Volume 10, Issue 5, May – 2025 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 [Link]

Knowledge Graph from Unstructure Data

G. Newton1; M. Shajith Revanth 2; P. Minish 3; K. Tharuni Reddy4
Assistant Professor1, Students 2,3,4
1,2,3,4
Department CSE, Nalla Narasimha Reddy Education Society's Group of Institutions, Hyderabad, India.

Publication Date: 2025/08/13

Abstract: This project presents a client-side, knowledge graph system that dynamically extracts and visualizes semantic
relationships from unstructured natural language input. Unlike traditional keyword-based methods, this system uses
lightweight Natural Language Processing (NLP) to interpret the contextual meaning of user queries. Unlike traditional
keyword-based methods, this system uses lightweight Natural Language Processing (NLP) to interpret the contextual
meaning of user queries. It identifies key entities and their relationships through in-browser logic and parsing, transforming
them into nodes and edges rendered instantly as a knowledge graph. Built with React, TypeScript (TSX), and ReactFlow,
the interface offers an intuitive experience for exploring semantic structures without relying on any backend or external
database. This fully browser-based architecture ensures fast, private, and responsive interaction. The system is well- suited
for applications such as semantic search, concept discovery, educational tools, and interactive data exploration—enabling
users to better understand and navigate the relationships embedded in text.

Keywords: Knowledge Graph, NLP, Semantic Parsing, Entity Extraction, Relationship Mapping, React, Typescript, Client-Side
Processing, Graph Visualization, Unstructured Text, Dynamic UI, Contextual Analysis, Backend-Free Architecture.

How to Cite: G. Newton; M. Shajith Revanth; P. Minish; K. Tharuni Reddy (2025) Knowledge Graph From Unstructure Data.
International Journal of Innovative Science and Research Technology, 10(5), 4821-4824.
[Link]

I. INTRODUCTION retrieval, making it a valuable tool for navigating and

understandinglarge-scaletextualdatasets.
In the era of rapidly expanding digital content,
organizing large volumes of unstructured text data has II. LITERATURE REVIEW
become a significant challenge. Traditional document
clustering techniques, which primarily rely on keyword Traditional methods for analyzing text have mostly
frequency and syntactic patterns, often fall short in capturing focused on basic techniques like keyword frequency and
the true semantic relationships between documents. This pattern matching. These approaches often rely on counting
project introduces a semantic-based document clustering words or identifying specific structures, but they fall short
system that addresses these limitations by leveraging advanced when it comes to understanding the actual meaning behind
Natural Language Processing (NLP) techniques. By utilizing the text. To address these shortcomings, more recent
pre-trained language models such as BERT and Sentence- approaches have explored the idea of understanding the
BERT, the system transforms each document into a dense meaning and context of text—moving beyond simple pattern
vector representation that encapsulates its recognition. This shift has led to the development of systems
that aim to extract entities and understand their relationships
Contextual and semantic meaning. These more like a human would, rather than just based on word
embeddings serve as the foundation for accurate position or frequency. In the context of this project, the focus
comparison and grouping of documents based on their is on building a lightweight, browser-based knowledge graph
intrinsic content, rather than mere word occurrence. To system that can interpret natural language input without
effectively cluster these semantic vectors, the system employs relying on a backend or external data services. Instead of
robust unsupervised learning algorithms like K-Means and using heavy machine learning models, this system applies
DBSCAN, along with cosine similarity to measure the custom logic and in- browser NLP to extract key elements like
closeness between document vectors. The result is a set of people, objects, actions, and their interconnections. Nodes
coherent and contextually meaningful clusters that represent entities, while edges show the semantic relationships
outperform traditional keyword-based approaches in between them. This approach allows users to see how
terms of relevance and interpretability. This semantic concepts are related in real time, giving a clearer
clustering framework has broad applications in content understanding of the text’s meaning through interactive visual
organization, topic discovery, and intelligent information representation. Compared to traditional system where key

