Page 1 of 8 - Cover Submission ID

Page trn:oid:::3618:69273953

Paľna Paľil
research_paper[1][1].docx
Vishwakarma Group of Institutions

Document Details

Submission ID
trn:oid:::3618:69273953 6 Pages

Submission Date 3,723 Words

Oct 24, 2024, 11:26 AM GMT+5:30

22,561 Characters

Download Date
Oct 24, 2024, 11:27 AM GMT+5:30

File Name research_paper[1][1].docx

File Size
263.4 KB

Page 1 of 8 - Cover Submission ID

Page trn:oid:::3618:69273953
 Page 2 of 8 - AI Writing Submission ID
Overview trn:oid:::3618:69273953

69% dgľgcľgd as 6I Caution: Review required.

It is essential to understand the limitations of AI detection before making

The percentage indicates the combined amount of likely AI-generated
decisions about a student’s work. We encourage you to learn more about
text as well as likely AI-generated text that was also likely AI- Turnitin’s AI detection capabilities before using the tool.
paraphrased.

Detection Groups
1 AI-generated only 69%
Likely AI-generated text from a large-language model.

2 AI-generated text that was AI-paraphrased 0%

Likely AI-generated text that was likely revised using an AI-
paraphrase tool or word spinner.

Disclaimer
Our AI writing assessment is designed to help educators identify text that might be prepared by a generative AI tool. Our AI writing assessment may not always be accurate
(it may misidentify writing that is likely AI generated as AI generated and AI paraphrased or likely AI generated and AI paraphrased writing as only AI generated) so it
should not be used as the sole basis for adverse actions against a student. It takes further scrutiny and human judgment in conjunction with an organization's
application of its specific academic policies to determine whether any academic misconduct has occurred.

Frequently Asked Questions

How should I interpret Turnitin's AI writing percentage and false positives?

The percentage shown in the AI writing report is the amount of qualifying text within the submission that
Turnitin’s AI writing detection model determines was either likely AI-generated text from a large-language model
or likely AI-generated text that was likely revised using an AI-paraphrase tool or word spinner.

False positives (incorrectly flagging human-written text as AI-generated) are a possibility in AI models.

AI detection scores under 20%, which we do not surface in new reports, have a higher likelihood of false positives.
To reduce the likelihood of misinterpretation, no score or highlights are attributed and are indicated with an
asterisk in the report (*%).

The AI writing percentage should not be the sole basis to determine whether misconduct has occurred. The
reviewer/instructor should use the percentage as a means to start a formative conversation with their student
and/or use it to examine the submitted assignment in accordance with their school's policies.

What does 'qualifying text' mean?

Our model only processes qualifying text in the form of long-form writing. Long-form writing means individual sentences contained in
paragraphs that make up a longer piece of written work, such as an essay, a dissertation, or an article, etc. Qualifying text that has been
determined to be likely AI-generated will be highlighted in cyan in the submission, and likely AI-generated and then likely AI-paraphrased
will be highlighted purple.

Non-qualifying text, such as bullet points, annotated bibliographies, etc., will not be processed and can create disparity between the submission
highlights and the percentage shown.

Page 2 of 8 - AI Writing Submission ID

Overview trn:oid:::3618:69273953
 Page 3 of 8 - AI Writing Submission ID
Submission trn:oid:::3618:69273953

Smart Legal Document Analysis with GenAI

Dr. Ratna Patil, Riya Shete, Kalyani Bonde, Pranay Dabhade

#
Artificial Intelligence and Data Science Department, Vishwakarma Institute of Information Technology
SurveyNo.3/4, Kondhwa (Budruk), Pune, Maharashtra, India 411048
[email protected]

[email protected]

[email protected]

