pages/Workflow & Model Overview.py

import streamlit as st

# Custom CSS for better styling
st.markdown("""

    <style>

        .main-title {

            font-size: 36px;

            color: #4A90E2;

            font-weight: bold;

            text-align: center;

        }

        .sub-title {

            font-size: 24px;

            color: #4A90E2;

            margin-top: 20px;

        }

        .section {

            background-color: #f9f9f9;

            padding: 15px;

            border-radius: 10px;

            margin-top: 20px;

        }

        .section h2 {

            font-size: 22px;

            color: #4A90E2;

        }

        .section p, .section ul {

            color: #666666;

        }

        .link {

            color: #4A90E2;

            text-decoration: none;

        }

    </style>

""", unsafe_allow_html=True)

# Main Title
st.markdown('<div class="main-title">State-of-the-Art Named Entity Recognition with Spark NLP (Italian)</div>', unsafe_allow_html=True)

# Introduction
st.markdown("""

<div class="section">

    <p>Named Entity Recognition (NER) is the task of identifying important words in a text and associating them with a category. For example, we may be interested in finding all the personal names in documents, or company names in news articles. Other examples include domain-specific uses such as identifying all disease names in a clinical text, or company trading codes in financial ones.</p>

    <p>NER can be implemented with many approaches. In this post, we introduce a deep learning-based method using the NerDL model. This approach leverages the scalability of Spark NLP with Python.</p>

</div>

""", unsafe_allow_html=True)

# Introduction to Spark NLP
st.markdown('<div class="sub-title">Introduction to Spark NLP</div>', unsafe_allow_html=True)
st.markdown("""

<div class="section">

    <p>Spark NLP is an open-source library maintained by John Snow Labs. It is built on top of Apache Spark and Spark ML and provides simple, performant & accurate NLP annotations for machine learning pipelines that can scale easily in a distributed environment.</p>

    <p>To install Spark NLP, you can simply use any package manager like conda or pip. For example, using pip you can simply run <code>pip install spark-nlp</code>. For different installation options, check the official <a href="https://nlp.johnsnowlabs.com/docs/en/install" target="_blank" class="link">documentation</a>.</p>

</div>

""", unsafe_allow_html=True)

# Using NerDL Model
st.markdown('<div class="sub-title">Using NerDL Model</div>', unsafe_allow_html=True)
st.markdown("""

<div class="section">

    <p>The NerDL model in Spark NLP is a deep learning-based approach for NER tasks. It uses a Char CNNs - BiLSTM - CRF architecture that achieves state-of-the-art results in most datasets. The training data should be a labeled Spark DataFrame in the format of CoNLL 2003 IOB with annotation type columns.</p>

</div>

""", unsafe_allow_html=True)

# Setup Instructions
st.markdown('<div class="sub-title">Setup</div>', unsafe_allow_html=True)
st.markdown('<p>To install Spark NLP in Python, use your favorite package manager (conda, pip, etc.). For example:</p>', unsafe_allow_html=True)
st.code("""

pip install spark-nlp

pip install pyspark

""", language="bash")

st.markdown("<p>Then, import Spark NLP and start a Spark session:</p>", unsafe_allow_html=True)
st.code("""

import sparknlp



# Start Spark Session

spark = sparknlp.start()

""", language='python')

# Example Usage with NerDL Model in Italian
st.markdown('<div class="sub-title">Example Usage with NerDL Model in Italian</div>', unsafe_allow_html=True)
st.markdown("""

<div class="section">

    <p>Below is an example of how to set up and use the NerDL model for named entity recognition in Italian:</p>

</div>

""", unsafe_allow_html=True)
st.code('''

from sparknlp.base import *

from sparknlp.annotator import *

from pyspark.ml import Pipeline

from pyspark.sql.functions import col, expr, round, concat, lit



# Document Assembler

document_assembler = DocumentAssembler() \\

    .setInputCol("text") \\

    .setOutputCol("document")



# Tokenizer

tokenizer = Tokenizer() \\

    .setInputCols(["document"]) \\

    .setOutputCol("token")



# Word Embeddings

embeddings = WordEmbeddingsModel.pretrained('glove_840B_300', lang='xx') \\

    .setInputCols(["document", "token"]) \\

    .setOutputCol("embeddings")



# NerDL Model

ner_model = NerDLModel.pretrained('ner_wikiner_glove_840B_300', 'xx') \\

    .setInputCols(["document", "token", "embeddings"]) \\

    .setOutputCol("ner")



# NER Converter

ner_converter = NerConverter() \\

    .setInputCols(["document", "token", "ner"]) \\

    .setOutputCol("ner_chunk")



# Pipeline

pipeline = Pipeline(stages=[

    document_assembler,

    tokenizer,

    embeddings,

    ner_model,

    ner_converter

])



# Example sentence

example = """

Giuseppe Verdi nacque a Le Roncole, una frazione di Busseto, il 10 ottobre 1813. 

Era un compositore italiano, noto per le sue opere come La Traviata e Rigoletto.

