Generating and Analyzing Chatbot Responses using

Natural Language Processing

Abstract—Customer support has become one of the most

Artificial intelligence (AI) improves digital marketing in a
important communication tools used by companies to provide
number of different areas from banking, retail, and travel to
before and after-sale services to customers. This includes
communicating through websites, phones, and social media healthcare and education. While the idea of using human
platforms such as Twitter. The connection becomes much faster language to communicate with computers holds merit, AI
and easier with the support of today's technologies. In the field of scientists underestimate the complexity of human language, in
customer service, companies use virtual agents (Chatbot) to both comprehension and generation. The challenge for
provide customer assistance through desktop interfaces. In this computers is not just understanding the meanings of words,
research, the main focus will be on the automatic generation of but understanding expression in how those words are
conversation “Chat” between a computer and a human by collocated. Moreover, a chatbot is an example of a virtual
developing an interactive artificial intelligent agent through the conversational service robot that can provide human-computer
use of natural language processing and deep learning techniques interaction. Companies use robotic virtual agents (Chatbot) to
such as Long Short-Term Memory, Gated Recurrent Units and assist customers through desktop interfaces [1, 2].
Convolution Neural Network to predict a suitable and automatic
response to customers’ queries. Based on the nature of this Natural language processing (NLP) is a subfield of
project, we need to apply sequence-to-sequence learning, which computer science that employs computational techniques for
means mapping a sequence of words representing the query to learning, understanding and producing human language
another sequence of words representing the response. Moreover, content. NLP can have multiple goals; it can aid human-
computational techniques for learning, understanding, and human communication, such as in machine translation and aid
producing human language content are needed. In order to human- machine communication, such as with conversational
achieve this goal, this paper discusses efforts towards data agents. Text mining and natural language processing are
preparation. Then, explain the model design, generate responses, widely used in customer care applications to predict a suitable
and apply evaluation metrics such as Bilingual Evaluation response to customers, which significantly reduces reliance on
Understudy and cosine similarity. The experimental results on call center operations [3].
the three models are very promising, especially with Long Short-
Term Memory and Gated Recurrent Units. They are useful in AI and NLP have emerged as a new front in IT customer
responses to emotional queries and can provide general, service chatbots. The importance of these applications appears
meaningful responses suitable for customer query. LSTM has when no technicians manage the customer service office due
been chosen to be the final model because it gets the best results to the end of working time or their presence outside the office
in all evaluation metrics. [4].
Keywords—Chatbot; deep learning; natural language In this project, the main focus will be on the automatic
processing; similarity generation of conversation ”Chat” between a computer and a
human by developing an interactive artificial intelligent agent
I. INTRODUCTION using deep learning. This will provide customers with the right
With the arrival of the information age, customer support information and response from a trusted source at the right
has become one of the most influential tools companies use to time as fast as possible.
communicate with customers. Modern companies opened up This project aims to build an automated response system
communication lines (conversations) with clients to support (Chatbot) that responds to customer queries on social
them regarding products before and after-sales through networking platforms (Twitter) to accelerate the performance
websites, telephones, and social media platforms such as of the service. Also, to keep the simplicity in mind while
Twitter. This communication becomes faster and much easier designing the system to enhance its efficiency.
with the support of the technologies that are being used today.