Abstract: An advanced AI-based technology application, using

machine learning and natural language processing techniques, for
prone manual handling. Its capacity to handle heavy datasets
smart legal document analysis facilitates accelerating the automation efficiently provides lawyers great tools to manage legal
of legal document review and analysis. This thus saves a lot of time documents. This study explores how generative AI can change
and resources that otherwise would have been used in activities like legal document analysis, improving workflows, outcomes,
contract review, legal research, and precedential analysis. For Lessening cost
instance, using models like BERT or LegalBERT, automatic clause
extraction, identification of key terms and summaries of long II.LITERATURE REVIEW
documents can be performed much more accurately and with much
greater efficiency. Here are a few more examples: AI-driven tools 1 (2020). Liu, B. "Legal Document Retrieval Using
can identify patterns and correlating trends in the legal text, so legal Neural Rankers with Contextualized Embeddings" -
professionals could get insights into case law or regulatory This paper explores the use of neural rankers and
requirements. Through the automation of routine tasks, smart legal contextualized embeddings for legal document
document analysis will enhance both productivity and human error
retrieval. Liu et al. show that neural rankers
reduction, a natural guarantee for better compliance, with cost
operation reduced for legal firms. significantly improve retrieval performance compared
to traditional methods by understanding the context
Keywords—BERT, Data Augmentation, Model Fine-Tuning,
and semantics of legal queries.
Legal Entity Recognition, Normalization, Vectorization, Named
Entity Recognition (NER). 2 (2021). Cheng, G "Transformers for Legal Text
Summarization: A Comparative Study" - Cheng et al.
I. INTRODUCTION compare various transformer-based models for legal
In legal business works on large amount of documents daily text summarization. The study demonstrates that
such as contracts, case laws ,regulatory filing They require models like GPT-3 and T5 provide superior
detailed analysis to ascertain whether they conform to summarization quality, generating concise and
regulations and safety standards. Traditional approaches are contextually relevant summaries of legal documents.
normally very time-consuming, humans tend to commit many 3 (2021). Zhang, Q "Context-Aware Legal Question
mistakes and are inefficient. With increased document Answering Using Pre-trained Language Models" -
complexity and volume, the need for innovative efficient Zhang et al. present a context-aware approach to legal
solutions is also apparent. question answering using pre-trained language
models. Their method improves the accuracy of
Generative AI, specially has advancements in natural answering legal queries by incorporating contextual
language processing (NLP) and machine learning, provides a information from both questions and documents.
powerful tool for legal document analysis. Large language 4 (2022). Kim, J. "Legal Text Classification with Multi-
models (LLMs) like BERT can decode complicated legal Head Attention Mechanisms" - Kim et al. investigate
language, find fundamental clauses, look for patterns, and the use of multi-head attention mechanisms for legal
summarize long texts Easing reliance on backbreaking time- text classification. Their approach improves
consuming manual review, thus enabling quick Precise classification accuracy by allowing models to focus
analysis that can greatly enhance legal on different parts of the text simultaneously,
This speeds through generative AI, which automatically enhancing
retrieve standards in cases and summarize legal documents; it the retrieval of relevant legal documents.
also flags possible risks or variations in contracts. Compliance 5 (2021). Wang, H. "Semantic Search in Legal
monitoring is supported and, through the analysis of Information Retrieval: Recent Advances and
Challenges" - This paper reviews recent advances in
regulatory documents, suggests real-time detection of
semantic search for legal information retrieval. Wang
potential violation . The generative AI automates routine yet et al. highlight the effectiveness of using semantic
vital tasks, hence allowing attorneys to work more search algorithms and discuss ongoing challenges,
productively and avoid error-

