Spaces:
Sleeping
Sleeping
Commit
·
bbed939
1
Parent(s):
2293223
Implement term ranking and plot comparison
Browse files- README.md +1 -1
- app.py +168 -63
- requirements.txt +4 -1
README.md
CHANGED
@@ -4,7 +4,7 @@ emoji: 🏢
|
|
4 |
colorFrom: gray
|
5 |
colorTo: indigo
|
6 |
sdk: gradio
|
7 |
-
sdk_version:
|
8 |
app_file: app.py
|
9 |
pinned: false
|
10 |
short_description: final project for IAT360
|
|
|
4 |
colorFrom: gray
|
5 |
colorTo: indigo
|
6 |
sdk: gradio
|
7 |
+
sdk_version: 5.8.0
|
8 |
app_file: app.py
|
9 |
pinned: false
|
10 |
short_description: final project for IAT360
|
app.py
CHANGED
@@ -1,77 +1,182 @@
|
|
1 |
import gradio as gr
|
2 |
from transformers import pipeline
|
3 |
import PyPDF2
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
|
5 |
# Load the token classification pipeline
|
6 |
model_name = "jjzha/jobbert_knowledge_extraction"
|
7 |
-
pipe = pipeline("token-classification", model=model_name)
|
8 |
-
|
9 |
-
#
|
10 |
-
def
|
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 |
interface = gr.Interface(
|
63 |
-
fn=
|
64 |
inputs=[
|
65 |
-
gr.
|
66 |
-
gr.
|
67 |
],
|
68 |
outputs=[
|
69 |
-
|
70 |
-
gr.
|
71 |
-
gr.
|
|
|
72 |
],
|
73 |
-
title="Resume vs Job Posting
|
74 |
-
description="Upload your resume and enter a job posting. The app will highlight key
|
75 |
)
|
76 |
|
77 |
# Launch the Gradio app
|
|
|
1 |
import gradio as gr
|
2 |
from transformers import pipeline
|
3 |
import PyPDF2
|
4 |
+
from PIL import Image
|
5 |
+
import matplotlib.pyplot as plt
|
6 |
+
from io import BytesIO
|
7 |
+
import pandas as pd # For displaying rankings in a table
|
8 |
+
import re
|
9 |
+
import math
|
10 |
|
11 |
# Load the token classification pipeline
|
12 |
model_name = "jjzha/jobbert_knowledge_extraction"
|
13 |
+
pipe = pipeline("token-classification", model=model_name, aggregation_strategy="first")
|
14 |
+
|
15 |
+
# Aggregate overlapping or adjacent spans into 1 entity
|
16 |
+
def aggregate_span(results):
|
17 |
+
new_results = []
|
18 |
+
current_result = results[0]
|
19 |
+
for result in results[1:]:
|
20 |
+
if result["start"] == current_result["end"] + 1:
|
21 |
+
current_result["word"] += " " + result["word"]
|
22 |
+
current_result["end"] = result["end"]
|
23 |
+
else:
|
24 |
+
new_results.append(current_result)
|
25 |
+
current_result = result
|
26 |
+
new_results.append(current_result)
|
27 |
+
return new_results
|
28 |
+
|
29 |
+
# Extract knowledge entities from job posting
|
30 |
+
def ner(text):
|
31 |
+
output_knowledge = pipe(text)
|
32 |
+
for result in output_knowledge:
|
33 |
+
if result.get("entity_group"):
|
34 |
+
result["entity"] = "Knowledge"
|
35 |
+
del result["entity_group"]
|
36 |
+
if len(output_knowledge) > 0:
|
37 |
+
output_knowledge = aggregate_span(output_knowledge)
|
38 |
+
|
39 |
+
return {"text": text, "entities": output_knowledge}
|
40 |
+
|
41 |
+
# Extract text from input PDF
|
42 |
+
def extract_pdf(pdf_file):
|
43 |
+
reader = PyPDF2.PdfReader(pdf_file)
|
44 |
+
text = ''
|
45 |
+
for page in reader.pages:
|
46 |
+
text += page.extract_text()
|
47 |
+
return text
|
48 |
+
|
49 |
+
def rank_knowledge(entities, job_posting_text, resume_text):
|
50 |
+
scores = {}
|
51 |
+
priority_keywords = ["must-have", "required", "preferred", "key", "important"]
|
52 |
+
|
53 |
+
for entity in entities:
|
54 |
+
term = entity["word"].lower()
|
55 |
+
term_score = 0
|
56 |
+
|
57 |
+
# Count exact matches of the term in the job posting
|
58 |
+
term_score += len(re.findall(rf'\b{re.escape(term)}\b', job_posting_text.lower()))
|
59 |
+
|
