def for_chinese(date):
"""[summary]
Arguments:
date {str} -- Must be prefixed with underscore, e.g. '20180303_'
"""
# 1. Load the post object generated by _2_remove_unrelated_data
all_posts = load_posts('analysis/_2_remove_unrelated_data/chinese.json')
# 2. Load negative and positive chinese keywords
positive_kp = load_semantic_keywords_processor(date, True, False,
'chinese')
negative_kp = load_semantic_keywords_processor(date, False, True,
'chinese')
# 3. Match those keywords against every post object
log("Labelling semantic of chinese post", 1)
for p in all_posts:
matching_positive_keywords = positive_kp.extract_keywords(p["value"])
matching_negative_keywords = negative_kp.extract_keywords(p["value"])
if (len(matching_positive_keywords) > 0):
p["semantic_value"]["positive"] = True
if (len(matching_negative_keywords) > 0):
p["semantic_value"]["negative"] = True
# 4. Save the labelled post as chinese.json
save_posts(all_posts, f'analysis/_3_label_semantic/chinese.json')
def save_posts(list_of_posts, file_name):
log("Saving output to " + file_name, 1)
with open(file_name, 'w+', encoding='utf-8') as file:
if hasattr(list_of_posts[0], '__dict__'):
json.dump([ob.__dict__ for ob in list_of_posts], file, indent=4, ensure_ascii=False)
else:
json.dump([ob for ob in list_of_posts], file, indent=4, ensure_ascii=False)
def main(language):
log(f"Analyzing {language} source", 1)
dir1 = '_1_process_raw_data/output' # use dir1 to analyze raw source
dir2 = '_2_remove_unrelated_data' # use dir2 to analyze filtered source
all_posts = load_posts(f'analysis/{dir1}/{language}.json')
all_posts = [x for x in all_posts if x["date"] != '']
standardized = standardize_date_format(all_posts)
dic = {}
date_list = []
for p in standardized:
source = p['source']
date = p['date']
if not source in dic:
dic[source] = []
if date > MIN_DATE:
dic[source].append(date)
date_list.append(date)
date_list = filter_date(date_list)
for source_name in dic:
dic[source_name] = filter_date(dic[source_name])
print(source_name)
print(len(dic[source_name]))
dic[source_name] = group_dates(MIN_DATE, MAX_DATE, dic[source_name])
with open(f'analysis/_5_analyze_source/{language}_source.json',
'w') as outfile:
print("Saving ", {language})
json.dump(dic, outfile)
def parse_files_from(self, directory, parser):
log("Parsing files from " + directory, 2)
posts = []
for file in os.listdir(directory):
if not file.endswith('.csv') and not file.endswith('.json'):
continue
posts += parser(directory + file)
return posts
def main(jobs, language):
posts = []
log(f"Parsing {language} posts", 1)
# Run every job
for job in jobs:
posts += job.run()
save_posts(posts, f'analysis/_1_process_raw_data/output/{language}.json')
log(f"Number of {language} posts created : " + str(len(posts)), 1)
def label_post(all_posts, all_labels):
log("Labelling post", 1)
for post in all_posts:
post['related_to'] = []
for label in all_labels:
# This will make sure 'lks' would not match 'talks'
pattern = f"(^|[^a-z]){label}([^a-z]|$)"
if re.search(pattern, post['value']) is not None:
if (label not in post['related_to']):
post['related_to'].append(label)
def plot_wordcloud(word_freq, title):
from wordcloud import WordCloud
if('chinese' in title):
wc = WordCloud(font_path='/usr/share/fonts/opentype/noto/NotoSansCJK-Light.ttc').fit_words(word_freq)
else:
wc = WordCloud().fit_words(word_freq)
import matplotlib.pyplot as plt
plt.imshow(wc, interpolation='bilinear')
plt.axis("off")
target_path = f'analysis/_6_analyze_keyword/{title}.png'
log("Saving word cloud as " + target_path, 2)
plt.savefig(target_path)
def for_english():
# all_posts = load_posts('analysis/_2_remove_unrelated_data/english.json') # this line is to check whether this new algorithm differ from the previous algo
all_posts = load_posts('analysis/_1_process_raw_data/output/english.json')
leaders = json.load(open(f'keywords/target/leader.json'))
parties = json.load(open(f'keywords/target/party.json'))
combined = {**leaders, **parties}
keyword_dict = {}
for key, value in combined.items():
keyword_dict[key] = [key] + combined[key]["alias_en"]
kp = KeywordProcessor()
kp.add_keywords_from_dict(keyword_dict)
for p in all_posts:
p["related_to"] = list(set(kp.extract_keywords(p["value"])))
purified = [x for x in all_posts if len(x['related_to']) > 0]
log(f"Number of removed posts = " + str(len(all_posts) - len(purified)), 1)
save_posts(purified, f'analysis/_2_remove_unrelated_data/english.json')
def load_labels(file_path, language):
log(f"Loading {language} labels from {file_path}", 1)
result = []
with open(file_path, encoding='utf8') as file:
dic = json.load(file)
for key in dic:
if language == 'english':
