars (optional)
    palette=custom_palette,
    edgecolor="white"
)

# Title and labels
plt.title("Emotion Breakdown by Star Rating", fontsize=16, fontweight='bold')
plt.suptitle("(Most popular fast food restaurants in Philadelphia)", fontsize=14)
plt.xlabel("Star Rating", fontsize=14)
plt.ylabel("Count of Reviews", fontsize=14)

# Display the plot
plt.tight_layout()
plt.show()
No description has been provided for this image

Question: What does your chart indicate for the effectiveness of our sentiment analysis? Does our original hypothesis about fast food restaurants seem plausible?

Annotation: The distribution of emotions closely aligns with star ratings. One- and two-star reviews are more associated with feelings of sadness and anger, while joy becomes more prevalent as the ratings increase. Among the most popular fast-food restaurants in Philadelphia, one-star reviews significantly outnumber those with higher ratings, reflecting a generally negative attitude among citizens.