1|Page
 This project centers around the study of deep learning manual intervention and supervision, which affects the speed
models, natural language generation, and the evaluation of the and quality of processes execution. IT service providers are
generated results. under competitive pressure to continually improve their
We believe that this contribution can add improvement by service quality and reduce operating costs through automation.
applying the right preprocessing steps which may organize Hence, they need the adoption of chatbots in order to speed up
sentences in a better way and help in generating proper the work and ensure its quality [10].
responses. On the other hand, we start with the existing text On the medical side, the field of healthcare has developed
generative models CNN and LSTM and then try to improve a lot, lately. This development appears with the use of
them as well as develop a new model such as GRU to compare information technology and AI in the field. In [11], the authors
results. We focus on evaluating the generated responses from proposed a mobile healthcare application as a chatbot to give a
two aspects: the number of words matches between the fast treatment in response to accidents that may occur in
reference response and the generated response and their everyday life, and also in response to the sudden health
semantic similarity. changes that can affect patients and threaten their lives.
The rest of this paper is organized as follows. Section II Customer services agent is an application of applying
provides reviews of the related works. The methodological chatbot technologies in businesses to solve customer problems
approach is described in Section III. Moreover, dataset and help the sales process. As companies become globalized
collection and analysis in details are provided in Section IV. in the new era of digital marketing and artificial intelligence,
The implementation strategy and results of this project are brands are moving to the online world to enhance the
discussed in section V. Finally, the conclusion of the project customer experience in purchasing and provide new technical
and its future work are provided in Sections VI and VII support ways to solve after-sales problems. Moreover, fashion
respectively. brands such as Burberry, Louis Vuitton, Tommy Hilfiger,
II. LITERATURE REVIEW Levi's, H&M, and eBay are increasing the popularity of e-
service agents [1].
Developing computational conversational models
(chatbots) took the attention of AI scientists, for a number of B. Natural Language Processing
years. Modern intelligent conversational and dialogue systems NLP allows users to communicate with computers in a
draw principles from many disciplines, including philosophy, natural way. The process of understanding natural language
linguistics, computer science, and sociology [5]. This section can be decomposed into the syntactic and semantic analysis.
will explore the previous work of chatbots and their Syntactic refers to the arrangement of words in a sentence
implementations. such that they make grammatical sense. Moreover, syntactic
A. Chatbots Applications and Uses analysis transforms sequences of words into structures that
show how these words are related to each other. On the other
Artificial dialogue systems are interactive talking hand, semantic refers to the meaning of each word and
machines called chatbots. Chatbot applications have been sentence. The semantic analysis of natural language content
around for a long time; the first well-known chatbot is Joseph captures the real meaning; it processes the logical structure of
Weizenbaum‟s Eliza program developed in the early 1960s. sentences to find the similarities between words and
Eliza facilitated the interaction between human and machine understand the topic discussed in the sentences [12].
through a simple pattern matching and a template-based
response mechanism to emulate the conversation [6, 7]. As part of the text mining process, the text needs many
modification and cleaning before using it in the prediction
Chatbot became important in many life areas; one of the models. As mentioned in [13], the text needs many
primary uses of chatbots is in education as a question preprocessing steps which include: removing URLs,
answering system for a specific knowledge domain. In [8], the punctuation marks and stop words such as a, most, and, is and
authors proposed a system that has been implemented as a so on in the text because those words do not contain any
personal agent to assist students in learning Java programming useful information. In addition, tokenizing, which is the
language. The developed prototype has been evaluated to process of breaking the text into single words. Moreover, text
analyze how users perceive the interaction with the system. needs stemming, which means changing a word into its root,
Also, the student can get help in registering and dropping such as “happiness” to “happy”. For features extraction, the
courses by using a chatbot spatialized in student authors use Bag of Words (BoW) to convert the text into a set
administrative problems, as mentioned in [9]. The of features vector in numerical format. BoW is the process of
administrative student‟s chatbot helps the colleges to have transforming all texts into a dictionary that consist of all words
24*7 automated query resolution and helps students have the in the text paired with their word counts. Vectors are then
right information from a trusted source. formed based on the frequency of each word appearing in the
On another hand, information technology (IT) service text.
management is an important application area for enterprise Before entering the data into a model or a classifier, it is
chatbots. In many originations and companies, IT services necessary to make sure that the data are suitable, convenient,
desk is one of the essential departments that helps to ensure and free of outliers. In [14], the authors explain how to
the continuity of work and solving technical problems that preprocess the text data. The main idea was to simplify the
employees and clients are facing. This variability demands text for the classifier to learn the features quickly. For
example, the