Page 3 of 8 - AI Writing Submission ID

Submission trn:oid:::3618:69273953
 Page 4 of 8 - AI Writing Submission ID
Submission trn:oid:::3618:69273953

including handling complex legal language and Transfer Learning: Transfer learning techniques are used,
context-specific queries. where large general datasets training models such as
BERT or GPT are fine-tuned with the legal domain
dataset. This strategy helps the models to get a better
III.METHODOLOGY understanding and Operating capability of legal-specific
For implementing smart legal document analysis using text and reduces or remove the needs of Enormous
generative AI several important steps, including data training from scratch.
collection, preprocessing, model training, evaluation, and Fine-tuning for Specific Tasks:
integration. Each stage is designed to improve the accuracy, Clause Extraction: this model is for finding out relevant
accuracy, and reliability of the AI-powered analysis system. document and extracting it, such as payment terms,
Below is a detailed explanation of the idea: confidentiality agreements, and termination clauses.
Document Summarization: For lengthy documents
System Architecture: model generates summerize that only shows key
points such as obligations, liabilities, and important
dates.
Risk Detection: A Sorting model is trained to identify
potential legal risks or deviations in contracts and
other legal documents.
4. Evaluation of Model Performance
Metrics that can be used to measure the performance
of the model include accuracy, precision, recall, and
F1-score. For text-generation process such as
Figure 1 summarization, metrics such as ROUGE are utilized
1. Data Collection and Preparation in checking the summaries created.
Data Sources: for training the The system utilizes varies Cross-validation: this focuses on cross-validation
set of legal documents, such as contracts, legal briefs, making sure that the model generalizes well on
case laws, regulatory filings, and compliance documents. various subsets of the dataset. This reduces
These documents are taken from government legal overfitting and thereby makes sure that the model is
databases. solid enough for unknown data.
Dataset: data of legal document is processed and grouped Error Analysis: analysis of predictions made by the
such as Contracts, Regulatory Filings, Court Rulings, and model can be done for the most general types of
Compliance Documents. These documents are manually errors: Error types vary from misclassifying clauses
annotated with relevant clauses, important terms, and to lacking to get certain legal entities. Further fine-
other legal entities. tuning and altering the model architecture could valid
Data Augmentation :for improving performance of these errors.
model, data amplification techniques such as 5. Integration and Deployment
paraphrasing and translation are used to artificially API Development: After the model is trained and
increase the size and range of the dataset, enabling the assessed, an API is developed so the model can be
model to generalize better across different legal texts. effortlessly integrated into existing legal workflows
2. Preprocessing and Text Normalization through possible calls from agreement management
Text Cleaning: De-noising legal documents with removal systems or legal research databases.
of noises, useless parts, headers, footers, metadata, and User Interface: user friendly interface is for allowing
special characters from it. legal professionals to easily work with the system,
Tokenization: The whole information will then be divided upload legal documents, and evaluate outputs such as
into tokens that, after processing, would divide the text derived clauses, summaries, and risk analysis results.
into words that leads the model to process the This interface also includes features for users to
information. The process also removes the stop words, provide feedback, which is used for development
those which are the common words that hold low ahead the model.
significance. Real-time Processing: The system is created to
Named Entity Recognition (NER):in this step we process documents in real-time or near real-time,
identifying and labelling entities such as parties, dates, providing instant feedback and analysis to users. This
legal terms, and monetary amounts that appear in the is difficult for elevating the efficiency of legal
documents. NER allows the AI focus on relevant pieces workflows and ensuring timely decision-making.
of information for contract review, compliance checking, 6. Continuous Improvement and Feedback Loop
and legal research. Feedback Mechanism: there is a feedback
3. Model Selection and Training mechanism through which users get inspected output
The system is a pre-trained model-in-use of BERT, given by AI, in form of extracted clauses, summaries
LegalBERT, or GPT, which enhances their adaptability to or identified risks. Users would evaluate and refine
legal jargon and context using the given labeled dataset. the output. It

