Knowledge Graph and Similarity Based Retrieval Method for Query Answering System

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 10 | Oct 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 464
Knowledge Graph and Similarity Based Retrieval Method for Query
Answering System
Aaditi Narkhede 1, Prof. M.U.Kulkarni2, Prof. S.R.Naik 3
1 Student, aaditirajeshn99@gmail.com, Dept. Of Computer Science and Technology, VJTI, Mumbai, India
2 Professor, mukulkarni@ce.vjti.ac.in, Dept. Of Computer Science and Technology, VJTI, Mumbai, India
3 Professor, sraksha@it.vjti.ac.in, Dept. Of Computer Science and Technology, VJTI, Mumbai, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - A semantic network that displays the connections
between entities is known as a knowledge graph. Data can be
visualised to help with information analysis and
comprehension through the use of the knowledge graph.
Knowledge graphs can help professionals with complex
analysis applications and decision support while also
erecting barriers in the market. A good construction system
can help companies construct knowledge graphs efficiently
and quickly. However, Extractionofvaluable informationfrom
such huge, complex and unstructured data requires a human
approach to handle user queries related to data, which causes
delays and uncertainty in decision making and strategy
planning. In this paper, we propose an approach for au-
tomatic knowledge graph construction and automated
querying engine to answer user queries using generated
knowledge graph. The proposed system will benefit
Professionals with faster, easierunderstandingandanalysis of
complex and huge unstructured data. Experimental results
show that our proposed solution is more effective in
constructing a generic knowledge graph.
Key Words: Knowledge graph, Question Answering
system, spaCy, Natural Language Processing, Named
Entity Recognition.
1. INTRODUCTION
The goal of a knowledge graph is usually to collect,
connect, and show information. It offers a high level of inter-
pretability. Knowledge Graphs help in establishing
purposefulrelationshipsoforganisationalknowledgethrough
classifying various content into different categories [1]. By
grouping different types of content into distinct categories,
knowledge graphs assist in establishing intentional
relationships between organisational knowledge. Large
volumes of unstructured data are produced daily. This
information is presentedinreports,researchpapers,patents,
scholastic articles, book chapters, essays, and speeches,
among other formats. Identifying key patterns in vast
amounts of unstructured data is crucial in to- day’s
environment.Itischallengingtounderstandandevaluatethe
implications provided in an organization’s data since the
essential informationisdispersedacrossthelargevolumesof
data.
Due to the absence of boundaries between the items that
need to be retrieved, the target entities’ context dependency,
the variability in language patterns, and the limits of
statisticalapproaches,automaticinformationextractionfrom
such vast amounts of data is challenging. The fact that this
type of datais frequently available as unstructuredtextsorin
PDF format presents another challenge when trying to
extract information from it. As a result, either laborious
manual preprocessing is required or sophisticated ETL
(Extract, transform, load)systems are used to automatically
ingest data. To handle this challenge,requireddatai.e.textual
data will be first extracted from PDFs for our research work
usingFine Tuneddetectron2basedmodelandpytesseractocr
and will be stored in a text filewhich will be used further for
information extraction purposeswhile building Knowledge
Graph.
The proposed approach aims to resolve issues of ambigu-
ity, abbreviations and semantics of text while constructing a
knowledge graph, achieved with the use of spaCy based NER
for extracting the entity-pairs and relations from the data.
Based on triplets obtained while information extraction, the
Knowledge graph is constructed. In the proposed Question-
Answering system, for Query analysis a similar spaCy based
approach is used for entity-pair and relation extraction from
user query. In the Answer Extraction module, the
combination of approaches is used, such as information
retrieval based on the featureinformationofrelevantentities
in sentences and uses trained feature classifiers to sort the
candidate answers and obtain the solutions; along with
matching query triplets with knowledge graph database
using generic linguistics rules designed to obtain the
solutions.
1.1 Organization Of the Paper
Section 2. of the paper describes the previous and current
study being carried out in the field of Knowledge Graphs
and Question-Answering Systems. It also states the
drawbacksand the problems faced in existing approaches.
Section 3.of the Paper explains the proposed methodology
and the step by step execution of the same with the help of a
few examples. Section 4. of the Paper summarizes and
analyses the Results obtained and gives an insight regarding
how the proposed system can be deployed for Different