result.append(key)
result += dic[key]['alias_en']
elif language == 'chinese':
x = dic[key]['name_cn']
if x is not None:
result.append(x)
result += dic[key]['alias_cn']
else:
raise Exception(
"Invalid language argument. Expected (chinese/english) but got "
+ language)
log(f"{len(result)} {language} labels loaded", 2)
return result
def main(language):
log(f"Loading {language} posts", 1)
# 1. Load the post data
posts = load_posts(f'analysis/_1_process_raw_data/output/{language}.json')
# 2. Load the labels, AKA keywords for political figure/party (example "Najib", "Pakatan Harapan")
labels = get_labels(language)
# 3. Match the labels with every post object
label_post(posts, labels)
# 4. Remove post that is not related to any keywords
log(f"Removing unrelated posts", 1)
purified = [x for x in posts if len(x['related_to']) > 0]
# 5. Save the remaning post object
log(f"Number of removed posts = " + str(len(posts) - len(purified)), 1)
save_posts(purified, f'analysis/_2_remove_unrelated_data/{language}.json')
# 6. This step is to save the post object that is not related to any keywords
SAVE_DUMPED_POST = False
if SAVE_DUMPED_POST:
dumped = [x for x in posts if len(x['related_to']) <= 0]
save_posts(
dumped, f'analysis/_2_remove_unrelated_data/dumped_{language}.json')
def plot_hbar(y_labels, x_values, title):
import matplotlib.pyplot as plt
import numpy as np
# Fixing random state for reproducibility
plt.rcdefaults()
fig, ax = plt.subplots()
fig.set_size_inches(19, 20)
y_pos = np.arange(len(y_labels))
error = np.random.rand(len(y_labels))
ax.barh(y_pos, x_values, xerr=error, align='center',
color='green', ecolor='black')
ax.set_yticks(y_pos)
ax.set_yticklabels(y_labels)
ax.invert_yaxis() # labels read top-to-bottom
ax.set_xlabel('Frequencies')
ax.set_title(title)
# plt.show()
target_path = f'analysis/_6_analyze_keyword/{title}.png'
log("Saving graph as " + target_path, 2)
plt.savefig(target_path)
def main(
language,
number_of_keywords_to_be_shown,
semantic_type
):
log(f'Analyzing {semantic_type} keywords for {language} data', 1)
if semantic_type == 'positive':
keyword_processor = load_semantic_keywords_processor('', True, False, language)
elif semantic_type == 'negative':
keyword_processor = load_semantic_keywords_processor('', False, True, language)
else:
raise Exception("Invalid argument")
posts = load_posts(f'analysis/_3_label_semantic/{language}.json')
dic = {}
for p in posts:
matching_keywords = keyword_processor.extract_keywords(p["value"])
for word in matching_keywords:
if not word in dic:
dic[word] = []
dic[word].append(p["origin"])
json.dump(dic, open(
f'analysis/_6_analyze_keyword/{language}_keyword_freq.json', 'w'), ensure_ascii=False)
flattened_dic = {}
for key in dic:
flattened_dic[key] = len(dic[key])
tuples = sorted(flattened_dic.items(),
key=operator.itemgetter(1), reverse=True)
y_labels = []
x_values = []
for t in tuples:
y_labels.append(t[0])
x_values.append(t[1])
y_labels = y_labels[0: number_of_keywords_to_be_shown + 1]
x_values = x_values[0: number_of_keywords_to_be_shown + 1]
file_name = f'analysis/_6_analyze_keyword/{language}_top_{semantic_type}_keywords.txt'
with open(file_name, 'w') as file:
for value in y_labels:
file.write(value + "\n")
log(f'Top 100 keywords are save as {file_name}', 2)
# log('Plotting graph', 2)
# plot_hbar(y_labels, x_values, f'{semantic_type}_keyword_frequencies')
# log('DONE', 1)
log('Plotting word cloud', 2)
plot_wordcloud(dict(zip(y_labels, x_values)), f'{language}_{semantic_type}_keywordcloud')
def main(language):
# 1. Load the post object generated by _3_label_semantic
post_str = str(load_posts(f'analysis/_3_label_semantic/{language}.json'))
# 2. Replace synonyms, (for example: replace 'Najib','PM' with 'najib')
post_str = replace_synonyms(post_str, "../../keywords/target/")
all_posts = eval(replace_synonyms(post_str, get_keywords(language)))
# 3. Standardize date format of every Post object
log(f"Standardizing date format of each {language} posts", 0)
standardized = standardize_date_format(all_posts)
log(f"Sorting post based on date", 1)
# 4. Sort the Post object based on date
sorted_posts = sorted(standardized, key=lambda x: x['date'])
# 5. Extract the data for the sorted post
extracted = extract_data(sorted_posts, START_DATE, END_DATE, language)
log(f'Storing results to analysis/results/{language}_extracted.json', 1)
# 6. Save the extracted data as XXXX_extracted.json
json.dump(extracted,
open(f'analysis/results/{language}_extracted.json', 'w'),
ensure_ascii=False)
def load_posts(file_path):
log("Loading posts from " + file_path, 1)
with open(file_path, 'rb') as file:
posts = json.load(file)
return posts
