Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -5,9 +5,93 @@ from transformers import pipeline, AutoModelForTokenClassification, AutoTokenize
|
|
5 |
import PyPDF2
|
6 |
import docx
|
7 |
import io
|
8 |
-
import re
|
9 |
|
10 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
11 |
|
12 |
def create_mask_dict(entities):
|
13 |
mask_dict = {}
|
@@ -28,6 +112,7 @@ def create_masked_text(input_text, mask_dict):
|
|
28 |
masked_text = re.sub(r'\b' + re.escape(word) + r'\b', mask, masked_text)
|
29 |
return masked_text
|
30 |
|
|
|
31 |
Run_Button = st.button("Run")
|
32 |
|
33 |
if Run_Button and input_text:
|
@@ -49,34 +134,47 @@ if Run_Button and input_text:
|
|
49 |
entity['end'] += offset
|
50 |
|
51 |
all_outputs.extend(output)
|
|
|
52 |
|
53 |
# Combine entities
|
|
|
54 |
output_comb = entity_comb(all_outputs)
|
55 |
|
56 |
# Create mask dictionary
|
57 |
mask_dict = create_mask_dict(output_comb)
|
58 |
|
59 |
-
|
60 |
-
masked_text = create_masked_text(input_text, mask_dict)
|
61 |
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
df = pd.DataFrame
|
70 |
-
|
71 |
-
|
|
|
|
|
|
|
72 |
|
73 |
-
|
74 |
-
|
|
|
75 |
spacy_display = {"ents": [], "text": input_text, "title": None}
|
76 |
for entity in output_comb:
|
77 |
if entity['entity_group'] not in ['CARDINAL', 'EVENT']:
|
78 |
-
label = mask_dict[entity['word']]
|
79 |
-
|
|
|
|
|
80 |
|
81 |
html = spacy.displacy.render(spacy_display, style="ent", minify=True, manual=True)
|
82 |
-
st.write(html, unsafe_allow_html=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
import PyPDF2
|
6 |
import docx
|
7 |
import io
|
|
|
8 |
|
9 |
+
def chunk_text(text, chunk_size=128):
|
10 |
+
words = text.split()
|
11 |
+
chunks = []
|
12 |
+
current_chunk = []
|
13 |
+
current_length = 0
|
14 |
+
|
15 |
+
for word in words:
|
16 |
+
if current_length + len(word) + 1 > chunk_size:
|
17 |
+
chunks.append(' '.join(current_chunk))
|
18 |
+
current_chunk = [word]
|
19 |
+
current_length = len(word)
|
20 |
+
else:
|
21 |
+
current_chunk.append(word)
|
22 |
+
current_length += len(word) + 1
|
23 |
+
|
24 |
+
if current_chunk:
|
25 |
+
chunks.append(' '.join(current_chunk))
|
26 |
+
|
27 |
+
return chunks
|
28 |
+
|
29 |
+
st.set_page_config(layout="wide")
|
30 |
+
|
31 |
+
# Function to read text from uploaded file
|
32 |
+
def read_file(file):
|
33 |
+
if file.type == "text/plain":
|
34 |
+
return file.getvalue().decode("utf-8")
|
35 |
+
elif file.type == "application/pdf":
|
36 |
+
pdf_reader = PyPDF2.PdfReader(io.BytesIO(file.getvalue()))
|
37 |
+
return " ".join(page.extract_text() for page in pdf_reader.pages)
|
38 |
+
elif file.type == "application/vnd.openxmlformats-officedocument.wordprocessingml.document":
|
39 |
+
doc = docx.Document(io.BytesIO(file.getvalue()))
|
40 |
+
return " ".join(paragraph.text for paragraph in doc.paragraphs)
|
41 |
+
else:
|
42 |
+
st.error("Unsupported file type")
|
43 |
+
return None
|
44 |
+
|
45 |
+
st.title("Turkish NER Models Testing")
|
46 |
+
|
47 |
+
model_list = [
|
48 |
+
'girayyagmur/bert-base-turkish-ner-cased',
|
49 |
+
'savasy/bert-base-turkish-ner-cased',
|
50 |
+
'xlm-roberta-large-finetuned-conll03-english',
|
51 |
+
'asahi417/tner-xlm-roberta-base-ontonotes5'
|
52 |
+
]
|
53 |
+
|
54 |
+
st.sidebar.header("Select NER Model")
|
55 |
+
model_checkpoint = st.sidebar.radio("", model_list)
|
56 |
+
|
57 |
+
st.sidebar.write("For details of models: 'https://huggingface.co/akdeniz27/")
|
58 |
+
st.sidebar.write("Only PDF, DOCX, and TXT files are supported.")