domain-data. Section 5. of the Paper concludes the Research
regarding the generic approach for Knowledge graph
Construction and query answering system. Section 6. of the
Paper summarizes the References used/followed for the
survey purpose.
2. RELATED WORK
Our work is inspired by two threads of research: Key
techniques in KG construction and Key techniques in QA
system.
2.1 Key Techniques in KG Construction
A typical knowledge-graph construction process consists
of three main components: information extraction,
knowledge fusion, and knowledge graph building.
Information Extraction : The goal of information
extraction is to find and separate entities in a data source, as
well as their qualities and relationships with other entities
with the means of entity recognition and relation extraction.
Entity extraction, also known as named entity recognition
(NER), refers to the process of identifying accurately named
entities from data, especially text data [7]. There are certain
paradigms that have been developed for NER activities such
as Rule based, Machine learning based and Deep learning
based approaches. The underlying concept behind most
common NER approaches was to create a small set of rules
by hand and then look for strings in the text that fit these
criteria. Usage of Rule based NER can not handle multiple
unforeseen patterns and thus does not possess strong
interpretability of semantics while information extraction
[2]. Traditional NER approaches, such as rule-based and
template-based, are expensive and rely too much on
procedures like rule development and feature engineering.
The machine learning based approach, tags the named
entities to the words even when the words are not listed in
the dictionary and the context is not described in the rule
set. K. Khadilkar, S. Kulkarni and S. Venkatraman [1],
proposed a method for summarizing multilingual vocal as
well as written paragraphs and speeches, using semantic
Knowledge Graphs. For the selection of relation extraction
API, experiments were performed on NLTK and Stanford
CoreNLP; out of which Stanford CoreNLP resulted with
highest accuracy of 92.23% which uses CRF algorithm
internally having ability to consider context of previous text
in detecting entities and relation; but fails to consider
context level dependency in relation extraction. Different
from traditional machine learning, deep learning can
automatically extract high-level abstract features from a
large amount of data to perform model training [4]. In
particular, word vector representation has provided a
powerful driving force for the typical serialized labelling
problems of NER. Z. Dai, X. Wang has discussed various
neural network models [3], where the BERT-BiLSTM-CRF
model achieved higher F1 score compared to other models.
BERT, captures a general language representation from
large-scale corpora, but lacks domain-specific knowledge.
Two approaches such as SpaCy and BERT are discussed in
[6] for NER task for Tourism dataset where experimental
results shows that SpaCy NER outperforms BERT NER as
achieved accuracies are 95% and 70% respectively.
Performance of BERT is lowered as it is unable to tokenize
special multi word names which are proper nounsvery well.
Syntactic Dependency Parsing is also offered by SpaCy’s
dependency parser [6] for relation extraction.
Relation Extraction is a major field in NLP. Using Relation
Extraction, data can be transformed into a 3-tuple format of
[entity, relationship, entity]. The patterns are generated
through text analysis and represent the unique language
constructions which are used to describe a particular
Entity/Relation in [2]. These patternsarethenmatchedwith
processed text to discover and extract required pieces of
information; extracted relationship names between two
entities using object property of an ontological concept.
Recognizing the relationships using the spaCymodel maybe
done in two ways. The first method is to utilise spaCy’s
dependency parser. The second method istocreate newtags
based on relation keywords. [6] showed comparison
between spaCy’s dependency parser and BERT with BIO
tagging where spaCy achieved higher accuracy of 95% and
also it was able to consider context level dependency in
relation extraction.
Knowledge Fusion : In Knowledge fusion, last step of
knowledge graph construction, Entity Disambiguation and
entity linking is carried out to preserve the semantic
information of text and handles uneven knowledge
expression. During the Entity Disambiguation job,
ambiguous entity mentions are connected to their referent
entities in the Knowledge graph, which is accomplished by
employing a clustering technique in [9]. A method based on
Word2Vec cosine similarity calculation is proposed in [8] to
complete entity linking for entities obtained from question
sentences. Word2Vec is used in [8] to convert recognition
entity and candidate entity into corresponding wordvectors
and implemented cosine similarity method to calculate
similarity value between recognition entity and candidate
entity. For the Question-answering system in[11],relevance
scoring based on QA context is used to link and find relevant
entities from KG. To link the primary entities with their
preposition occurrences in the text or their multiple
reference entities, coreference resolution is used [1]. For
topmost winning models described in [7], Coreference
resolution is achieved using NeuralCoref offered by spaCy.
Rather thanadoptingcomprehensivecoreferenceresolution,
[12] tackle the problem by recognising the pronouns in the
input content and replacing them with the corresponding
subject or object. further, accuracy of coreferenceresolution
is improved by analysing the gender of an entity. For the