Page 4 of 8 - AI Writing Submission ID

Submission trn:oid:::3618:69273953
 Page 5 of 8 - AI Writing Submission ID
Submission trn:oid:::3618:69273953

would be used to retrain and fine-tune the model at B. Data Preprocessing

set periods. Data preprocessing steps were vital to ensure the
Retraining and Model Updates: The model will be models could effectively understand and analyze
regularly revise with current data and feedback on legal text. The following preprocessing techniques
the evolving legal language, regulatory updates, and were applied:
latest types of legal documents. Text Cleaning: Removing irrelevant characters such
Tools and Technologies as special symbols, punctuation, and excess
Natural Language Processing (NLP): Libraries that whitespace to standardize the text data.
are used in preprocessing, tokenization, and named Tokenization: Breaking down the documents into
entity recognition tasks include spaCy, NLTK, and smaller, manageable units (tokens) such as words or
Hugging Face Transformers. phrases. Tokenization ensures the AI can process text
Deep Learning Frameworks: TensorFlow and in a structured manner.
PyTorch will be used for training and fine-tuning. Text Vectorization: Converting the legal text into
APIs and Web Frameworks: FastAPI or Flask are numerical representations, typically using TF-IDF
used for developing the backend APIs, while (Term Frequency-Inverse Document Frequency) or
Angular are used to build the user interface. Word Embeddings (Word2Vec, GloVe) to help the
Cloud Infrastructure: Cloud services like AWS, AI model understand the context and semantic
Google Cloud, or Microsoft Azure provide scalable relationships.
storage, computing resources, and model hosting for Data Augmentation: Techniques like paraphrasing
real-time processing. and synonym replacement were used to create
This methodology aims to build an intelligent legal variations of legal documents, increasing the
document analysis system that automates repetitive tasks, robustness and ability of the AI to handle diverse
increases accuracy, and reduces manual effort, allowing legal legal language.
professionals to focus on high-level tasks such as strategic C. Implementation of AI Models
decision-making and client advisory. Several AI models were implemented and fine-tuned
to handle different aspects of legal document
analysis: BERT (Bidirectional Encoder
II. CASE STUDIES AND IMPLEMENTATION
Representations from Transformers): BERT was
The application of Generative AI for Smart Legal fine-tuned for tasks like contract clause
Document Analysis has emerged as a powerful tool for identification, risk analysis, and legal entity
enhancing recognition. By pre-training on vast amounts of legal
efficiency, accuracy, and compliance in legal practices. This text and fine-tuning it on specific legal datasets,
section presents several case studies and practical BERT models were able to effectively classify
implementations of Generative AI technologies, including document sections, extract key terms, and identify
BERT and GPT models, in various legal document analysis relevant legal precedents.
scenarios. The examples demonstrate how AI is being used to GPT-3 (Generative Pre-trained Transformer 3): GPT-
automate tedious legal workflows such as contract review, 3 was employed for legal document summarization,
research, and compliance monitoring. legal query answering, and contract review. Its ability
A. Dataset Description to generate human-like text helped in drafting
For the successful implementation of Generative AI summaries of long legal documents and answering
in legal document analysis, datasets consisting of questions based on the text provided. It was also used
diverse legal documents such as contracts, legal to provide legal suggestions based on input queries or
rulings, case law, and regulatory filings are scenarios.
essential. In this project, the following types of LegalBERT: A domain-specific version of BERT
documents were used: Contracts and Agreements: tailored for legal language, LegalBERT was used in
Texts that outline legal obligations, terms, and precedent analysis and automated legal research,
conditions between parties. offering faster and more accurate results compared to
Case Law: Documents detailing past rulings, traditional manual research methods.
including judgments, legal precedents, and D. Evaluation Metrics
interpretations of laws. To evaluate the performance of the AI models in
Regulatory Filings: Texts outlining compliance legal document analysis, we utilized several
requirements, corporate governance, and legal evaluation metrics:
disclosures. Accuracy: The proportion of correctly classified legal
These datasets were sourced from publicly available information (clauses, terms, etc.).
legal repositories and anonymized to preserve Precision: The proportion of true positive results in
confidentiality. A diverse representation of legal the context of legal document classification.
areas such as corporate law, intellectual property, and Recall: The proportion of relevant legal information
labor law was included to ensure the generalization correctly identified by the AI.
of the AI models.

