CHATBOT INTERACTION IN E-COMMERCE

USING FUZZY LOGIC WITH NLP

Dhruv Sangal Vishal Nishad Nikhil Kumar
CSE (DATA SCIENCE) CSE (DATA SCIENCE) CSE (DATA SCIENCE)
G.L. Bajaj Institute of Technology and G.L. Bajaj Institute of Technology and G.L. Bajaj Institute of Technology and
Management Management Management
[email protected] [email protected] [email protected]

Abstract—Computer have made everyone’s life easier.

The most important part of computer science today is the II. LITERATURE REVIEW
ability to understand technology and automate it to save time,
which are important skill for life. SSS chatbot is a program Recommendations are important in helping customers
designed to answer the most common question people have find products of interest in web-based transactions (such
in their minds. This article is only taken from text chatbot. as e-commerce). Therefore, with the emerging need for a
SSS chatbot validates user input and accesses data to provide new approval process that is more accurate, more precise,
predefined validation using standard comparisons with the and focused on customer needs, researchers have chosen
SSS database. For example, if the user provides the bot with to use fuzzy logic as a possible method to solve the bias in
a statement like “What is the supported email address?” FAQ user behavior and preference. This section discusses
chatbot can usually be answered by looking at the database- previous research on e-commerce recommendations,
“This [email protected] When the entry is entered into especially in fuzzy logic, to increase efficiency and
the database, the response is given personally using the pre- effectiveness. The use in recommendations also implies
FAQ chatbot model, the order is the entire sentence is the ability to improve recommendations. Since the
recognized and the response model is the appropriate model combination of fuzzy sets and rules can confuse and make
of known sentence hash events. The answer to the questions customers indecisive, the application of fuzzy logic in the
are based on the depth and volume of knowledge-[1]-chatbot decision-making process improves recommendations,
is a new tool that will be found in many places in the future especially in unstable e-commerce situations [1]. This
of chatbot. This article covers the process of creating and method corrects some of the shortcomings of traditional
using FAQ chatbots. recommendations based on explicit instructions. (2021)
presented the essence of fuzzy logic methods in e-
commerce recommendation systems. They stated that
I. INTRODUCTION fuzzy logic is suitable for systems with a large number of
users because it can be adapted to fuzzy decision making.
In recent years, e-commerce has become a way of eating by growing This shows that combining fuzzy logic machines with
every year. As the variety and number of products offered in the other technologies such as collaborative filtering
online research environment increase, consumer often have machines, content-based filtering tools can lead to better
difficulty finding products that suit their tastes. A good way to accuracy in terms of user satisfaction. This integration
provide service. These systems can create recommended lists that helps make recommendations more dynamic and
will increase customer satisfaction by processing information such adaptable to the needs of different users at different times
as users’ access frequency and past purchase history. Minimum [3]. Good recommendations and collaboration emerged
information; The changes in the conditions required for the by filtering. The system uses fuzzy inference to improve
provision of approval services with the development of e- video recommendations produced by traditional filtering
commerce are also valid for the development of approval standards. techniques. Therefore, using fuzzy logic, the system can
Capturing people’s decisions that are open to interpretation. How make the desired recommendations correctly even if some
ever the global design model has some important problems that can information is omitted. Many authors have demonstrated
be overcome by using fuzzy logic, because it will be able to manage this fact by experimenting with theoretical simulations of
errors and different language models. A new modeling approach the described hybrid process and comparing them with the
that uses fuzzy logic to develop recommendations in e-commerce. simple identification process and the real truth [4]. Zhang,
By providing fuzzy reasoning capabilities to the conceptual system, Cui, and Ling (2018) studied the application of fuzzy logic
it can learn and manipulate uncertain preferences and create in e-commerce recommendations, especially with regard
dynamic interfaces based on preferences. Changing patterns in user to the uncertainty of user ratings and preferences. Their
behavior provide direct recommendations based on binary method combines fuzzy clustering with fuzzy
decisions. collaborative filtering to enable the system to capture
The second part of this study addresses the problem of similar customer preferences and thus improve recommendations
studies and draws attention to the inadequacy of the current system.
This article describes the planning process as well as experimental
design and performance evaluation according The to
traditional methods. results of the study show that the use of fuzzy system logic helps t o improve the
 performance of the overall model, as it expands the probability of
reaching correct predictions without affecting user satisfaction, a Fuzzification: By defining the membership function, many open
feature that can be used in large e-commerce sites with different concepts such as user ratings and product features are converted into
information. and confusion is an important problem that cannot be fuzzy loss. This member function is shown for keywords such as "low",
avoided.
"medium" and "high". For example, the user's conversion rate will be
changed to the word "medium-high" / on the contrary, the product type
III. METHODOLOGY will be created as "electronics", "clothing" etc.