"""



data = spark.createDataFrame([[example]]).toDF("text")



# Transforming data

result = pipeline.fit(data).transform(data)



# Select the result, entity, and confidence columns

result.select(

    expr("explode(ner_chunk) as ner_chunk")

).select(

    col("ner_chunk.result").alias("result"),

    col("ner_chunk.metadata").getItem("entity").alias("entity"),

    concat(

        round((col("ner_chunk.metadata").getItem("confidence").cast("float") * 100), 2),

        lit("%")

    ).alias("confidence")

).show(truncate=False)

''', language="python")

st.text("""

+--------------+------+----------+

|result        |entity|confidence|

+--------------+------+----------+

|Giuseppe Verdi|PER   |66.19%    |

|Le Roncole    |LOC   |51.45%    |

|Busseto       |LOC   |85.39%    |

|La Traviata   |MISC  |69.35%    |

|Rigoletto     |MISC  |93.34%    |

+--------------+------+----------+

""")

# Benchmark Section
st.markdown('<div class="sub-title">Benchmark</div>', unsafe_allow_html=True)
st.markdown("""

<div class="section">

    <p>Evaluating the performance of NER models is crucial to understanding their effectiveness in real-world applications. Below are the benchmark results for the "ner_wikiner_glove_840B_300" model on Italian text, focusing on various named entity categories. The metrics used include precision, recall, and F1-score, which are standard for evaluating classification models.</p>

</div>

""", unsafe_allow_html=True)

st.markdown("""

<div class="sub-title">Detailed Results</div>



| Entity Type | Precision | Recall | F1-Score | Support |

|-------------|:---------:|:------:|:--------:|--------:|

| **B-LOC**   | 0.88      | 0.92   | 0.90     | 13,050  |

| **I-ORG**   | 0.78      | 0.71   | 0.74     | 1,211   |

| **I-LOC**   | 0.89      | 0.85   | 0.87     | 7,454   |

| **I-PER**   | 0.93      | 0.94   | 0.94     | 4,539   |

| **B-ORG**   | 0.88      | 0.72   | 0.79     | 2,222   |

| **B-PER**   | 0.90      | 0.93   | 0.92     | 7,206   |



<div class="sub-title">Averages</div>



| Average Type | Precision | Recall | F1-Score | Support |

|--------------|:---------:|:------:|:--------:|--------:|

| **Micro**    | 0.89      | 0.89   | 0.89     | 35,682  |

| **Macro**    | 0.88      | 0.85   | 0.86     | 35,682  |

| **Weighted** | 0.89      | 0.89   | 0.89     | 35,682  |



<div class="sub-title">Category-Specific Performance</div>



| Category | Precision | Recall | F1-Score | Support |

|----------|:---------:|:------:|:--------:|--------:|

| **LOC**  | 86.33%    | 90.53% | 88.38    | 13,685  |

| **MISC** | 81.88%    | 67.03% | 73.72    | 3,069   |

| **ORG**  | 85.91%    | 70.52% | 77.46    | 1,824   |

| **PER**  | 89.54%    | 92.08% | 90.79    | 7,410   |



<div class="sub-title">Additional Metrics</div>



- **Processed Tokens:** 349,242

- **Total Phrases:** 26,227

- **Found Phrases:** 25,988

- **Correct Phrases:** 22,529

- **Accuracy (non-O):** 85.99%

- **Overall Accuracy:** 98.06%

""", unsafe_allow_html=True)

# Summary
st.markdown('<div class="sub-title">Summary</div>', unsafe_allow_html=True)
st.markdown("""

<div class="section">

    <p>In this article, we discussed named entity recognition using a deep learning-based method with the "wikiner_840B_300" model for Italian. We introduced how to perform the task using the open-source Spark NLP library with Python, which can be used at scale in the Spark ecosystem. These methods can be used for natural language processing applications in various fields, including finance and healthcare.</p>

</div>

""", unsafe_allow_html=True)

# References
st.markdown('<div class="sub-title">References</div>', unsafe_allow_html=True)
st.markdown("""

<div class="section">

    <ul>

        <li><a class="link" href="https://sparknlp.org/api/python/reference/autosummary/sparknlp/annotator/ner/ner_dl/index.html" target="_blank" rel="noopener">NerDLModel</a> annotator documentation</li>

        <li>Model Used: <a class="link" href="https://sparknlp.org/2021/07/19/ner_wikiner_glove_840B_300_xx.html" target="_blank" rel="noopener">ner_wikiner_glove_840B_300</a></li>

        <li><a class="link" href="https://nlp.johnsnowlabs.com/recognize_entitie" target="_blank" rel="noopener">Visualization demos for NER in Spark NLP</a></li>

        <li><a class="link" href="https://www.johnsnowlabs.com/named-entity-recognition-ner-with-bert-in-spark-nlp/">Named Entity Recognition (NER) with BERT in Spark NLP</a></li>

    </ul>

</div>

""", unsafe_allow_html=True)

# Community & Support
st.markdown('<div class="sub-title">Community & Support</div>', unsafe_allow_html=True)
st.markdown("""

<div class="section">

    <ul>

        <li><a class="link" href="https://sparknlp.org/" target="_blank">Official Website</a>: Documentation and examples</li>

        <li><a class="link" href="https://join.slack.com/t/spark-nlp/shared_invite/zt-198dipu77-L3UWNe_AJf4Rqb3DaMb-7A" target="_blank">Slack Community</a>: Connect with other Spark NLP users</li>

        <li><a class="link" href="https://github.com/JohnSnowLabs/spark-nlp" target="_blank">GitHub Repository</a>: Source code and issue tracker</li>

    </ul>

</div>

""", unsafe_allow_html=True)