Volume 7, Issue 11, November – 2022 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Hooman – An Online Dating Website

Jayasekara K.R.T. Manori Gamage
Department of Information Technology SLIIT Department of Information Technology SLIIT
Malabe, Sri Lanka Malabe, Sri Lanka

Ginigaddare G.D.U.D. Hansika Mahadikara

Department of Information Technology SLIIT Department of Information Technology SLIIT
Malabe, Sri Lanka Malabe, Sri Lanka

Wijerathne M.K.N.B Senarathna A.M.M.D.N

Department of Information Technology SLIIT Department of Information Technology SLIIT
Malabe, Sri Lanka Malabe, Sri Lanka

Abstract:- In society there are two types of people some person to spend the rest of their life, such as having a
are introverts and extroverts. Introverts prefer not to be personalitytest using an algorithm, the system calculates the
socialized much and could be shy to talk to people, but personalityof the user and suggests mutual users same as the
in the end, we are all humans we all need someone to be user’spersonality, but this is not enough if the users are more
with not to be lonely, soat a point in their life they matured with the system and found a good match for them
need a mate to live with, so they could date or marry their next step would be the marriage or some of the people
someone, some of the introverts don’t know how to start start their relationship by matching horoscopes this is mostly
a conversation even if they do most of the time they would happening in Sri Lanka because we have a culture to maintain
feel awkward when having a conversation, to overcome and most family members approve the partner only if the
this new advanced series of features have been added. horoscope getsmatched. So as our second feature we can use
Using a profile matcher with a series of questions shows a horoscope matching feature, but this is not enough to get to
what would be their personality and by grading them, know one another because the horoscope doesn’t say
users who have the same grades will be matched, this anything about what kind of a person they are, but what if they
would make it easier for people to get connected with could know what the strengths and weaknesses are of each
one another because they might have the same hobbies other in aromantic relationship, this way they can get an idea
and same interests, so the bonding experience is easy, as about each other. A game will analyze the strengths and
an advanced feature, having a horoscope matcher isvery weaknesses ofthe player. Finally, an intelligent chat service
valuable for the people in Asia, especially in Sri Lanka most introverts need to be trained before talking to people
because of the culture. Here you need to give the image of because most of them don’t know how to respond to other
the horoscope and it automatically matches with others people’s texts sothe intelligent message supporter gives them
and gives suggestions, an intelligent message supporter as suggestions from previous chats, if a similar conversation
a feature, some introverts have no idea what they have happened before then this chatbot gives what to say, so you
to say if another person is texting them, so using this can select it and sendit. Regarding the game, two game levels
intelligent message supporter we suggest what are the have been developed where each level predicts a strength or
responses that user can give to the other person. Video a weakness. In level one, the listening skill of the player will
games tend to attract more people just because they are be predicted. In level one player must navigate through a
interactive and fun and eventful, so a video game is maze with the help of audio clips [6][7]. There are invisible
perfect to predict the strengths and weaknesses of the score points inside the maze if the player goes along the
player in order to give the player’s partner an idea about correct path according to theaudio clips score points will be
them. added to the score and there arepenalty points that will be used
to deduct the score points of the player if they went off course.
Keywords:- Introverts, Profile Matcher, Horoscope Matcher, A Machine Learning model will help to predict the strengths
Intelligent Message Supporter, Video Game. and weaknesses of the player. The second level is somewhat
trickier because it measures problem-solving skills. It’s a
I. INTRODUCTION puzzle-solving game and this level will improve and predict
the user’s problemsolving skills. According to Mayer and
When it comes to online dating, the person we like can Wittrock (1996, 2006) [5], there area few characteristics of
be a psychopath or worse even though from the first problem-solving.
impressionto be the most honest and innocent person. But 1) It is a cognitive process.
if we can get to know this person to some extent the risks can 2) It is goal oriented.
be mitigated, so using some advanced features with dating 3) Complexity (and hence difficulty) of the problem de-
sites will help users to be more careful and to pick a decent pends on one’s current knowledge and skills. and since

IJISRT22NOV1405 www.ijisrt.com 1688

Volume 7, Issue 11, November – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
puzzle games have these characteristics and cantap into
multiple cognitive abilities [8] puzzle games can be used
to measure and improve problem-solving skills of a
player.