There are several stages in the development of fuzzy logic as a strategic Fuzzy knowledge base: A summary of fuzzy rules generated by expert
approach to e-commerce offerings: data capture, FIS development, knowledge (or patterns found in historical data). For example, a rule
integration with unified filtering sharing, and evaluation system. The might say: "If the user likes electronics (and has a medium budget), then
process used in this guide is compared to fuzzy logic as accepted by the
person in the e-business environment. The main steps of the plan are recommend electronics in medium cost Inference Engine: A fuzzy
discussed below. inference engine (that processes fuzzy input) based on fuzzy output rules.
This engine evaluates the rules based on fuzzy inputs (user preferences,
1. Exploratory Data Analysis: products, etc.) and results in negative objects representing the strength of
The first phase, expected to be completed in the first quarter, will collect the words beautiful.
e-commerce data such as browsing history, purchase data, reviews
received if any, age, and gender from platform users. This data is then pre- Defuzzification: The fuzzy output is then converted into intelligent
cleaned to remove noise, null values, and variations. Feature scaling is to
results that make it recognizable and generate recommendations for the
ensure that all variables have the same scale so that fuzzy inference
systems (FIS) can better interpret data such as object measurements and customer. The defuzzification mechanism converts the strength of the
preferences. fuzzy propositions into meaningful numbers that represent the reliability
(or truth) of each element to the user. The fuzzy logic approach is good
because it resolves the uncertainty and ambiguity in the customer's
preferences and work equipment; but it also makes the process more
acceptable.

3. Hybrid with collaborative filtering:

To improve the accuracy (and individuality) of recommendations, a
hybrid approach (combining fuzzy logic with collaborative filtering) is
adopted. Collaborative filtering uses a matrix of user interactions (such as
ratings or clicks) to identify similar users and recommend products based
on their interests using other users’ likes.
One of the most significant challenges in data collection is determining the
factors that drive user choices. These factors can include:
By customers: Identify similar users based on preferences (ratings,
User behavior data: web browsing history data, click activity, and survey
comments, etc.) and create recommendations based on products liked by
questions. Age; address, etc. Customer reviews and various ratings.
similar users. However, this approach also has its limitations. Although it
is very useful for users, it sometimes blocks the view of available options.
2. Application of Fuzzy Logic Modeling to User Preferences:
Therefore, other factors should be taken into account in the approval
FIS is introduced to use fuzzy rules to mimic the user’s behavior based on
process (thus providing more choices).
their preferences and interests. FIS has the following components:

Project-based collaboration: Project-based collaborative filtering

(IBCF) identifies similar products based on the impact of user
interactions, instead of focusing on similar users; The combination of
fuzzy logic and collaborative filtering makes it possible to deal with
situations where user preferences are vague, incomplete or ambiguous (as
demonstrated in previous works or [1], [2], [4]). Fuzzy logic systems
improve the effectiveness of collaborative filtering models by correcting
uncertainties in customer preferences and product features. However, this
approach increases the overall effectiveness of the recommendation
process because it adapts to the complexity of the user's behavior.