IJISRT25MAY2182 [Link] 4821

Volume 10, Issue 5, May – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 [Link]
words are used in the knowledge graph system we can observe  Server-Dependent Knowledge Graph Tools
a way to explore and organize information.. It opens up new A few systems allow users to build knowledge graphs
possibilities for semantic search, idea mapping, and from text, but they depend on backend services or databases to
intelligent information retrieval—all handled completely process and store data. This setup can be overkill for
within the user's browser. lightweight or personal use, and it makes the system harder to
maintain or deploy independently.
III. METHODOLOGY
V. ARCHITECTURE
The methodology of the knowledge graph deduction
system is structured around a set of client-side processes that
extract and visualize semantic relationships from unstructured
natural language input in real time. The process begins with
user input, where the system accepts free-form natural
language text through a browser-based interface. Instead of
relying on external APIs or server-side models, the text is
parsed and processed directly in the browser using
lightweight NLP techniques. This includes basic text
preprocessing steps such as lowercasing, punctuation
removal, and tokenization, followed by custom logic to
identify important entities .Once the relevant entities and
relationships are identified, they are transformed into
structured graph components. It all starts when the user types
or pastes some text into the system. The first step is to clean
it up a bit—things like removing punctuation, converting
everything to lowercase, and breaking the text into smaller
pieces so it’s easier to work with. This structured data is then
passed to the ReactFlow library, which dynamically renders
the knowledge graph on the screen using an intuitive and
interactive visual layout. The graph layout adapts in real time,
allowing users to explore the connections and meanings
embedded in their input. For example, in a sentence like “The
doctor treats the patient,” it would recognize “doctor” and
“patient” as important ideas and “treats” as the connection
between them. The methodology supports applications such
as semantic search, concept mapping, interactive education,
and exploratory data analysis—making it easier to understand
and navigate the relationships within natural language
content.

IV. EXISTING SYSTEM

 Pre-Trained Model-Based Clustering

Some platforms use powerful language models to
convert text into numbers that represent its meaning. These Graph 1 Knowledge Graph from Unstructured Data
"embeddings" can then be grouped using clustering methods. Architecture
It’s useful for understanding which pieces of text are similar
in meaning, even if the words are different. A user begins by inputting free-form text into the system
through a simple, browser-based interface. After this, the
 Cloud NLP Platforms system uses in-browser logic to analyze the structure of the
Some APIs help extract useful details from text, like sentences and identify important elements like subjects,
important keywords or the general tone. While they don’t objects, and verbs.
create clusters themselves, the extracted data can be used later
to build connections between pieces of information. The system then examines sentence structure to extract
core entities and the relationships between them, such as
 Heavy Cloud-Based APIS actions, links, or contextual connections. These extracted
Some platforms use advanced AI models hosted in the elements are translated into a graph structure, where each
cloud to analyze text, extract entities, and find relationships. entity becomes a node and each relationship forms a
While powerful, these tools require internet connectivity, user connecting edge. This structure forms the foundation of a
data sharing, and can be slow for real-time interaction. They knowledge graph, which is immediately visualized using
also limit customization and control. React Flow in a dynamic and interactive interface.

IJISRT25MAY2182 [Link] 4822

Volume 10, Issue 5, May – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 [Link]
Users can explore this graph in real-time, gaining a embeddings enable the system to represent complex textual
clearer view of how different ideas are connected. The entire information accurately as nodes (entities) and edges
process runs without any backend, making it lightweight and (relationships) within the knowledge graph. However,
accessible while still offering rich insights into the input text. running BERT requires substantial computational power, so
the system optimizes usage by selectively processing relevant
VI. DATAFLOW text portions and leveraging browser-efficient
implementations.

 DBSCAN
DBSCAN is a density-based clustering algorithm that
identifies clusters of semantically related entities and
relationships within the embedding space. In the knowledge
graph project, DBSCAN is used to cluster embeddings
generated by BERT to discover natural groupings of related
concepts without prior assumptions about cluster counts. This
flexibility allows the system to reveal both small and large
clusters of closely related information, even when clusters
have irregular shapes. Additionally, DBSCAN labels low-
density points as noise, effectively filtering out outliers and
irrelevant data that might otherwise clutter the knowledge
graph. This leads to clearer, more meaningful clusters and
ultimately a more coherent and insightful graph
representation.

 K-Means Clustering:
K-Means is a widely-used clustering algorithm that
partitions data into a pre-defined number of clusters by
iteratively assigning points to the nearest cluster center
(centroid) and recalculating centroids based on current
members. In the context of the knowledge graph project, K-
Means is applied to the dense embeddings obtained from
BERT to group semantically similar entities or concepts
Fig 2 Dataflow together. This method works well when the number of
clusters is known or can be estimated, allowing the system to
The process starts with a user who provides raw text or organize knowledge graph elements into meaningful
documents as input. These texts are temporarily held in a Text categories or themes.
Repository, serving as the initial data source. The input is
then processed by a Natural Language Processing (NLP) Additional Notes on Dimensionality Reduction and
module that uses lightweight language parsing and custom Visualization (Optional): While the core algorithms are
logic to identify important entities and relationships within BERT, DBSCAN, and K-Means, some implementations may
the text. This module extracts semantic information and include dimensionality reduction techniques such as t-SNE or
transforms it into structured data elements. UMAP to visualize high-dimensional embeddings in 2D or
3D space, making clusters more interpretable. Combined with
The structured semantic data is then used to generate an frontend tools like ReactFlow, the reduced embeddings
interactive knowledge graph, where entities become nodes enable dynamic, real-time visualization of the knowledge
and their connections form edges. This graph is immediately graph, where users can explore entities, their connections, and
visualized through a user-friendly interface, allowing users to overall graph structure in a user- friendly manner.
interact with and explore the complex web of information
intuitively and effectively. VIII. CONCLUSION