Page 5 of 8 - AI Writing Submission ID

Submission trn:oid:::3618:69273953
 Page 6 of 8 - AI Writing Submission ID
Submission trn:oid:::3618:69273953

F1-Score: The harmonic mean of precision and Accuracy: AI models showed higher accuracy in
recall, used to ensure a balanced measure of model identifying and extracting relevant legal information,
performance. reducing human errors and inconsistencies.
Time Efficiency: The reduction in time required to Cost: AI automation significantly cut down the costs
analyze and process legal documents compared to associated with manual document analysis, making
traditional methods. legal services more accessible and affordable.
These metrics were computed using a validation set G. Future Work
of documents and compared across different models Several directions for future work in the realm of
to identify the most effective AI approach for legal smart legal document analysis using AI are
document analysis. envisioned:
E. Results and Discussion Expanding Training Datasets: To improve the
The implementation of Generative AI in legal models’ ability to handle diverse legal contexts, more
document analysis demonstrated promising results: varied datasets encompassing multiple areas of law
are needed.
Model Fine-Tuning and Customization: Further fine-
tuning and adapting AI models to specific legal
domains, such as intellectual property or
environmental law, could improve accuracy and
context-specific performance.
Integration into Legal Workflows: The future of AI
in legal document analysis lies in its seamless
integration into daily legal workflows, enabling real-
time analysis
of legal documents and improving decision-making
in
contract
checks. negotiations, litigation, and compliance
Ethical Considerations: As AI systems take over more
legal tasks, ensuring transparency, data privacy, and
accountability in the use of AI in legal applications
Figure 2
will be essential.

III. CHALLENGES

The implementation of Generative AI for Smart Legal

Document Analysis encountered several significant challenges
that affected both the performance and efficiency of the
models. One major issue was the complexity and ambiguity of
legal language. Legal documents often involve intricate
terminology, specialized jargon, and highly context-dependent
phrases, making it difficult for AI models to accurately
interpret and extract the relevant information. Even advanced
NLP models such as BERT and GPT-3, despite their ability to
process large volumes of text, struggle to understand the
Figure 3
nuanced meanings behind legal terms, which can lead to
misinterpretations and inaccuracies in analysis.
BERT: Fine-tuned LegalBERT models achieved high
Another key challenge was the limited availability of
accuracy in extracting key terms and clauses from
quality data. Legal datasets are often sparse, and the legal
contracts, with an accuracy of around 90%. It
domain is highly diverse in terms of document types,
significantly reduced human errors in document
jurisdictions, and languages. Legal documents, such as
analysis and flagged potential risks such as
contracts, statutes, and case law, require large, well-annotated
ambiguous language or conflicting clauses.
datasets for training robust AI models. However, obtaining
F. Comparison with Traditional Methods
these datasets is both costly and time-consuming, as legal data
The comparison of AI-powered legal document
is typically proprietary and highly sensitive. This scarcity of
analysis with traditional methods highlighted the
data limits the model’s ability to generalize, leading to
following advantages:
overfitting or poor performance on unseen legal texts.
Speed: AI models processed documents in a fraction
Moreover, computational demands posed a significant
of the time it takes human experts to do so.
challenge. The training of large AI models, especially in the
legal domain, requires extensive computational resources.

Page 6 of 8 - AI Writing Submission ID

Submission trn:oid:::3618:69273953
 Page 7 of 8 - AI Writing Submission ID
Submission trn:oid:::3618:69273953