2|Page
names can be replaced with one feature {{Name}} in the The customer service agent is an important chatbot that is
feature set, instead of having the classifier to learn 100 names used to map conversations from request to the response using
from the text as features. This will help in grouping similar the sequence to sequence model. Moreover, a sequence to
features together to build a better predicting classifier. On sequence models has two networks one work as an encoder
another hand, emoticons and punctuation‟s marks are that maps a variable-length input sequence to a fixed-length
converted to indicators (tags). Moreover, a list of emoticons is vector, and the other work as a decoder that maps the vector to
compiled from online sources and grouped into categories. a variable-length output sequence. In [4], the authors generate
Other punctuation marks that were not relevant to the coding word-embedding features and train word2vec models. They
scheme are removed. trained LSTMs jointly with five layers and 640 memory cells
Chat language contains many abbreviations and using stochastic gradient descent for optimization and gradient
contractions in the form of short forms and acronyms that clipping. In order to evaluate the model, the system was
have to be expanded. Short forms are shorter representations compared with actual human agents responses and the
of a word which are done by omitting or replacing few similarity measured by human judgments and an automatic
characters, e.g., grp → group and can‟t → cannot. The authors evaluation metric BLEU.
created a dictionary of these words from the Urban Dictionary As a conclusion of reviewing works concerned with the
to replace abbreviations by expansions. Spell checking is conversational system, text generation in English language
performed as the next step of the pre-processing pipeline on and the collaboration of social media in customer support
all word tokens, excluding the tagged ones from the previous service, this paper proposes a work that aims to fill the gap of
steps [14]. limited works in the conversational system for customer
Minimizing the words during the text pre-processing phase support field, especially in the Twitter environment. The
as much as possible is very important to group similar features hypothesis of this project was aiming to improve the
and obtain a better prediction. As mentioned in [15], the automated responses generated by different deep learning
authors suggest processing the text through stemming and algorithms such as LSTM, CNN, and GRU to compare results
lower casing of words to reduce inflectional forms and and then evaluate them using BLEU and cosine similarity
derivational affixes from the text. The Porter Stemming techniques. As a result, this project will help to improve the
algorithm is used to map variations of words (e.g., run, text generation process in general, and customer support field
running and runner) into a common root term (e.g., run). in particular.
Words can not be used directly as inputs in machine III. METHODOLOGICAL APPROACH
learning models; each word needs to be converted into a
This section discusses the background of the implemented
vector feature. In [4], the authors adopt the Word2vec word
methods, explain why these methods are appropriate and give
embedding method to learn word representations of customer
an overview of the project methodology.
service conversations. Word2vec's idea is that each dimension
of inclusion is a possible feature of the word, which can A. Text Generative Model
capture useful grammatical and semantic properties. Based on the nature of this project, which is generating a
Moreover, they tokenize the data by building a vocabulary of proper response to every customer query in social media,
the most frequent 100K words in the conversations. applying sequence-to-sequence learning are needed.
C. Machine Learning Algorithm and Evaluation Moreover, sequence-to-sequence means mapping a sequence
of words representing the query to another sequence of words
A large number of researchers use the idea of artificial representing the response, the length of queries and responses
intelligence and deep learning techniques to develop chatbots can be different. This can be applied by the use of NLP and
with different algorithms and methods. As mentioned in [16], deep learning techniques.
the authors use a repository of predefined responses and a
model that ranks these responses to pick an appropriate Sequence-to-sequence models are used in many fields,
response for a user‟s input. Besides, they proposed topic including chat generation, text translation, speech recognition,
aware convolutional neural tensor network (TACNTN) model and video captioning. As shown in Fig. 1, a sequence-to-
to classify whether or not a response is proper for a message. sequence model consists of two networks, encoder, and
The matching model used to select a response for a user decoder. The input text enters the encoder network in reverse
message. Specifically, it has three-stages that include: pre- order, then it is converted into a sequence of fixed length
processing the message, retrieving response candidates from context vector, which is then used by the decoder to generate
the pre-defined message-response pair index, then ranking the the output sequence [18].
response candidates with a pre-train matching model.
In [17], the authors train two word-based machine learning
models, a convolutional neural network (CNN) and a bag of
words SVM classifier. Resulting scores are measured by the
Explanatory Power Index (EPI). EPI used to determine how
much words contribute to the classification decision and filter
relevant information without an explicit semantic information
extraction step. Fig. 1. Sequence to Sequence Model.

3|Page
 Before inserting the sequence of words into the encoder
model, it needs to be converted into a numerical format; this
can be applied by using NLP techniques. This project focused
on Bag of Words, or BoW vector representations, which is the
most commonly used traditional vector representation for text
generating models. BoW is used to transforms all texts into a
dictionary that consists of all words that appear in the
document [13]. It then creates a set of features in real number
inside a vector for each text.
B. Deep Learning Models Fig. 3. The Architecture of RNN.