60 |
+
# Proximity to priority keywords
|
61 |
+
term_positions = [m.start() for m in re.finditer(rf'\b{re.escape(term)}\b', job_posting_text.lower())]
|
62 |
+
for keyword in priority_keywords:
|
63 |
+
keyword_positions = [m.start() for m in re.finditer(rf'\b{re.escape(keyword)}\b', job_posting_text.lower())]
|
64 |
+
for t_pos in term_positions:
|
65 |
+
for k_pos in keyword_positions:
|
66 |
+
if abs(t_pos - k_pos) < 20: # Within 20 characters
|
67 |
+
term_score += 1
|
68 |
+
|
69 |
+
scores[term] = term_score
|
70 |
+
|
71 |
+
# Normalize
|
72 |
+
max_score = max(scores.values(), default=1)
|
73 |
+
ranked_entities = [
|
74 |
+
{
|
75 |
+
"Term": k,
|
76 |
+
"Score": (math.log1p(v) / math.log1p(max_score)) * 100, # Log scaling
|
77 |
+
"In Resume": "Yes" if k in resume_text.lower() else "No"
|
78 |
+
}
|
79 |
+
for k, v in scores.items()
|
80 |
+
]
|
81 |
+
|
82 |
+
ranked_entities.sort(key=lambda x: x["Score"], reverse=True)
|
83 |
+
return ranked_entities
|
84 |
+
|
85 |
+
# Compare extracted knowledge entities with the resume
|
86 |
+
def compare_with_resume(output_knowledge, resume_file):
|
87 |
+
resume_text = extract_pdf(resume_file) if resume_file else ''
|
88 |
+
matched_knowledge = []
|
89 |
+
unmatched_knowledge = []
|
90 |
+
|
91 |
+
for entity in output_knowledge:
|
92 |
+
if entity["word"].lower() in resume_text.lower():
|
93 |
+
matched_knowledge.append(entity["word"])
|
94 |
+
else:
|
95 |
+
unmatched_knowledge.append(entity["word"])
|
96 |
+
return matched_knowledge, unmatched_knowledge
|
97 |
+
|
98 |
+
def plot_comparison(matched_knowledge, unmatched_knowledge):
|
99 |
+
labels = ['Matched', 'Unmatched']
|
100 |
+
values = [len(matched_knowledge), len(unmatched_knowledge)]
|
101 |
+
total = sum(values)
|
102 |
+
percentages = [f"{(value / total * 100):.1f}%" for value in values]
|
103 |
+
|
104 |
+
plt.figure(figsize=(6, 4))
|
105 |
+
bars = plt.bar(labels, values, color=['green', 'red'])
|
106 |
+
plt.xlabel('Knowledge Match Status')
|
107 |
+
plt.ylabel('Count')
|
108 |
+
plt.title('Knowledge Match Comparison')
|
109 |
+
plt.tight_layout()
|
110 |
+
|
111 |
+
# Add percentage labels above bars
|
112 |
+
for bar, percentage in zip(bars, percentages):
|
113 |
+
plt.text(bar.get_x() + bar.get_width() / 2, bar.get_height() + 0.1, percentage, ha='center', fontsize=10)
|
114 |
+
|
115 |
+
buf = BytesIO()
|
116 |
+
plt.savefig(buf, format='png')
|
117 |
+
buf.seek(0)
|
118 |
+
plt.close()
|
119 |
+
|
120 |
+
return Image.open(buf)
|
121 |
+
|
122 |
+
def plot_pie_chart(ranked_knowledge, threshold=50):
|
123 |
+
# Filter terms above the threshold
|
124 |
+
filtered_terms = [term for term in ranked_knowledge if term["Score"] >= threshold]
|
125 |
+
matched_terms = sum(1 for term in filtered_terms if term["In Resume"] == "Yes")
|
126 |
+
unmatched_terms = len(filtered_terms) - matched_terms
|
127 |
+
|
128 |
+
# Data for pie chart
|
129 |
+
labels = ['Matched', 'Unmatched']
|
130 |
+
values = [matched_terms, unmatched_terms]
|
131 |
+
|
132 |
+
# Create pie chart
|
133 |
+
plt.figure(figsize=(6, 4))
|
134 |
+
plt.pie(values, labels=labels, autopct='%1.1f%%', colors=['green', 'red'], startangle=90)
|
135 |
+
plt.title(f"Terms Above Threshold (Score >= {threshold})")
|
136 |
+
buf = BytesIO()
|
137 |
+
plt.savefig(buf, format='png')
|
138 |
+
buf.seek(0)
|
139 |
+
plt.close()
|
140 |
+
return Image.open(buf)
|
141 |
+
|
142 |
+
def ner_and_compare_with_plot_and_rank(job_posting_text, resume_file):
|
143 |
+
"""Combined function to process NER, comparison, ranking, and visualization."""