II. LITERATURE REVIEW

Over time there are many features came up with dating

platforms and the community also talked about developing
some advanced features, but these were harder to do when
we are talking about Tinder, they just give photos of people
to swipe left or right if they like them. This was not an
accurate approach to matching someone. When it comes to
online dating there is a saying, “you have to kiss a lot of frogs
to find a prince”. Meeting someone online is fundamentally
Fig.1. overall diagram
different than meeting someone in real life [1]. Therefore,
having a personality test describes more about the user to the
Intelligent Message Supporter. In the conversational
system then using machine learning we can suggest what are
dataset, responses are grouped as sender and receiver
the best matches for them so the user can see only people
removing the stop words and making a bag of words so the
who have similar personalities just like them which makes
unique and meaningful words are in that bag so the response
it more exciting. If the user trusts the system, then the user
is generated from this bag of words, definitely, these
will not be afraid and move on with the date. When it comes
sentences are not going to make any sense because they
to matching people with other people within Sri Lanka or
cannot be grammaticallycorrect since there are no stop
within South Asia horoscope can play a main role but this
words and might not make any sense either, because
feature is not implemented on any dating website because
processing sentences, so while we are doing feature
people had not thought to a greater extent than people have
engineering, we make two arrays to store these sender and
proposal matching websites[5] but you need to have a partner
responders values after that using seq2seq model and
to match your proposal this is always not easy so if we
LSTM[1] we make a relationship between these two groups
use those implementations in a dating website people can
and categorical cross entropy as the loss function, therefore,
use their horoscopes and match it with the person who they
it will calculate how bad the guess is and give it to the
got suggested with if it is and if the user likes the matched
optimizer Adam so it can make a better prediction asfor
person, then they can even build up a future without
the next prediction. This is the machine learning model
hesitating. This horoscope matching is included in our
training process, now this model is stored in the frontend and
system, and itis consulted and instructed to be made by a real
when the user types in the chat the text passes into the Machine
astrologer because most of the time online horoscope
Learning model and predicts what would be the response, and
matching could be wrong and end up in failed marriages [2].
using the tokenization method, the system tokenizes the text,
so it has a numerical value so it’s easier to find the similar
III. METHODOLOGY
sentences which have the same words and token values as the
response, this can be a range of words and limited it to
To implement this we used multiple number of data sets
asthree sentences these stored responses are used by the users
because each of these systems we different from each other
who used the chat earlier. Seq2seq problems are a special
and only two of them need to have data sets that would
class of Sequence modeling problems, both these problems
be the intelligent chat bot and “profile suggesting”
are regarding input and output of sequences. Specially
components, both these data sets were gained from Kaggle,
Encoder- Decoder models were used to solve Seq2Seq
one dataset is used to analyze the conversations of people in
problems. For the encoder we pass the input sequence
movies and if the models were trained with them then
according to this scenario we pass the text which user types
model learns howto think like a normal human, the main
as the message, from the encoder it generates a vector, this is
purpose is to get information about the user and suggest the
a collection of tokens and this is always fixed length this is
best match possible for them therefore when it comes to
called as ‘context vector’ thisheld the whole meaning of the
horoscope matching all need to be researched from many
sentence as a vector, this vector is making the decoder easier
articles and find the patternsin them and understand how the
to make the accurate predictions. The decoder reads this
concept work this was done by learning from astrologer to
vector and goes from each token to token and targets the
match the patterns and this“ information ” is made by the
sequence token by token [2].
researcher himself and two array sets have been used to train
the gaming model to get the relationship between the score
of the game(1st array set - 5,10, 15, ...) and a value that
need to be given (2nd array set -0, 0, 1, 1,...) which is
between 0 and 1.

IJISRT22NOV1405 www.ijisrt.com 1689

Volume 7, Issue 11, November – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
main technologies that are used to build this component are
image processing and CNN (Convolutional Natural Network)
which are used for building deep learning models. This CNN
behaves like a human brain when it is identifying images
[10],[11]. When we use this CNN, it needs a data set to train
the Machine Learning model to identify the combinations
betweenhoroscopes since there were no found data for this
purpose to be found on the internet this data was made by the
researcher himself with the help of a professional Astrologer.
There are 12 astrological signs and this astrological sign when
we are writing in Sinhala it can be different in many ways it
Fig.2. Basic process structure in Intelligent Message changes from person to person therefore only from one
Supporter (Chatbot) astrological sign researcher took 250 to 300 images to train
what is the written name looks like to train the model to
capture the imageusing CNN so that model knows exactly
what the astrological sign is accurately. So, it has a collective
knowledge of around more than 3000 images, so the model
gets the value correct way [14]. Now for the Machine
Learning algorithm this component passes the Astrological
sign that’s detected anddo the prediction using this method
matching the horoscopeis promising and the results are very
accurate also since this needs to improve much more later
as for the start this is a very good start and a very good
method to math horoscopes, the accuracy of the Machine
Learning model is high more like85as shown in the Fig 4),
when considering accuracy this isa good rate
Fig.3. Summary of the model

Profile Verification System. A Convolutional Neural

net- work, which is one of the main image processing
modelsused in neural networks has been trained in identifying
char- acteristics of images related to face attributes. As per
image preprocessing, all the images are converted to
grayscale to reduce the noise of the images. The following
classifications were conducted by processing images using
CNN’s. Four models have been built using CNN in order to
classify age, gender, hair, and nose. Based on the model and
its classifica- tion accuracy, convolution layers were used
with an activation function. The dense layer of the output
layer has a softmax activation function according to
categories that needs to be classified by the model.
Fig.5. Model training Diagram