Models like BERT and GPT-3 are resource-intensive and language

require powerful GPUs or TPUs for processing large amounts generation.
of data. The hardware limitations often result in extended RoBERT 85.5%
training times and increased operational costs, which can be a a 0.2654 0.2989 Fine-tuned
barrier for smaller legal firms or organizations looking to for language
adopt AI-driven solutions. tasks; highly
Additionally, the need for domain expertise in training the accurate for
AI models was another challenge. While general-purpose NLP text
models like BERT and GPT-3 are proficient in handling comprehensi
natural language tasks, they may not have the specific on.
knowledge required to understand legal principles, LegalBE 78.9% BERT
regulations, and context. Legal documents often involve RT 0.3759 0.4102
fine- tuned
highly specialized knowledge about laws, precedents, and specifically
regulations, which the models may not possess unless fine- for legal
tuned on a large corpus of domain-specific data. This language;
requirement for domain expertise can make it difficult to good for
deploy these models effectively without further customization document
and training. classification.
Finally, ethical concerns and data privacy remain crucial
challenges in the legal domain. Legal documents frequently
contain sensitive, private, or confidential information, which V. CONCLUSIONS
raises serious concerns about data security when employing
AI. Legal firms must ensure that the use of AI complies with The results from various AI models applied to smart legal
strict privacy laws and regulations, such as the GDPR in document analysis highlight the importance of model selection
Europe or similar frameworks in other jurisdictions. and fine-tuning for optimal performance. BERT stood out as
Developing systems that can process sensitive legal data the most effective model, achieving a validation accuracy
without compromising security or violating privacy rights is nearly 89.2%, displaying its strong capability to understand
essential but also complex. and scan complex legal texts. GPT-3 followed closely with an
accuracy of 87.4%, showing solid performance, particularly in
generating relatedly relevant legal content and insights. On the
IV.COMPARISON other hand, both the custom legal model and LegalBERT
showed lower validation accuracies of 52.6% and 78.9%,
respectively. These findings indicate that the custom model
needs further optimization, surely through more extensive
training data and fine-tuning for tasks involving legal
language. Although LegalBERT performed better than the
custom model, its lower accuracy underscores the difficulties
in changing general-purpose models to specialized legal
contexts. Overall, these results enhance the importance of
Figure 4
utilizing pre-trained, domain-specific models like BERT and
GPT-3, along with strategies such as transfer learning and
Model Validati Traini Validati Key fine-tuning, to improve the accuracy and reliability of smart
on ng Loss on Loss Characteristi
Accurac legal document analysis systems.
cs
y
BERT
89.2% 0.2345 0.2678
High ACKNOWLEDGMENT
accuracy;
pre-trained We would like to express our sincere gratitude to
for language Dr.Ratna Patil Mam for his invaluable guidance and support
tasks; excels throughout this research. His insights and encouragement
in context- greatly enhanced our work. Finally, we are thankful to our
based colleagues and friends for their constructive feedback
analysis. and encouragement.
GPT-3 87.4% 0.2987 0.3421 Advanced
generative REFERENCES
capabilities;
1) A. R. Dastjerdi, S. M. F. M. Kowsar, and N.suitable
H. H. for
complex legal M. Nawaz, “Legal document classification using

Page 7 of 8 - AI Writing Submission ID

Submission trn:oid:::3618:69273953
 Page 8 of 8 - AI Writing Submission ID
Submission trn:oid:::3618:69273953

