def get_danmaku_from_range(all_children, he_range):
start, end = he_range
start += 45
end += 45
return [
item.text for item in all_children
if item.tag == 'd' and start <= get_time(item) <= end
]
def get_heat_time(all_children):
interval = 2
center = 0
cur_entry = 0
final_time = get_time(all_children[-1])
cur_heat = 0
danmaku_queue = deque()
heat_time = [[], []]
while True:
if center > final_time:
break
start = center - interval
end = center + interval
while cur_entry < len(all_children) and get_time(
all_children[cur_entry]) < end:
cur_danmaku = all_children[cur_entry]
danmaku_queue.append(cur_danmaku)
cur_heat += get_value(cur_danmaku)
cur_entry += 1
while len(danmaku_queue) != 0 and get_time(danmaku_queue[0]) < start:
prev_danmaku = danmaku_queue.popleft()
cur_heat -= get_value(prev_danmaku)
heat_time[0] += [center]
heat_time[1] += [cur_heat]
center += 1
heat_value = heat_time[1]
heat_value_gaussian = half_gaussian_filter(heat_value, sigma=50)
heat_value_gaussian2 = half_gaussian_filter(heat_value, sigma=1000) * 1.2
he_points = [[], []]
cur_highest = -1
highest_idx = -1
he_start = -1
he_range = []
for i in range(len(heat_value_gaussian)):
if highest_idx != -1:
assert he_start != -1
if heat_value_gaussian[i] < heat_value_gaussian2[i]:
he_points[0] += [highest_idx]
he_points[1] += [cur_highest]
he_range += [(he_start, i)]
highest_idx = -1
he_start = -1
else:
if heat_value_gaussian[i] > cur_highest:
cur_highest = heat_value_gaussian[i]
highest_idx = i
else:
assert he_start == -1
if heat_value_gaussian[i] > heat_value_gaussian2[i]:
cur_highest = heat_value_gaussian[i]
highest_idx = i
he_start = i
# Usually the HE point at the end of a live stream is just to say goodbye
# if highest_idx != -1:
# he_points[0] += [highest_idx]
# he_points[1] += [cur_highest]
return heat_time, heat_value_gaussian / np.sqrt(
heat_value_gaussian2), np.sqrt(
heat_value_gaussian2), he_points, he_range
if args.he_map is not None:
he_pairs = heat_values[3]
all_timestamps = heat_values[0][0]
heat_comments = []
xml_list_iter = iter(xml_list)
tr4s = TextRank4Sentence()
for start, end in tqdm(heat_values[4]):
comment_list = []
while True:
try:
element = next(xml_list_iter)
except StopIteration:
break
if get_time(element) <= start + 45:
continue
if get_time(element) > end + 45:
break
if element.tag == 'd':
text = element.text
if text is not None and not text.replace(
" ", "").replace("哈", "") == "":
comment_list += [text]
print(len(comment_list))
if len(comment_list) > 1000:
comment_list = random.sample(comment_list, 1000)
tr4s.analyze("\n".join(comment_list),
lower=True,
source='no_filter')
key_sentences = tr4s.get_key_sentences(1)