Fig.4. Model Training in the Profile Verification System

A training test split is used to sample the training

dataset against the test data set. Model checkpoint has been
used tosave the best model with the highest accuracies.
Fig.6. Model training in Horoscope Matching Component
After training is finished Machine Learning model
Horoscope Matching System Users must upload an needs to be deployed to the backend so from the frontend
im- age of their horoscope. Using image processing, The image is passed to the backend of our system as a POST

IJISRT22NOV1405 www.ijisrt.com 1690

Volume 7, Issue 11, November – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
request through the API. Earlier CNN was used to process
the data of the image and the accuracy of the image
detection was high from that but before moving on to the
Machine Learning model Optical Character Recognition
(OCR) is used to detect the image’s data, to process the
image and get the data. Todo this we have used 550 images
to train the model, but the accuracy of the result is very
poor (accuracy of 34Because of that we moved on to CNN,
OCR process is done by the predefined model itself, but this
CNN model is made by the researcher himself.

Strength and Weakness finder (Game) Mainly used

lo- gistic regression to implement this component. To reduce
codecomplexity (Unity accesses the ONNX model as an asset Fig.7. Chat screen of the Intelligent Message Supporter
so there is no need for hard coding in unity). Especially in a
videogame, we should maintain less code complexity because
it willcause GPU CPU to be in high usage and video games
use GPU and CPU heavily so code complexity will affect
GPU and CPU as a result it will affect the game
performance asthe game grows and becomes more complex
with the future works. We use previously trained data (an
array set of the score given by each level and an array for
results according to the score) which will lead to more
accurate results (accuracy- 0.89) than not using Machine
Learning. For example, if we use if-else we will only know if
the player has good listening skills or poor listening skills. If Fig.8. Profile Verification System result interface
we want to get the percentage of the listening skill, we have
to write more code which will lead to more complexity of the
game (as mentioned above). It is better to use a Machine
Learning algorithm to predict the player’s strengths and
weaknesses which was trained using training data to get a
more accurate result with the percentage of the listening skill
because even if the player got the full score, it doesn’t mean
that the player has the best listeningskill so prediction using
previously trained data is the best option here. There are two
levels in this game for the second level we mainly focus on
how good the user is when it comes to solving problems, for
this purpose a jigsaw puzzle is used, When the player solves
the puzzle and if they complete the puzzle with extra time left
on the timer, they get extra points but if they take longer to
complete the puzzle than the given time a penalty will Fig.9. Horoscope matching result
impose. So, the result that goes into the system for the
Machine Learning model will be reduced. And for every
correct placement, the game score will be increased.

IV. RESULTS AND DISCUSSION

It’s a highly advanced system and much harder to

maneuver since we have to predict everything that user does
when they logged into the system, there is always a
probability the rate ofbeing successful, and models need to be
trained more, that’s what we have found so far but the
probability of the success rate is high since we use actual
conversations of people and model prediction is most of the
Fig.10. Gamification interface
time suggests pretty much the correct reply that we need to
use for but the accuracy is not much likable and need to
So, what we really need to work on more right now
develop a lot more than right now.
isnot about the optimizers or the loss function but the data
preprocessing part since this is a product that needs to
be with people to help get their shyness and unfamiliarity
about how to talk. The chatbot can help them a lot