1) Convolutional Neural Network (CNN) Model: In this C. Project Methodology

project, CNN is chosen mainly for its efficiency, since CNN is
In order to implement this project, several preprocessing
faster compared to other text representation and extraction
and modeling steps are performed. First, split the original
methods [19]. The CNN consists of the convolution and dataset into train and test sets. Then, prepare the dataset for
pooling layers and provides a standard architecture that takes a modeling. The preparing process includes preprocessing steps
variable-length sequence of words as an input and then passes and features extraction. After that, train models using train set
it to a word embedding layer. The embedding layer maps each with LSTM, GRU, and CNN. Finally, prepare the test set and
word into a fixed dimensional real-valued vector then passes it use it for evaluating the models. Fig. 4 illustrates the
to the 1D convolutional layer. The output is then further methodology steps.
down- sampled by a 1D max-pooling layer. Outputs from the
pooling layers are then fed into the final output layer to
produce a fixed-length feature vector [20]. CNN has been
widely used in image and video recognition systems, and,
lately, they have shown promising results in NLP applications
[21]. Fig. 2 shows the standard architecture of the CNN
model.
2) Recurrent Neural Network (RNN) Model: In a
traditional neural network, all inputs and outputs are
independent of each other, which is not useful when working
with sequential information. Predicting the next word in a
sentence requires knowing the sequence of the words in the
sentence that come before the predicted word. Among all
models for learning sentence representations, recurrent neural
network (RNN) models, especially the Long Short Term Fig. 4. The General Implementation Steps.
Memory (LSTM) model, are the most appropriate models for
processing sentences, as they have achieved substantial
success in text categorization and machine translation [22].
Therefore, this project applies LSTM and Gated Recurrent
Units (GRU) as a newer generation of Recurrent Neural
Networks. Fig. 3 illustrates the basic architecture of RNN.
Hochreiter & Schmidhuber introduced Long Short Term
Memory Networks in 1997. They solve the problem of
vanishing and exploding gradient problem that is prevalent in
a simple recurrent structure, as it allows some states to pass
without activation. In 2014, Cho et al developed GRU
networks in an effort to design recurrent encoder-decoder
architecture [23]. They are relatively more straightforward
than LSTM and retain a majority of its advantages.

Fig. 2. The Architecture of CNN.

4|Page
 IV. IMPLEMENTATION STRATEGY The experiments are applied using three different models
In this section, we are going to explain the methodology LSTM, GRU, and CNN. The models use a training dataset of
followed for this project. At first, prepare the dataset for around 700k pairs of queries and responses and a testing dataset
modeling. The preparing process includes preprocessing step of 30k of unseen data. Training time is between 5 and 12 hours,
and features extraction then train the models using a training depending on the model ( see Table III).
set and evaluate them with a test set.
A. Data Preprocessing
A data analyst cannot handle raw text directly to suit
machine learning or deep learning methods. Therefore, it is
necessary to work on texts‟ preprocessing from all existing
impurities, for example, punctuation, expression code, and
non-English words (Chinese, Spanish, French, and others). In
order to do this, a number of python NLP libraries such as
regular expression (RE), unicodedata, langdetect, and
contractions are used.
In this project, the performed preprocessing steps include:
remove links, images, Twitter ID, numbers, punctuation,
emoji, non-English words and replace abbreviations with long
forms. Table II illustrates the changes in the dataset before and
after applying all the previous preprocessing steps.
The preprocessing steps are chosen carefully; not all
preprocessing techniques are suitable for this kind of projects.
For example, removing stopwords and text stemming cannot
be applied because it will affect the sentences structures as
well as the text generation process.
B. Feature Extraction
Before doing any complex modeling, the dataset needs to
be transformed into a numerical format suitable for training.
The Bag of Words (BOW) concept is applied to extract
features from the text dataset. First, all of the texts in the
dataset are split into an array of tokens (words). Then, a
vocabulary dictionary is built with all of the words in the
dataset and its corresponding index value. The array of words
is then converted to an array of indexes. This process can be
applied by the use of the sklearn‟ predefined method called
CountVectorizer.
In order to handle variable length, the maximum sentence
length needs to be decided. Moreover, all remaining vector
positions should be filled with a value („1‟ in this case) to
make all sequences have the same length. On the other hand,
words not in the vocabulary dictionary will be represented
with UNK as a shortcut of unknown words. Moreover, each
output text in the dataset will start with a start flag („2‟ in this
case) to help in training. Now the dataset is ready for training.
C. Modeling
The infrastructure used for experimentation involves
google colaboratory and Crestle cloud services which are
GPU-enabled Jupyter environments with powerful computing
resources. All popular scientific computing and deep learning
packages are pre-installed and configured to run on a GPU.

