File size: 4,497 Bytes
c4aa562 aa1d6fa c4aa562 aa1d6fa c4aa562 aa1d6fa c4aa562 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 |
import streamlit as st
import spacy
import numpy as np
from gensim import corpora, models
from itertools import chain
from sklearn.preprocessing import MultiLabelBinarizer
from sklearn.metrics.pairwise import cosine_similarity
from itertools import islice
from scipy.signal import argrelmax
nlp = spacy.load('en_core_web_sm')
def window(seq, n=3):
it = iter(seq)
result = tuple(islice(it, n))
if len(result) == n:
yield result
for elem in it:
result = result[1:] + (elem,)
yield result
def get_depths(scores):
def climb(seq, i, mode='left'):
if mode == 'left':
while True:
curr = seq[i]
if i == 0:
return curr
i = i-1
if not seq[i] > curr:
return curr
if mode == 'right':
while True:
curr = seq[i]
if i == (len(seq)-1):
return curr
i = i+1
if not seq[i] > curr:
return curr
depths = []
for i in range(len(scores)):
score = scores[i]
l_peak = climb(scores, i, mode='left')
r_peak = climb(scores, i, mode='right')
depth = 0.5 * (l_peak + r_peak - (2*score))
depths.append(depth)
return np.array(depths)
def get_local_maxima(depth_scores, order=1):
maxima_ids = argrelmax(depth_scores, order=order)[0]
filtered_scores = np.zeros(len(depth_scores))
filtered_scores[maxima_ids] = depth_scores[maxima_ids]
return filtered_scores
def compute_threshold(scores):
s = scores[np.nonzero(scores)]
threshold = np.mean(s) - (np.std(s) / 2)
return threshold
def get_threshold_segments(scores, threshold=0.1):
segment_ids = np.where(scores >= threshold)[0]
return segment_ids
def print_list(lst):
for e in lst:
st.markdown("- " + e)
st.subheader("Topic Modeling with Segmentation")
uploaded_file = st.file_uploader("choose a text file", type=["txt"])
if uploaded_file is not None:
st.session_state["text"] = uploaded_file.getvalue().decode('utf-8')
st.write("OR")
input_text = st.text_area(
label="Enter text separated by newlines",
value="",
key="text",
height=150
)
button=st.button('Get Segments')
if (button==True) and input_text != "":
texts = input_text.split('\n')
sents = []
for text in texts:
doc = nlp(text)
for sent in doc.sents:
sents.append(sent)
MIN_LENGTH = 3
tokenized_sents = [[token.lemma_.lower() for token in sent if
not token.is_stop and not token.is_punct and token.text.strip() and len(token) >= MIN_LENGTH]
for sent in sents]
st.write("Modeling topics:")
np.random.seed(123)
N_TOPICS = 5
N_PASSES = 5
dictionary = corpora.Dictionary(tokenized_sents)
bow = [dictionary.doc2bow(sent) for sent in tokenized_sents]
topic_model = models.LdaModel(corpus=bow, id2word=dictionary, num_topics=N_TOPICS, passes=N_PASSES)
st.write("inferring topics ...")
THRESHOLD = 0.05
doc_topics = list(topic_model.get_document_topics(bow, minimum_probability=THRESHOLD))
k = 3
top_k_topics = [[t[0] for t in sorted(sent_topics, key=lambda x: x[1], reverse=True)][:k]
for sent_topics in doc_topics]
WINDOW_SIZE = 3
window_topics = window(top_k_topics, n=WINDOW_SIZE)
window_topics = [list(set(chain.from_iterable(window))) for window in window_topics]
binarizer = MultiLabelBinarizer(classes=range(N_TOPICS))
encoded_topic = binarizer.fit_transform(window_topics)
st.write("generating segments ...")
sims_topic = [cosine_similarity([pair[0]], [pair[1]])[0][0] for pair in zip(encoded_topic, encoded_topic[1:])]
depths_topic = get_depths(sims_topic)
filtered_topic = get_local_maxima(depths_topic, order=1)
threshold_topic = compute_threshold(filtered_topic)
threshold_segments_topic = get_threshold_segments(filtered_topic, threshold_topic)
segment_ids = threshold_segments_topic + WINDOW_SIZE
segment_ids = [0] + segment_ids.tolist() + [len(sents)]
slices = list(zip(segment_ids[:-1], segment_ids[1:]))
segmented = [sents[s[0]: s[1]] for s in slices]
for segment in segmented[:-1]:
print_list([s.text for s in segment])
st.markdown("""---""")
print_list([s.text for s in segmented[-1]]) |