Spaces:
Runtime error
Runtime error
GameReview
commited on
Commit
•
87a4736
1
Parent(s):
cc02372
Upload initial attempt at launching app
Browse files- app.py +163 -0
- requirements.txt +3 -0
app.py
ADDED
@@ -0,0 +1,163 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
from googleapiclient.discovery import build
|
3 |
+
import pandas as pd
|
4 |
+
import numpy as np
|
5 |
+
from urllib.parse import urlparse, parse_qs
|
6 |
+
from transformers import pipeline
|
7 |
+
from transformers import AutoTokenizer
|
8 |
+
from transformers import AutoModelForSequenceClassification
|
9 |
+
from scipy.special import softmax
|
10 |
+
import gradio as gr
|
11 |
+
|
12 |
+
api_key = os.environ['api_key']
|
13 |
+
youtube_api = build('youtube','v3',developerKey=api_key)
|
14 |
+
|
15 |
+
#Get top 100 comments and make a dataframe
|
16 |
+
def get_comment_data(youtube_id):
|
17 |
+
request = youtube_api.commentThreads().list(part="snippet", videoId= youtube_id, maxResults=100, order="relevance", textFormat="plainText")
|
18 |
+
response = request.execute()
|
19 |
+
comments = [[comment['snippet']['topLevelComment']['snippet']['textDisplay'], comment['snippet']['topLevelComment']['snippet']['likeCount']] for comment in response['items']]
|
20 |
+
df = pd.DataFrame(comments, columns=['Comment_Text', 'Like_Count'])
|
21 |
+
return df
|
22 |
+
|
23 |
+
#In case we ever want all comments
|
24 |
+
def get_all_comments(youtube_id):
|
25 |
+
comments = [[]]
|
26 |
+
next_page_token = None
|
27 |
+
while True:
|
28 |
+
request = youtube_api.commentThreads().list(part="snippet", videoId= youtube_id, maxResults=100, pageToken=next_page_token, order="relevance", textFormat="plainText")
|
29 |
+
response = request.execute()
|
30 |
+
|
31 |
+
for item in response['items']:
|
32 |
+
comments.append([item['snippet']['topLevelComment']['snippet']['textDisplay'], item['snippet']['topLevelComment']['snippet']['likeCount']])
|
33 |
+
|
34 |
+
if 'nextPageToken' in response:
|
35 |
+
next_page_token = response['nextPageToken']
|
36 |
+
else:
|
37 |
+
break
|
38 |
+
df = pd.DataFrame(comments, columns=['Comment_Text', 'Like_Count'])
|
39 |
+
return df
|
40 |
+
|
41 |
+
#Get all videos from a creator
|
42 |
+
def get_channel_videos(channel_id):
|
43 |
+
all_videos=[]
|
44 |
+
# Initial request to retrieve the channel's uploaded videos
|
45 |
+
request = youtube_api.search().list(
|
46 |
+
part='id',
|
47 |
+
channelId=channel_id,
|
48 |
+
maxResults=50 # Adjust as needed
|
49 |
+
)
|
50 |
+
|
51 |
+
while request is not None:
|
52 |
+
response = request.execute()
|
53 |
+
|
54 |
+
for item in response.get('items', []):
|
55 |
+
if item['id']['kind'] == 'youtube#video':
|
56 |
+
all_videos.append(item['id']['videoId'])
|
57 |
+
|
58 |
+
request = youtube_api.search().list_next(request, response)
|
59 |
+
|
60 |
+
return all_videos
|
61 |
+
|
62 |
+
#Pass a valid youtube video url or else function will not work
|
63 |
+
def get_video_id(url):
|
64 |
+
parsed_url = urlparse(url)
|
65 |
+
return parse_qs(parsed_url.query)['v'][0]
|
66 |
+
|
67 |
+
#Set up the model and tokenizer
|
68 |
+
MODEL = f"cardiffnlp/twitter-roberta-base-sentiment"
|
69 |
+
MODEL2 = "SamLowe/roberta-base-go_emotions"
|
70 |
+
tokenizer = AutoTokenizer.from_pretrained(MODEL)
|
71 |
+
tokenizer2 = AutoTokenizer.from_pretrained(MODEL2)
|
72 |
+
model = AutoModelForSequenceClassification.from_pretrained(MODEL)
|
73 |
+
model2 = AutoModelForSequenceClassification.from_pretrained(MODEL2)
|
74 |
+
classifier = pipeline(task="text-classification", model="SamLowe/roberta-base-go_emotions", top_k=None)
|
75 |
+
|
76 |
+