5|Page
 TABLE. II. THE CHANGES IN TEXT BEFORE AND AFTER
APPLYING PREPROCESSING STEPS

Before preprocessing After preprocessing

@115743 C91. Feel free to keep an

feel free to keep an eye on the ps
eye on the PS Blog for news and
blog for news and updates
updates: https://t.co/aLtfBAztyC

@133100 We do our best to clear as we do our best to clear as many

many upgrades as we can, send us a upgrades as we can send us a dm
DM with the reservation you're with the reservation you are referring
referring to and we'll take a look. to and we will take a look

@129388 We'd like to look into this

we would like to look into this with
with you. To confirm, did you update
you to confirm did you update to ios
to iOS 11.1? Please DM us here:
please dm us here
https://t.co/GDrqU22YpT

TABLE. III. TRAINING TIME IN HOURS

Model Training Time in Hours

LSTM 12
GRU 8
CNN 5

In the experiments, multiple parameters are tested, and

their effects are addressed. All models are tested with
varying dimensionality of the word embeddings (100, 300
and 640), it was observed that models perform better and
faster with 100- word embedding size.
The dataset is large, the number of vocabularies is
388,950 unique words, and our computers cannot handle it.
So, only the frequent words appeared in the dataset should
be used. The most frequent words are decided by the
max_features parameter in the CountVectorizer function
which sort words by its frequency then choose the most
frequent words. The first vocabulary size in the experiments
is 8000 and then it increases, taking into consideration
memory limitation. A slight improvement has been
recognized in all models and because of the memory
limitation, only 10,000 of the vocabularies are used.
Moreover, the GRU model was trained for eight epochs but
without significant improvement. The three models are all
trained under the same conditions. Table IV shows the
common parameters used in all models.

TABLE. IV. THE COMMON PARAMETERS USED IN LSTM, GRU AND CNN
MODELS

6|Page
Parameter Value
Word embedding dimension size 100
Vocabulary size 10,000
Context dimension size 100
Learning rate 0.001
Optimization function Adam
1000 (the max that our computer can
Batch size
handle)
Max message length 30

Fig. 8. Good Result Example.

The following are the common layers used in the models,

starting from inserting the sequence of words into the model to
generating the responses:
 Last Word Input Layer: Inputs the last word of the
sequence.
 Encoder Input Layer: Inputs sequence data and pass it to
the embedding layer.
 Embedding Layer: Used to create word vectors for
incoming words.
 Encoder Layer (LSTM, GRU, CNN): Creates a temporary
output vector from the input sequence.
 Repeated Vector Layer: Used like an adapter to fit the
encoder and decoder parts of the network together. It can be
configured to repeat the fixed-length vector one time for each
time step in the output sequence.
 Concatenate Layer: Takes inputs and concatenates them
along a specified dimension.
 Decoder Layer (LSTM, GRU, CNN)(Dense): Used as the Fig. 9. Bad Result Example.
output for the network.
 Next Word Dense Layer: Takes inputs from the previous
layer and outputs a one vector representing the target word.
 Next Word softmax Layer: Applies a softmax function E. Result and Discussion
that turns the dense layer output into a probability distribution,
from to pick the most likely next word. In this project, the process of generating responses take
around 6 hours for each model to be accomplished. Moreover,
calculating BLEU and cosine similarity scores takes around 4
hours.
D. Generating Responses :
The models are evaluated automatically based on the words
After training the models, the generating responses process using BLEU score. The BLEU is applied for 1-gram, 2-gram, 3-
is started using the 30k test set. The following are samples of gram, and 4-gram in order to explore the strength of the models.
the generated responses from all models (see Fig. 8 and 9). It can be seen that LSTM and GRU models outperform the
official baseline LSTM model [4] with respect to the 4-gram
BLEU score. Fig. 10, shows in details the performance of
models in each n-gram.
Hence it can be seen that LSTM achieves the highest
evaluation scores for all grams, but it takes a long time in
training. Moreover, the GRU model has very close evaluation
scores to LSTM. In the other hand, the CNN model has the
lowest evaluation scores compared with all RNN models but
achieves high-speed performance, which can be useful in
application trained on large datasets.