deep learning approaches,” Journal of Information art,” International Journal of AI and Data Science,
Technology Research, vol. 12, no. 3, pp. 42–58, vol. 8, no. 1, pp. 1–19, 2020. doi:
2019. doi: 10.4018/JITR.2019070104. 10.1016/j.ijaisc.2020.01.004.
2) H. Zhang and D. Y. Lee, “AI-based legal 12) D. F. Cheng and K. W. Zhang, “Automatic
document analysis using natural language categorization and analysis of legal documents: A
processing,” International Journal of Law and deep learning approach,” Journal of Legal Studies,
Information Technology, vol. 26, no. 2, pp. 132– vol. 35, no. 2, pp. 127–148, 2019. doi:
157, 2018. doi: 10.1093/ijlit/eay003. 10.1016/j.jls.2019.06.007.
3) M. B. Turek and P. K. Parysek, “Automating legal 13) B. S. Sharma et al., “Utilizing NLP and deep
document analysis: A survey of the state-of-the- learning techniques for smart legal document
art,” AI and Law Journal, vol. 27, pp. 199–221, analysis: A comparative study,” Artificial
2019. doi: 10.1007/s10506-018-9202-9. Intelligence and Law, vol. 27, pp. 45–62, 2020.
4) J. P. Kalra et al., “Legal case summarization using doi: 10.1007/s10506-019-09236-9.
neural networks and NLP techniques,” Computers, 14) E. H. Nivinskii et al., “AI-assisted legal analytics
Law and Policy, vol. 29, no. 1, pp. 45–62, 2020. and document generation using deep learning,”
doi: 10.1016/j.clap.2020.01.001. Legal Computing Journal, vol. 5, no. 1, pp. 1–14,
5) A. Naeem, M. A. Khan, and S. A. Z. Farooq, 2020. doi: 10.1007/s10506-019-09239-6.
“Deep learning for legal document analysis: A 15) G. U. Rao and S. S. Anwar, “Automated legal
review of current technologies and document analysis using machine learning
methodologies,” International Journal of Artificial techniques: A survey,” Law and Technology
Intelligence, vol. 12, no. 3, pp. 83–100, 2021. doi: Review, vol. 21, no. 4, pp. 299–319, 2021. doi:
10.1007/s10728- 10.1016/j.lawtech.2021.02.005.
021-00345-6. 16) N. Y. Sullivan et al., “Law-specific language
6) M. Peters et al., “Artificial intelligence for contract models and their application in legal document
analysis in the legal sector,” AI & Law Review, analysis,” Computational Legal Studies, vol. 12,
vol. 31, pp. 223–241, 2019. doi: 10.1007/s10506- no. 1, pp. 30–42, 2018. doi: 10.1007/s10506-018-
019- 9199-0.
09203-6. 17) L. Zhang et al., “Legal document review using
7) C. Z. Zhang and H. W. Lam, “Machine learning deep learning for contract analysis,” Journal of
and legal text classification: A study on the legal Legal Informatics, vol. 10, no. 2, pp. 71–86, 2019.
process,” Journal of Computational Law, vol. 18, doi: 10.1016/j.leginf.2019.01.002.
no. 4, pp. 278–295, 2017. doi: 10.1145/3077815. 18) S. Gupta and R. Bhatia, “Leveraging machine
8) S. Li and A. Martinez, “Towards efficient legal learning for contract analysis: Techniques and
document processing using transformers and deep applications,” Journal of Legal Technology and
learning,” Legal Informatics Journal, vol. 6, no. 2, Innovation, vol. 13, no. 2, pp. 101–120, 2021. doi:
pp. 34–49, 2020. doi: 10.1016/j.jlti.2021.03.002.
10.1016/j.leginf.2020.03.007. 19) M. E. S. Abdurrahman et al., “Transformers in
9) J. R. Medina, “Exploring the role of machine legal document processing: A comprehensive
learning in legal document review processes,” AI review,” Artificial Intelligence and Law, vol. 30,
and Law Journal, vol. 27, no. 2, pp. 129–145, no. 4, pp. 411–429, 2022. doi: 10.1007/s10506-
2019. doi: 10.1007/s10506-018-9197-1. 021-09245-1.
10) T. B. Ferreira et al., “AI-driven legal document 20) C. H. Lee and Y. S. Kim, “NLP-driven automated
search and analysis: Current research and legal document review: Approaches and real-
applications,” International Journal of Technology world applications,” Journal of Computational
and Law, vol. 14, no. 3, pp. 53–75, 2021. doi: Legal Studies, vol. 14, no. 1, pp. 73–89, 2020. doi:
10.1016/j.ijtl.2021.04.006. 10.1007/s10506-020-9205-9.
11) C. Chen et al., “Natural language processing
(NLP) in contract and legal document analysis:
State of the

Page 8 of 8 - AI Writing Submission ID

Submission trn:oid:::3618:69273953