def generate_sentiments(df, progress=gr.Progress()):
|
77 |
+
#Set up lists to add to dataframe
|
78 |
+
pos_sent = []
|
79 |
+
neu_sent = []
|
80 |
+
neg_sent = []
|
81 |
+
|
82 |
+
feeling1 = []
|
83 |
+
feeling2 = []
|
84 |
+
feeling3 = []
|
85 |
+
|
86 |
+
for comment in progress.tqdm(df['Comment_Text'],desc="Analyzing Comments"):
|
87 |
+
#Encode the comment and run roberta on it
|
88 |
+
tokens = tokenizer.tokenize(comment)
|
89 |
+
if len(tokens) > 514:
|
90 |
+
tokens = tokens[:512]
|
91 |
+
comment = tokenizer.convert_tokens_to_string(tokens)
|
92 |
+
|
93 |
+
model_outputs = classifier(comment)
|
94 |
+
top_three_feelings = ""
|
95 |
+
|
96 |
+
#Top three sentiments, RoBERTa-based model
|
97 |
+
sentiment1 = list(model_outputs[0][0].values())[0]
|
98 |
+
sentiment2 = list(model_outputs[0][1].values())[0]
|
99 |
+
sentiment3 = list(model_outputs[0][2].values())[0]
|
100 |
+
|
101 |
+
feeling1.append(sentiment1)
|
102 |
+
feeling2.append(sentiment2)
|
103 |
+
feeling3.append(sentiment3)
|
104 |
+
|
105 |
+
encoded_comment = tokenizer(comment, return_tensors='pt')
|
106 |
+
output = model(**encoded_comment)
|
107 |
+
result = output[0][0].detach().numpy()
|
108 |
+
#Convert the numbers to be between 0 and 1 to do analysis with it
|
109 |
+
result = softmax(result)
|
110 |
+
#Add results to the lists
|
111 |
+
pos_sent.append(result[2])
|
112 |
+
neu_sent.append(result[1])
|
113 |
+
neg_sent.append(result[0])
|
114 |
+
#Add sentiments to the dataframe
|
115 |
+
new_df = df.copy()
|
116 |
+
new_df['Positive_Sentiment'] = pos_sent
|
117 |
+
new_df['Neural_Sentiment'] = neu_sent
|
118 |
+
new_df['Negative_Sentiment'] = neg_sent
|
119 |
+
|
120 |
+
new_df['Feeling 1'] = feeling1
|
121 |
+
new_df['Feeling 2'] = feeling2
|
122 |
+
new_df['Feeling 3'] = feeling3
|
123 |
+
|
124 |
+
return new_df
|
125 |
+
|
126 |
+
def addWeights(df,progress=gr.Progress()):
|
127 |
+
df1 = generate_sentiments(df,progress)
|
128 |
+
total_weights = df1['Like_Count'].sum()
|
129 |
+
df1['Weights'] = df1['Like_Count'] / total_weights
|
130 |
+
return df1
|
131 |
+
|
132 |
+
def getWeightSentimentAll(df, progress=gr.Progress()):
|
133 |
+
df1 = addWeights(df,progress)
|
134 |
+
#Start at default 0.5, add the results of positive sentiment and subtract negative sentiment
|
135 |
+
weighted_avg = (df1['Positive_Sentiment'] * df1['Weights']).sum()*0.5 - (df1['Negative_Sentiment'] * df1['Weights']).sum()*0.5 + 0.5
|
136 |
+
df['Weighted Average'] = weighted_avg
|
137 |
+
return weighted_avg
|
138 |
+
|
139 |
+
def rate(youtube_url, progress=gr.Progress()):
|
140 |
+
try:
|
141 |
+
vid_id = get_video_id(youtube_url)
|
142 |
+
vid_df = get_comment_data(vid_id)
|
143 |
+
#This step to be replaced with whatever final calculation we decide
|
144 |
+
vid_sent = getWeightSentimentAll(vid_df,progress)
|
145 |
+
return vid_sent
|
146 |
+
except:
|
147 |
+
raise gr.Error("Process failed. Ensure link is a valid YouTube URL")
|
148 |
+
|
149 |
+
|
150 |
+
with gr.Blocks() as app:
|
151 |
+
gr.Markdown("""
|
152 |
+
# Game Review Analysis Using Youtube
|
153 |
+
|
154 |
+
### Insert a YouTube URL to analyze the comments and get the population's review on the game!
|
155 |
+
"""
|
156 |
+
)
|
157 |
+
|
158 |
+
input = gr.Textbox(label="YouTube URL", placeholder = "Place link here")
|
159 |
+
output = gr.Textbox(label = "Community's Rating of the Game")
|
160 |
+
rate_btn = gr.Button("Rate!")
|
161 |
+
rate_btn.click(fn=rate, inputs=input,outputs=output)
|
162 |
+
|
163 |
+
app.launch()
|
requirements.txt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
transformers
|
2 |
+
tensorflow
|
3 |
+
google-api-python-client
|