7|Page
 Fig. 10. The BLEU Scores for 1, 2, 3 and 4 Grams.
i am sorry for the trouble with your order
GRU Generated please reach out to us here and we will look
Response into this for you please do not provide your
order details

hi there is not provide your order number and

CNN Generated
we can you please dm us a dm us a dm us a dm
Response
us a dm us a dm us
Furthermore, another evaluation metric cosine similarity
are applied to captures the semantics beyond responses and
gives similarity scores. It has been found that RNN models
capture the semantics in the responses and they are more
effective in improving the reply quality than the CNN model. TABLE. VII. EXAMPLE OF INFORMATIVE QUERY AND RESPONSES FROM ALL
MODELS:
Fig. 11 shows the similarity scores for each model.
After exploring the generated responses and get in-depth guys when are you going to open your services
Customer Query
in the good and bad results, it has been found that RNN in middle east
models, in general, are good in responses to emotional hulu store is only available in the us at this time
queries more than an informative one. The models can Customer Support
but we will share the interest in bringing our
Response
provide general, meaningful responses suitable for customer service to the middle east
query. Table VI shows an example of an emotional query. hi there we are sorry to hear about this please
LSTM Generated
On the other hand, the queries that are more informative dm us with your email address so we can
Response
connect
and ask about specific information are hard to generate, and
the generated responses become less efficient. Table VII hi there i am sorry to hear about this please
GRU Generated
shows an example of an informative query. Response
dm me the details of the issue you are having
with your services
By looking at the different responses from different
CNN Generated
models, it has been noticed that LSTM is generating better hi there is not have you are you
Response
sentences that make sense and it is hard to say if the response
is from a human or machine whereas GRU responses are not
as good as LSTM.
V. CONCLUSION
In this project, we build customer support chatbot that helps
companies to have 24 hours of automated responses. After
analyzing the dataset and understanding the importance to have
automated responses to customers and companies, we start
exploring existing techniques used for generating responses in
the customer service field. Then, we attempt to try three different
models LSTM, GRU, and CNN. The experimental results show
that LSTM and GRU models(with modified parameters) tend to
generate more informative and valuable responses compared to
CNN model and the baseline model LSTM. Besides, we used a
BLEU score and cosine similarity as evaluation measures to
support the final decision.

REFERENCES
Fig. 11. The Cosine Similarity Scores. [1] M. Chung, E. Ko, H. Joung, and S. J. Kim, “Chatbot e-service and
customer satisfaction regarding luxury brands,” J. Bus. Res., Nov. 2018.
TABLE. VI. EXAMPLE OF EMOTIONAL QUERY AND RESPONSES FROM ALL
[2] J. Hill, W. Ford, I. F.-C. in H. Behavior, and undefined 2015, “Real
MODELS
conversations with artificial intelligence: A comparison between human–
human online conversations and human–chatbot conversations,” Elsevier.
my package is days late and i am leaving [3] J. Hirschberg and C. D. Manning, “Advances in natural language
Customer Query tomorrow on holidays could you please help it processing,” Science (80-. )., vol. 349, no. 6245, pp. 261–266, Jul. 2015.
is extremely
[4] A. Xu, Z. Liu, Y. Guo, V. Sinha, and R. Akkiraju, “A New Chatbot for
Customer Service on Social Media,” in Proceedings of the 2017 CHI
Customer Support sorry to hear this please dm us your tracking Conference on Human Factors in Computing Systems - CHI ‟17, 2017,
Response and phone number pp. 3506–3510.

i am sorry for the trouble with your order [5] S. Oraby, P. Gundecha, J. Mahmud, M. Bhuiyan, and R. Akkiraju,
LSTM Generated
please report this to our support team here and “Modeling Twitter Customer ServiceConversations Using Fine-Grained
Response
we will check this Dialogue Acts,” in Proceedings of the 22nd International Conference on
Intelligent User Interfaces - IUI ‟17, 2017, pp. 343–355.
[6] H. Shah, K. Warwick, J. Vallverdú, and D. Wu, “Can machines talk?
Comparison of Eliza with modern dialogue systems,” Comput. Human
Behav., vol. 58, pp. 278–295, May 2016.

8|Page
[7] R. DALE, “The return of the chatbots,” Nat. Lang. Eng., vol. 22, no. 05,
pp. 811–817, Sep. 2016.
[8] M. Coronado, C. A. Iglesias, Á. Carrera, and A. Mardomingo, “A
cognitive assistant for learning java featuring social dialogue,” Int. J.
Hum. Comput. Stud., vol. 117, pp. 55–67, Sep. 2018.
[9] S. Jha, S. Bagaria, L. Karthikey, U. Satsangi, and S. Thota,
“STUDENT INFORMATION AI CHATBOT,” in International
Journal of Advanced Research in Computer Science, 2018, vol. 9, no.
3.

9|Page
10 | P a g e