coreference resolution task, it has been observed that
NeuralCoref by spaCy does the job in an efficient way.
2.2 Key Techniques in QA System
Question-answering research based on Knowledge graphs
is an important area of study. The basis of the knowledge
graph’s question answering mechanism isquestionanalysis.
The accuracy with which the semantic information in the
question sentence is mined impacts the actual effect of the
question answering system. Research based on financial
knowledge graph using ontology proposed Question
Answering system [2] to sequentially perform linguistic
analysis of query, do named entity extraction, entity / graph
search, fusion and ranking of possible answers.SPARQL is
used for querying.
To solve the problem of inadequate question’s semantic
information mining, [8] proposedBERTbasedKGQAsystem
which mainly includes Entity recognition and Relation
Recognition for question analysis where BERT-BiLSTM-CRF
model is used and calculates the similarity between the
entity obtained in the entity recognition and candidate
entity, and takes the entity with highest similarity as the
standard entity contained in the natural language question.
The mainstream implementation methods of a knowledge
based question answering system can be divided into
categories such as Semantic Parsing which involves some
linguistics and traditional NLP methods, requires many
manual design rules, and has high accuracy but lacks
generalization ability; and Information Extraction:Thiskind
of approach extracts the feature information of relevant
entities in sentences or knowledge bases and uses trained
feature classifiers to sort the candidate answers and obtain
the solutions. It is closely related to the traditional NLP
method and feature engineering, with strong generalization
ability but relatively weak accuracy. The combination of
these techniques is used in [9]. The key entities and
relationships are derived from the user’s natural language
questions, which are categorised and evaluated. Finally,
Cypher language is built to query the knowledge graph via
intention prediction. It used question and answer matching
technology to calculate the semantic similarity score
between the original problem and each candidate solution,
choose the best answer based on the highest score, and then
return the user’s response directly. Experimental results of
[9] shows that Question-Answer pairing achieves higher
accuracy than simple KG retrieval.
[12] introduces a graph-based QA system for reading
comprehension tests that pick out the sentence in the
passage that best answers a given question by extractingthe
relations. The proposed system consists of three main
modules –Document Processing, Query Processing and
Answer Extraction. In the Answer Extraction module,
comparison between generated graph from Document
processing and query triplets obtained from Query
processing is done to determine a set of matching sub-
graphs. A morphological analysis which includes a tense
variant check for the verb is carried out if no match occurs
in the comparison step which improves the accuracy of the
model. model achieve accuracy of 79.67%
2.3 GAP Analysis
The challenges in knowledge graph construction are :
1) information loss
2) information redundancy
3) information overlapping.
4) Knowledge noise
Information loss occurswheninformationextractionispoor,
resulting in an incomplete output graph. Information
redundancy refers to the repetition of the same entity with
different abbreviations or prepositions, as well as extra
concepts and relations that do not exist in the input text but
do exist in the background knowledge. The information
overlapping challenge refers to whether a knowledge graph
can encode the changing of an attribute. Too much
knowledge incorporation from LMs may divert the sentence
from its correct meaning, which is called knowledge noise
(KN) issue.
Several approaches on which these systems arebased,the
best known of which are based on Information Extraction
using Named Entity Recognition approaches for knowledge
graph construction and Traditional query processing and
pattern matching approaches for Question-answering
systems. The hybrid approach is an alternative trying to
merge the advantages of these methods to fill the weak
points.
3. METHODOLOGY
We have divided the research work broadly in certain
phases. All the phases are discussed separately in this
paper.
3.1 Data Acquisition and Processing
The Knowledge Graph and Question-Answering System
constructed in this paper is based on the Annual Reports of
certain Companies such as Apple, Facebook,ACCand RIL,etc
which discussesvariousfinancial statements,listedcompany
executive information, news, announcements and research
reports. In this research we are consideringonly textual data
for KG Construction. These Reports are generally in
unstructured format, PDF formatsandalsoconsistofvarious
graphics, charts, images, tables and text data which are in a