|
59 |
+
|
60 |
+
# Determine aggregation strategy
|
61 |
+
aggregation = "simple" if model_checkpoint in ["akdeniz27/xlm-roberta-base-turkish-ner", "xlm-roberta-large-finetuned-conll03-english", "asahi417/tner-xlm-roberta-base-ontonotes5"] else "first"
|
62 |
+
|
63 |
+
st.subheader("Select Text Input Method")
|
64 |
+
input_method = st.radio("", ('Write or Paste New Text', 'Upload File'))
|
65 |
+
|
66 |
+
if input_method == "Write or Paste New Text":
|
67 |
+
input_text = st.text_area('Write or Paste Text Below', value="", height=128)
|
68 |
+
else:
|
69 |
+
uploaded_file = st.file_uploader("Choose a file", type=["txt", "pdf", "docx"])
|
70 |
+
if uploaded_file is not None:
|
71 |
+
input_text = read_file(uploaded_file)
|
72 |
+
if input_text:
|
73 |
+
st.text_area("Extracted Text", input_text, height=128)
|
74 |
+
else:
|
75 |
+
input_text = ""
|
76 |
+
|
77 |
+
@st.cache_resource
|
78 |
+
def setModel(model_checkpoint, aggregation):
|
79 |
+
model = AutoModelForTokenClassification.from_pretrained(model_checkpoint)
|
80 |
+
tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)
|
81 |
+
return pipeline('ner', model=model, tokenizer=tokenizer, aggregation_strategy=aggregation)
|
82 |
+
|
83 |
+
@st.cache_resource
|
84 |
+
def entity_comb(output):
|
85 |
+
output_comb = []
|
86 |
+
for ind, entity in enumerate(output):
|
87 |
+
if ind == 0:
|
88 |
+
output_comb.append(entity)
|
89 |
+
elif output[ind]["start"] == output[ind-1]["end"] and output[ind]["entity_group"] == output[ind-1]["entity_group"]:
|
90 |
+
output_comb[-1]["word"] += output[ind]["word"]
|
91 |
+
output_comb[-1]["end"] = output[ind]["end"]
|
92 |
+
else:
|
93 |
+
output_comb.append(entity)
|
94 |
+
return output_comb
|
95 |
|
96 |
def create_mask_dict(entities):
|
97 |
mask_dict = {}
|
|
|
112 |
masked_text = re.sub(r'\b' + re.escape(word) + r'\b', mask, masked_text)
|
113 |
return masked_text
|
114 |
|
115 |
+
|
116 |
Run_Button = st.button("Run")
|
117 |
|
118 |
if Run_Button and input_text:
|
|
|
134 |
entity['end'] += offset
|
135 |
|
136 |
all_outputs.extend(output)
|
137 |
+
|
138 |
|
139 |
# Combine entities
|
140 |
+
|
141 |
output_comb = entity_comb(all_outputs)
|
142 |
|
143 |
# Create mask dictionary
|
144 |
mask_dict = create_mask_dict(output_comb)
|
145 |
|
146 |
+
masked_text = create_masked_text(input_text, output_comb, mask_dict)
|
|
|
147 |
|
148 |
+
# Apply masking and add masked_word column
|
149 |
+
for entity in output_comb:
|
150 |
+
if entity['entity_group'] not in ['CARDINAL', 'EVENT']:
|
151 |
+
entity['masked_word'] = mask_dict.get(entity['word'], entity['word'])
|
152 |
+
else:
|
153 |
+
entity['masked_word'] = entity['word']
|
154 |
+
print("output_comb", output_comb)
|
155 |
+
#df = pd.DataFrame.from_dict(output_comb)
|
156 |
+
#cols_to_keep = ['word', 'entity_group', 'score', 'start', 'end']
|
157 |
+
#df_final = df[cols_to_keep].loc[:,~df.columns.duplicated()].copy()
|
158 |
+
|
159 |
+
#st.subheader("Recognized Entities")
|
160 |
+
#st.dataframe(df_final)
|
161 |
|
162 |
+
|
163 |
+
|
164 |
+
# Spacy display logic with entity numbering
|
165 |
spacy_display = {"ents": [], "text": input_text, "title": None}
|
166 |
for entity in output_comb:
|
167 |
if entity['entity_group'] not in ['CARDINAL', 'EVENT']:
|
168 |
+
label = f"{entity['entity_group']}_{mask_dict[entity['word']].split('_')[1]}"
|
169 |
+
else:
|
170 |
+
label = entity['entity_group']
|
171 |
+
spacy_display["ents"].append({"start": entity["start"], "end": entity["end"], "label": label})
|
172 |
|
173 |
html = spacy.displacy.render(spacy_display, style="ent", minify=True, manual=True)
|
174 |
+
st.write(html, unsafe_allow_html=True)
|
175 |
+
|
176 |
+
st.subheader("Masking Dictionary")
|
177 |
+
st.json(mask_dict)
|
178 |
+
|
179 |
+
st.subheader("Masked Text Preview")
|
180 |
+
st.text(masked_text)
|