|
144 |
+
ner_result = ner(job_posting_text)
|
145 |
+
resume_text = extract_pdf(resume_file) if resume_file else ''
|
146 |
+
matched_knowledge, unmatched_knowledge = compare_with_resume(ner_result["entities"], resume_file)
|
147 |
+
comparison_result = {
|
148 |
+
"Matched Knowledge": matched_knowledge,
|
149 |
+
"Unmatched Knowledge": unmatched_knowledge,
|
150 |
+
}
|
151 |
+
bar_plot = plot_comparison(matched_knowledge, unmatched_knowledge)
|
152 |
+
|
153 |
+
# Ranking knowledge entities with "In Resume" column
|
154 |
+
ranked_knowledge = rank_knowledge(ner_result["entities"], job_posting_text, resume_text)
|
155 |
+
|
156 |
+
# Generate pie chart for a fixed threshold
|
157 |
+
pie_chart = plot_pie_chart(ranked_knowledge, threshold=50)
|
158 |
+
|
159 |
+
# Convert ranked knowledge to a DataFrame for better display
|
160 |
+
ranked_df = pd.DataFrame(ranked_knowledge)
|
161 |
+
|
162 |
+
return ner_result, ranked_df, bar_plot, pie_chart
|
163 |
+
|
164 |
+
|
165 |
+
# Gradio interface setup
|
166 |
interface = gr.Interface(
|
167 |
+
fn=ner_and_compare_with_plot_and_rank,
|
168 |
inputs=[
|
169 |
+
gr.Textbox(label="Enter Job Posting Text", lines=20, placeholder="Paste job posting text here..."),
|
170 |
+
gr.File(label="Upload a PDF of your resume")
|
171 |
],
|
172 |
outputs=[
|
173 |
+
"highlight", # Highlighted job posting text with extracted entities
|
174 |
+
gr.DataFrame(label="Ranked Knowledge"), # Ranked knowledge table
|
175 |
+
gr.Image(label="Pie Chart for Terms Above Threshold")
|
176 |
+
gr.Image(label="Comparison Chart"), # Bar chart visualization
|
177 |
],
|
178 |
+
title="Resume vs Job Posting Knowledge Match with Highlights and Rankings",
|
179 |
+
description="Upload your resume and enter a job posting. The app will highlight key knowledge from the job posting, check if they are present in your resume, visualize the comparison, and rank knowledge terms based on importance.",
|
180 |
)
|
181 |
|
182 |
# Launch the Gradio app
|
requirements.txt
CHANGED
@@ -2,4 +2,7 @@ torch
|
|
2 |
transformers
|
3 |
gradio # if using Gradio
|
4 |
streamlit # if using Streamlit
|
5 |
-
pyPDF2
|
|
|
|
|
|
|
|
2 |
transformers
|
3 |
gradio # if using Gradio
|
4 |
streamlit # if using Streamlit
|
5 |
+
pyPDF2
|
6 |
+
matplotlib
|
7 |
+
PIL
|
8 |
+
pandas
|