highly unstructured format; making it difficultfor extraction
of data from PDF reports.
To handle this, we extract the required textual data for
KG construction into a single text file for all companies.
For the Data extraction, we use Layoutparser which
performs Document Image Analysiswiththehelpofstate-of-
the-art detectron2 deep learning model which enables
identify and extract complicated document structures.
A custom dataset is being created and annotated using
labelImg tool by giving annotationsforportionsofdocument
to be extracted and fine tuned the pre-trained faster rcnn
detectron2 model on our custom dataset and annotations.
Finally, with the use of Fine Tuned model, Inferences are
made and extracted outputs are saved in a single text file
which will act as an input data for the constructionofKGand
QA System.
3.2 System Architecture
The proposed System architecture, which presents how the
user queries are processedandhowthesystem will generate
the results, is shown in Fig. 1.
Fig. 1. Proposed System Architecture
3.3 Coreference Resolution
Fig. 2. NeuralCoref identifies the coreferences for an entity
The ACC Company in the above example and resolves them
by replacing with an entity name.
To connect the primary entities with the prepositional
occurrences of those entities in the text, coreference
resolution is used. With this process pronouns get replaced
with proper nouns thereby generating more interpretable
graphs. Additionally, it is employed when a knowledge base
object is connected to several entityreferences.Forinstance,
since ”President Modi” and ”Narendra Modi” relate to the
same person, they should be combined before being linked
to an entity in the knowledge base.CoreferenceResolutionis
also a heavy NLP task. We have used NeuralCoref; a pipeline
extension for spaCy 2.1+ which annotates and resolves
coreference clusters using a neural network. It is integrated
in spaCy’s NLP pipeline. NeuralCoref has been proved to be
both more efficient and accurate for coreference resolution
tasks. Example of working of NeuralCoref pipeline is shown
in Fig. 2.
3.4 Information Extraction
The second phase in creating a knowledge graph is
information extraction. The main challenge is identifying
candidate knowledge units by autonomouslyextracting data
from data sources. Information extraction is challenging
since natural language processing (NLP) technology is
typically required when working with semi-structured or
unstructured data. Entity and RelationExtractionareamong
the important technologies.
The purpose of entity extraction, sometimes referred to
as named entity recognition (NER), is to create ”nodes”
in knowledge graphs.An essential component ofinformation
extraction, entity extraction has a significant impact on the
effectiveness and calibre of later knowledge acquisition.
Entityclass—which includes names of people, places, and
insti- tutions—timeclass—whichincludes datesand times—
and number class—which includes terms like money and
percent- age—are the three basic classes. It is possible to
expand these classes to accommodate various application
areas. In this step,we extract the pairs of entities which are

related to each other from each sentence of the data. We
employed the NLP library SpaCy, in order to categorise
words with their appropriate part-of-speech (POS) tags and
chunked noun and verb phrases extracted in accordance
with specified rules, whichareableto considerall dependent
noun and verbs in a sentence with the help of dependency
parser offered by spaCy. A verb chunk is defined as the verb
and any accompanying adpositions or particles, whereas a
noun chunk is defined as the words characterising the noun.
3.5 Knowledge Graph
We build knowledge graphs using a bottom-up strategy,
Fig. 3. Visualization of KG for Annual Reports ( with zoom
view)
After obtaining the entities i.e. nodes in a graph using
entity-pair extraction, the process then moves on to relation
extraction for edge construction. Extracting the relationship
i.e. edge between entities is required to gain semantic
information. In order to map pairings of entities, we first
extracted relation terms from sentences, such as verbs,
prepositions, and postpositions. We then combined each
relation phrase with its source and target entities to
generate triplets. Extracted triplets by using SpaCy and a set
of linguistic rules based on subjects, objects, predicates, and
prepositions so that it can work on domain independent
data. Along with normal triplet structure i.e. <ent1, relation,
ent2>; we have additionallyextractedauxiliaryrelation,time
and place such as <source, relation, aux rel, target, time ,
place> making it better for further QA System.