IJISRT22NOV1405 www.ijisrt.com 1691

Volume 7, Issue 11, November – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
V. CONCLUSION AND FUTURE WORK Workplace Teambuild- ing Activities: Games and
Exercised Designed to Build Trust Encourage
Since this model is deployed with the website it will Teamwork Among Employees, 2nd ed. Ocala, Florida:
learn new words, new sentences and everything but that’s not Atlantic Publish- ing Group Inc, 2016, p. 153.
enough, need to monitor how regularly people use this model [8]. P. Fissler, O. Küster, D. Laptinskaya, L. Loy, C. von
and the answers it suggests to them if they are using it more Arnim and I. Kolassa, ”Jigsaw Puzzling Taps Multiple
often then it’s great but what if the suggestions are not Cognitive Abilities and Is a Potential Protective Factor
suitableat all, so we have to monitor it, by collecting the for Cognitive Aging”, Frontiers in Aging Neuroscience,
data that users used to chat with each other, more like vol. 10, 2018. Available: 10.3389/fnagi.2018.00299.
chats peopleare having so we can detect what are the areas [9]. Q. Wang, L. Sun and Z. Chen, ”Sentiment Analysis of
that they are more commonly focused on and if that area Reviews Based on Deep Learning Model,” 2019
needs to have an improvement focusing on that area can IEEE/ACIS 18th International Conference on Computer
improve the chatbot a lot. Always this needs to be more and Information Science (ICIS), 2019, pp. 258-261, doi:
focused on the empathy side because all humans need to be 10.1109/ICIS46139.2019.8940267.
treated nicely and gently. As for the game component, it can, [10]. Huiping Li, D. Doermann and O. Kia, ”Automatic text
1) Use more levels to get more strengths or weaknesses of detection and tracking in digital video,” in IEEE
the player Transactions on Image Processing, vol. 9, no. 1, pp.
2) We can use this method to get not only strengths and 147- 156, Jan. 2000, doi: 10.1109/83.817607.
weaknesses but also personality traits too. [11]. Mattohti, M. Ibrahim, A. Dawut and A. Hamdulla,
”Effective Text Detection in Composite Images with
VI. ACKNOWLEDGMENT Simple Background,” 2019 2nd International
Conference on Information Systems and Computer
Authors are highly appreciating the lecturers at Sri Aided Education (ICISCAE), 2019, pp. 263-266, doi:
Lanka Institute of Information Technology for all the 10.1109/ICIS- CAE48440.2019.221631.
support they have given us and special thanks to the Kaggle [12]. P. Ranjitha, K. Rajashekar and Shamjith, ”A Review
Team because of them Machine Learning is made easier and on Text Detection from Multi-Oriented Text Images in
so descriptive about the codes they provide and Laurence Different Ap- proaches,” 2020 International
Moroney about the huge knowledge transferred through Conference on Electronics and Sus- tainable
YouTube and other course programs and every other senior Communication Systems (ICESC), 2020, pp. 240-245,
author who posts answers to wide web forums like Stack doi: 10.1109/ICESC48915.2020.9156002.
Overflow. [13]. L. Rajamanthri and S. Thelijjagoda, ”Plagiarism
Detection Tool for Sinhala Language with Internet
REFERENCES Resources Using Natural Language Processing,” 2021
10th International Conference on Information and
[1]. Y. Wan, ”Short-Text Sentiment Classification Based on Automation for Sustainability (ICIAfS), 2021, pp. 156-
Graph- LSTM,” 2019 International Conference on 160, doi: 10.1109/ICIAfS52090.2021.9605852.
Artificial Intelligence and Advanced Manufacturing [14]. S. Malakar and W. Chiracharit, ”Thai Text Detection
(AIAM), 2019, pp. 35-38, doi: and Classification Using Convolutional Neural
10.1109/AIAM48774.2019.00014. Network,” 2020 59th Annual Conference of the Society
[2]. L. Zhang and H. Deng, ”Sentence Simplification Based of Instrument and Control Engineers of Japan (SICE),
on Multi-Stage Encoder Model,” in IEEE Access, vol. 2020, pp. 99-102, doi:
7, pp. 174248-174256, 2019, doi: 10.23919/SICE48898.2020.9240290.
10.1109/ACCESS.2019.2957160. [15]. W. He, X. -Y. Zhang, F. Yin and C. -L. Liu,
[3]. S. Al-Rousan, A. Abuhussein, F. Alsubaei, O. ”MultiOriented and Multi-Lingual Scene Text
Kahveci, H. Farra and S. Shiva, ”Social-Guard: Detection With Direct Regression,” in IEEE
Detecting Scammers in Online Dating,” 2020 IEEE Transactions on Image Processing, vol. 27, no. 11, pp.
International Conference on Electro Information 5406-5419, Nov. 2018, doi:
Technology (EIT), 2020, pp. 416-422, doi: 10.1109/TIP.2018.2855399.
10.1109/EIT48999.2020.9208268. [16]. V. Skorpil, V. Oujezsky, P. Cika and M. Tuleja,
[4]. T. Ghosal, S. K. Das and S. Bhattacharjee, ”Sentiment ”Parallel Processing of Genetic Algorithms in Python
analysis on (Bengali horoscope) corpus,” 2015 Annual Language,” 2019 PhotonIcs Electromag- netics
IEEE India Conference (INDICON), 2015, pp. 1-6, doi: Research Symposium - Spring (PIERS-Spring), 2019,
10.1109/INDICON.2015.7443551. pp. 3727- 3731, doi: 10.1109/PIERS-
[5]. P. Alexander, P. Winne, R. Mayer and M. Wittrock, Spring46901.2019.9017332.
Handbook of educational psychology, 2nd ed.
Routledge Taylor and Francis Group, 2006, pp. 287-
303..
[6]. N. Mead and D. Rubin, ”Assessing Listening and
Speaking Skills. ERIC Digest.”, Ericdigests.org, 1985.

[7]. J. Peragine and G. Hudgins, 365 Low or No Cost

IJISRT22NOV1405 www.ijisrt.com 1692