VII. ALGORITHAMS The primary aim of this project was to develop an

intelligent knowledge graph system that can automatically
 BERT extract, represent, and visualize the semantic relationships
BERT (Bidirectional Encoder Representations from between entities within unstructured text data. Unlike
Transformers) is a powerful language model developed by conventional systems that depend heavily on keyword
Google that has transformed how computers understand matching or simple frequency-based methods, this project
human language. This bidirectional processing enables successfully leverages advanced natural language processing
BERT to understand the full context of each word within a techniques directly in the browser to understand and connect
sentence, capturing subtle nuances and relationships It key concepts in a more meaningful way.
transforms raw text into dense vector embeddings, which
encapsulate semantic and contextual information. These

IJISRT25MAY2182 [Link] 4823

Volume 10, Issue 5, May – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 [Link]
It provides a scalable, browser-based solution that [10]. T. Wang and L. Zhang, "Enhancing document
supports a variety of applications, including semantic search, clustering performance with contextual embeddings
concept linking, educational tools, and intelligent data and density-based algorithms," International Journal
analysis. The success of this knowledge graph framework of Data Science, vol. 8, no. 1, pp. 30–42, 2022.
lays the groundwork for future enhancements such as
incorporating more sophisticated NLP models, expanding
entity recognition capabilities, and enabling integration with
external knowledge bases to enrich the semantic context
further.

REFERENCES

[1]. M. Reimers, J. Dodge, J. Gilmer, M. D. Hoffman, and

M. Dredze, "Sentence-level representations for
document classification," Proceedings of the 2019
Conference of the North American Chapter of the
Association for Computational Linguistics,
Minneapolis, USA, 2019, pp. 700–707, doi:
10.18653/v1/N19-1070.
[2]. J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova,
"BERT: Pre-training of Deep Bidirectional
Transformers for Language Understanding,"
Proceedings of the 2019 Conference of the North
American Chapter of the Association for
Computational Linguistics: Human Language
Technologies, Minneapolis, 2019, pp. 4171–4186,
doi: 10.48550/arXiv.1810.04805.
[3]. N. Reimers and I. Gurevych, "Sentence-BERT:
Sentence Embeddings using Siamese BERT-
Networks," Proceedings of the 2019 Conference on
Empirical Methods in Natural Language Processing
and the 9th International Joint Conference on Natural
Language Processing, Hong Kong, 2019, pp. 3982–
3992, doi: 10.48550/arXiv.1908.10084.
[4]. C. C. Aggarwal and C. X. Zhai, "A Survey of Text
Clustering Algorithms," in Mining Text Data, Boston,
MA: Springer, 2012, pp. 77–128, doi: 10.1007/978-1-
4614-3223-4_4.
[5]. M. Ester, H.-P. Kriegel, J. Sander, and X. Xu, "A
density-based algorithm for discovering clusters in
large spatial databases with noise," Proceedings of the
Second International Conference on Knowledge
Discovery and Data Mining (KDD-96), Portland,
1996, pp. 226– 231.
[6]. J. Han, J. Pei, and M. Kamdar, Data Mining: Concepts
and Techniques, 4th ed., Elsevier, 2022, pp. 493–508,
ISBN: 978-0-12-818148-7.
[7]. J. Liu, X. Shen, W. Pan, and B. Liu, "Document
clustering via topic modeling using BERT
embeddings," Proceedings of the 2020 International
Conference on Artificial Intelligence and Computer
Engineering (ICAICE), Beijing, 2020, pp. 188–192,
doi: 10.1109/ICAICE51518.2020.00047.
[8]. W. X. Zhao, Y. Guo, and Y. He, "A Comparative
Study of Deep Learning Models for Semantic
Document Clustering," Information Sciences, vol.
576, pp. 55–72, 2021, doi: 10.1016/[Link].2021.07.055.
[9]. S. K. Singh and R. Sharma, "Semantic clustering using
transformer embeddings for document organization,"
Journal of Machine Learning and Data Mining, vol.
10, no. 3, pp. 145–160, 2023.

IJISRT25MAY2182 [Link] 4824