where entities and relationships are first pulled from the
data and the upper model layer is generatedvia data driving.
After entities are recorded in triple form, the knowledge
graph can then be visualizedusingtheNeo4Jgraphdatabase.
First, we encode the triplets that were recorded into a CSV
file using UTF-8, and then we used the Cypher import
command LOAD CSV to put the triplets’ data intheCSVfilein
accordance with the triplet ”ENTITY—[r:RELATIONSHIP]->
ENTITY” in the graph database. For the Visualization of a KG
for Annual reports following Cypher query is processed :
”’ LOAD CSV WITH headers from ”file:///Triplets
Database.csv” as row with row where row.source is not null
merge (n:Source {id:row.source}) merge (m:Target
{id:row.target}) merge (n)-[:TO{rel:row.relation}]- (m)
return * ”’
As the input triplet data is large, the generated KG is
huge. Here, Neo4j comes as the best solution as it allows
users to traverse and navigate through the entire KG and
visualize and interpret various properties and attributes
of nodes and edges by hovering over them. Visualization
results for KG for Annual reports are shown in Fig. 3.
We can Visualize triplets only for the specifiedrelation,for
example to visualize triplets only for relation ’excelled’using
following query :
”’match (n)-[:TO{rel:’excelled’}]->(m) return *”’
Results of above query can be seen in Fig. 4. Also by
hovering over the edges and nodes we can display and
interpret their attributes and properties.
Fig. 4. Visualization of Relation by hovering over edges.
3.6 Question-Answering System
The question-answering system suggested in this paper is
built using a combination of two approaches: semantic
parsing, which involves some linguistics andtraditional NLP
methods, requires rules; and another is extracting the
feature information of pertinent entities in the question and
finding matching triplets for the query from knowledge
databases to obtain the solutions. The primary entities and
relationships are determined by analysing theuser’snatural
language queries. Unlike other generalized QA systems
which usually works only for processing Factoid-type
Questions (‘Wh’ type objective questions), The proposed QA
System can work for Factoid type as well as Descriptivetype
of Questions such as questions that start with the keywords
of “why” and “how”.
1) Query Analysis and Processing : In relation to the
context of the query, the similarity computation is
performed to identify comparable sentences from data. In
this paper, the Cosine similarity discrimination method is
adopted. The recognition entityandcandidate entitymust be
transformed into corresponding word vectors in order to
assess how similar they are to one another. Cosinesimilarity
is calculated by the angle between two vectors, given as
follows :-
Two words’ vector directions are nearly identical if they are
synonyms. Their vector directions are essentially the
opposite if they are antonyms. As a result,thecriteria arethe
same if the two words’ semantics are more similar and their
vector directions frequently coincide.
Term Frequency -InverseDocumentFrequencyisreferred
to as TF-IDF. A numerical statistic calledtheTF-IDFratesthe
significance of each word in a document. To count the word
occurrence in each document, we use TfidfVectorizer
functions that are provided by Scikit-Learn library. Term
Frequency: Number of times a word appears in a text
document. Its formula is as follows :-
Inverse Document Frequency: Measure the word is a rare
word or common word in a document.
Finally, multiplying them can obtain TF-IDF values,

Using the cosine similarity, the indexes of text with similar
context as that of question are returned which gives the
sentences which are supposed to be candidate answers.
Once the candidate answers are obtained, the Query
processing is carried out.
In the Query Processing part, Input Query is processed to
find out the expected answer type according to the question
using POS tags and NER tags of the question sentence. With
the help of various linguistics based design rules, the task of
determining required entity type is carried out such as in
short way, following describes the design rules for finding
answer type for required solution :
“WHO” : “PERSON” or “ORG”
“WHEN” : “TIME”
“WHERE” : “GPE”
“HOW MUCH” : “MONEY”
“HOW MANY” : “QUANTITY”
“HOW” : “DESCRIPTION”
“WHAT” : “DESCRIPTION”
“WHY” : “DESCRIPTION”
2) Answer Matching: Once the answer type is estimated,
for Answer Extraction, thecorrespondingequivalent entities
are searched from candidate answers and returned as the
final solution to the query. In case, if the system cannot
identify the correct candidate answer and hence fails to
extract the required entities, then System will moveforward
to knowledge database retrieval. In the Knowledgedatabase
retrieval method, first the triplets are extracted from a
question which defines what information the user is
demanding. Then the answer isextracted bytriplet matching
where the question triplet is traversedand matchedover the
triplets database. For Descriptive Kind of questions, the
candidate answer found with usage of cosine similarity
feature is returned as the solution by system.
4. RESULTS AND ANALYSIS
We implemented the Knowledge Graph and Question
answering System for the Annual Financial Reports of
various organizations, such as Facebook, Apple, RIL, and
ACC,etc. We have constructed a single Knowledge graph for
textual data comprising data from all company’s Reports. In
the Knowledge graph Construction Process, the proposed
system is able to identify and extract a total 77,575 triplets
comprising entity-pairs and relations. With the use of
spaCy’s Dependency Parser, it achieved the extraction of all
possible entity-pairs and relations between them in a better
way for the complex grammar structured data of Annual
reports. Fig. 3. Shows the Knowledge Graph generated for
Annual Financial reports of Companies.
In the actual use of the QA system, the combination of
information retrieval from document and Knowledge graph
based Question-Answer triplet matching is used. With the
use of these combined approach, Proposed QA achieves the
functionality to work for both Factoid as well as Descriptive
type of questions. User queries’ results are relatively
accurate due to high quality data and strong Knowledge
fusion (Coreference resolution) steps. Currently, accuracyis
used as the basis for the performance test of the question
answering system. The accuracy increases with the number
of correct questions the question-answering system
answers. To handle QA system evaluation’s subjectivity,test
data is collected from users through Google Forms where
users were asked to provide question and answer for given
paragraph data. For the Annual Financial reports, the
proposed system achieved the accuracy of 81.34% by
answering 327 questions correctly out of 402 questions.
For testing the Generalization capability of the proposed
system, performanceofthesystemistestedonother-domain
data such as Ayurveda and Wildlife Management. PDF data
for Ayurveda and Wildlife Management is collected from
https://www.ayurveda.com/pdf/intro ayurveda.pdf and
https://old.mgkvp.ac.in/Uploads/Lectures/49/859.pdf
respectively. It has been observed that the system is able to
identify and extract all possible entity-pairandrelations. For
Empirical verification, Question-Answer dataset is collected
from users for Ayurveda and Wildlife management domain
through the Google Forms. Following table summarizes the
results :
Table -1: Results
Domain
Correctly
Answered
Questions
Total
Questions
Accuracy
Ayurveda 128 150 85.33%
Wildlife
Management
121 150 80.66%
Financial Annual
Reports
327 402 81.34%
Accuracy of 85.33% and 80.66% is achieved for other
domains, Ayurveda and Wildlife Management data
respectively.
For the comparative evaluation, we compared our system
with existing system [12], which uses a graph-based
approach for answer retrieval for QA system handling
reading comprehensiontestswithaccuracy79.67%.Existing

system performs the Document processing and Query
Processing parallely. Graphisgeneratedfortripletsobtained
from the document, extracts the triplets fromuserquestions
using spaCy and at last answer is extracted by subgraph
matching. So, the existing system fails if the exact matching
subgraph is not found. Also, it works for Factoid-type
questions only and can’t answer descriptive questions. To
address these issues, the proposed system uses cosine
similarity feature to find the candidate answers that match
the context of questions along with the answer retrieval
based on knowledge database triplet matching and
addresses descriptive questions as well.
However, the proposed system fails to answer the
analytical and confirmation(yes/no)typeofquestions.Fig.5
shows the summary of comparative analysis for graph-
based(existing system) and Combination of Knowledge
graph QA along with retrieval based on the feature
information(proposed system).
Fig. 5. Comparison graph in terms of Accuracy
5. CONCLUSIONS
This system establishes a generalized approach for
constructing Knowledge Graph and an automated querying
engine to answer user queries for faster and better
understanding of huge data. The system resolves the issues
of ambiguity and semantics of text by Named Entity
Recognition using spaCy which has been observed to be
successful for extracting entities from complex structured
text. Experimental results show that the proposed approach
works efficiently for the different domain data. The
combined approach, Knowledge graph QA along with
retrieval based on the feature information used in the
proposed QA system widens system scope, as it allows the
system to process and answer descriptive questions.
However, the system generates only one sentence answers
for the descriptive queries. Also, the system is unable to
answer confirmation and analytical questions correctly. In
the future work, more advancement can be done to the
system to provision processing of various types ofquestions
and inference based answer retrieval.
REFERENCES
[1] K. Khadilkar, S. Kulkarni and S. Venkatraman, ”A Knowl-
edge Graph Based Approach for Automatic Speech and
Essay Summarization,”2019 IEEE5thInternational Con-
ference for Convergence in Technology(I2CT),2019,pp.
1-6, doi: 10.1109/I2CT45611.2019.9033908.
[2] S. Zehra, S. F. M. Mohsin, S. Wasi, S. I. Jami, M. S. Sid-
diqui and M. K. -U. -R. R. Syed, ”Financial Knowledge
Graph Based Financial Report Query System,” in IEEE
Access, vol. 9, pp. 69766-69782, 2021, doi:10.1109/AC-
CESS.2021.3077916.
[3] Z. Dai, X. Wang, P. Ni, Y. Li, G. Li and X. Bai, ”Named
Entity Recognition Using BERT BiLSTM CRF for Chi-
nese Electronic Health Records,” 2019 12th
International Congress on Image and Signal Processing,
BioMedical Engineering and Informatics (CISP-BMEI),
2019, pp. 1- 5, doi: 10.1109/CISP-
BMEI48845.2019.8965823
[4] R. Miao, X. Zhang, H. Yan and C. Chen, ”A Dynamic
Financial Knowledge Graph Based on Reinforcement
Learning and Transfer Learning,” 2019 IEEE
International
[5] Liu, W., Zhou, P., Zhao, Z., Wang, Z., Ju, Q., Deng, H.,Wang,
P. (2019, September 17). K-Bert: Enabling language
representation with knowledge graph. arXiv.org.
[6] C. Chantrapornchai and A. Tunsakul, ”Information
Extraction based on Named Entity for Tourism Corpus,”
2019 16th International Joint Conference on Computer
Science and Software Engineering (JCSSE), 2019, pp.
187-192, doi: 10.1109/JCSSE.2019.8864166.
[7] X. Wu, J. Wu, X. Fu, J. Li, P. Zhou and X. Jiang, ”Automatic
Knowledge Graph Construction: A Report on the 2019
ICDM/ICBK Contest,” 2019 IEEE International
Conference on Data Mining (ICDM), 2019, pp. 1540-
1545, doi: 10.1109/ICDM.2019.00204.
[8] H. Jiang, B. Yang, L. Jin and H. Wang, ”A BERT-Bi-LSTM-
Based Knowledge Graph Question Answering Method,”
2021 International Conference on Communications,
Information System and ComputerEngineering(CISCE),
2021, pp. 308-312, doi:
10.1109/CISCE52179.2021.9445907

[9] Z. Jiang, C. Chi and Y. Zhan, ”Research on Medical
Question Answering System Based on Knowledge
Graph,” in IEEE Access, vol. 9, pp. 21094-21101, 2021,
doi: 10.1109/ACCESS.2021.3055371.
[10] Z. Chen, S. Yin and X. Zhu, ”Research and
Implementation of QA System Based on the Knowledge
Graph of Chinese Classic Poetry,” 2020 IEEE 5th
International Conference on Cloud Computing and Big
Data Analytics (ICCCBDA), 2020, pp. 495-499, doi:
10.1109/ICCCBDA49378.2020.9095587.
[11] Yasunaga, M. (2021, April 13). QA-GNN: Reasoning with
Language Models and Knowledge Graphs for. . .
ArXiv.Org. https://arxiv.org/abs/2104.06378
[12] G. Veena, S. Athulya, S. Shaji and D.Gupta,”Agraphbased
relation extraction method for question answering
system,” 2017 International Conference on Advances in
Computing, Communications and Informatics (ICACCI),
2017, pp. 944-949, doi: 10.1109/ICACCI.2017.8125963.
[13] Ostendorff, M. (2019, September 18). Enriching BERT
with Knowledge Graph Embeddings for Document
Classification. ArXiv.Org.
https://arxiv.org/abs/1909.08402
[14] K. Guo, T. Jiang and H. Zhang, ”Knowledge Graph
Enhanced Event Extraction in Financial Documents,”
2020 IEEE International Conference on Big Data (Big
Data), 2020, pp. 1322-1329, doi: 10.1109/Big-
Data50022.2020.9378471
[15] Hsu, C. (2019, December 3). Knowledge-EnrichedVisual
Storytelling. ArXiv.Org.
https://arxiv.org/abs/1912.01496
[16] X. Yang et al., ”A General Solution and Practice for
Automatically Constructing Domain KnowledgeGraph,”
2020 IEEE 6th International Conference on Computer
and Communications (ICCC), 2020, pp. 1675-1681, doi:
10.1109/ICCC51575.2020.9344946.

Knowledge Graph and Similarity Based Retrieval Method for Query Answering System

More Related Content

Similar to Knowledge Graph and Similarity Based Retrieval Method for Query Answering System (20)

More from IRJET Journal (20)

Recently uploaded (20)

Knowledge Graph and Similarity Based Retrieval Method for Query Answering System