Cosine similarity: Cosine similarity is often used to calculate the

similarity between users or objects. Cosine similarity measures the cosine
of the angle between two vectors (in this case, user ratings or product
features).
Combination compatibility: Add the similarity calculated by cosine
similarity and adjust the consensus score according to the similarity of
users or products. Aggregation can be done using the weighted average V. FUTURE SCOPE
method:
Chatbots are intelligent tools that interact with users through words, text,
4. In order to evaluate the effectiveness of the fuzzy logic based or voice. Chatbots can work in areas such as marketing, payments and
recommendation, various evaluation methods are adopted (including but operations, and services. Today, customers need 24/7 service in many
not limited to): areas, from banking to finance, health to wellness. Therefore, companies
are looking for ways to quickly create chatbots and virtual assistants to
answer questions that customers may have at any time of the day.
Precision and Recall: These metrics measure the accuracy of However, companies should not forget that chatbots can better direct
recommendations by measuring the proportion of products that users effective conversations to human representatives; they need to be able to
interact with or buy. Precision shows the number of relevant items in the solve customer requests. As more businesses and consumers use
offer, while regression checks how well the offer includes relevant Chatbots, the need to develop better Chatbots will also increase, making
it easier for companies to use them in their business. Data training is a
items. thus showing the accuracy of the recommendation.
major barrier to entry for businesses looking for chatbots. This is
especially true for more advanced chatbots that try to understand the
User satisfaction: Research and strategy development to determine user target and respond in human-like language. Language and
experience and satisfaction with personalized recommendations are communication skills must be developed.
essential to measure the effectiveness of recommendations. Differences Good robot language is not easy and difficult
in recommendations were also analyzed; different recommendations
International Journal of Computer Science and Engineering Vol. often
will increase the likelihood of discovering new products and encourage less than expected. The user experience should be consistent, clear, and
customer interaction. focused on the behaviors that enable and facilitate
people to interact. The machine learning system does not have the same
5. Implementation and evaluation of the system behavior because all the interactive responses are a combination of text
The implementation of fuzzy logic-based recommendations is carried from different sources.
out through software processes (such as Python, MATLAB) and data
processing, which are converted into a fuzzy logic model. The system is V1. CONCLUSION
tested on data obtained from real e-commerce, thus ensuring that the
proposed fuzzy logic approach remains flexible and adaptable.
Techniques such as content-based filtering and integrated filtering are Chatbot is a program that allows people to chat. It was designed to be the
combined; Multiple tests are performed to evaluate the system under ultimate virtual assistant. Chatbots Special FAQ Chatbots have become
different conditions, including sparse data, changing customer popular among the business world today due to their ability to reduce
preferences, and various features. However, the complexity of these customer service costs and manage multiple users simultaneously. In this
situations increases the need for good modeling techniques, as the article, we present the design of a FAQ chatbot that provides effective and
complexities involved can affect the quality of the work. accurate answers to all questions based on the FAQ dataset. We have
looked at the best ways to develop a simple chatbot.
IV. RESULT AND ANALYSIS The scheme is a simple and automated solution used to transfer data from
the computer without having to think about finding the necessary terms in
FAQ Chatbot is designed to be very simple and user-friendly. It is not as the search or checking the number of web pages to be collected. Messages
complex as other Chatbots. This FAQ Chatbot uses simple comparison
that allow organizations to handle many customers at the same time. By
patterns to represent input and output, while other
Chatbots use input rules, keyword patterns and output > code to generate using chatbots that complement human agents, you can save on personnel
the response. If no input is found in the file, a canned response is costs and prevent problems caused by human error. Users can easily write
generated. The input and output can be modified by the user. Depending queries and record data in natural language. This article provides
on the developer or user, requests and responses should be stored in a information on creating and using a simple, user-friendly, and interactive
repository. If the customer really needs human assistance, the Chatbot can FAQ chatbot.
also send them to a human support representative via email, as shown in
Figure 2 below. The FAQ function can be completed by clicking on the In this context, the general features that make fuzzy logic techniques
field. effective are: · Ability to adapt to uncertainty and changes in user
preferences. Therefore, it can work with a small number of features and
fuzzy information, which increases the accuracy, precision and diversity of
recommendations compared to traditional methods. However, there are still
areas where further progress can be made. Future efforts can expand the
question by improving the quality of the system, making it more capable
and able to respond faster to online big data platforms. In addition, the
integration of the use of fuzzy systems with other state-of-the-art machine
learning methods such as deep learning or hybrid models will increase the
value of recommendations. It shows that fuzzy logic operations can
improve the performance of recommendations in digital environments and
increase customer satisfaction; however, it will be particularly useful in
terms of improving business results in the e-commerce environment.
Although these methods are promising, their use should be considered
carefully as they can lead to problems. Therefore, careful analysis of the
relevant factors is necessary.
 VII. REFERENCES

[1] R. S. Russell, “Language Use, Personality and True Conversational

Interfaces”, Project Report of AI and CS University of Edinburgh,
Edinburgh, pp.1-80, 2002.

[2] Y. Zhou, X. Ziyu, A. W. Black, A. I. Rudnicky, “Chatbot

Evaluation and Database Expansion via Crowdsourcing”, Proc. Of the
Chatbot Workshop of LREC, US, pp. 16-19, 2016.

[3] C. R. Anik, C. Jacob, A. Mohanan, “A Survey on Web Based

Conversational Bot Design”, JETIR, Vol.3, Issue.10, pp. 96-99, 2016.

[4] J. Jia, “The Study of the Application of a Keywords-based Chatbot

System on the Teaching of Foreign Languages”, Report of University
of Augsburg, Augsburg, pp.1-36, 2003.

[5] H. L. Wang and T. L. Zhang, "Fuzzy Logic in Product

Recommendation Systems: A Survey and Future Trends,"
International Journal of Computer Applications, vol. 145, no. 7, pp.
88-93, Nov. 2020.

[6] E. H. Wu, C. Lin, Y. Ou, C. Liu, W. Wang and C. Chao,

"Advantages and Constraints of a Hybrid Model K-12 E-Learning
Assistant Chatbot," in IEEE Access, vol. 8, pp., 2020.

[7] Boris Galitsky, Dmitry Ilvovsky "Chatbot with a Discourse

Structure-Driven Dialogue Management", Association for
Computational Linguistics, 2017.

[8] L. Zhang, Y. Yang, J. Zhou, C. Chen and L. He, "Retrieval-

Polished Response Generation for Chatbot," in IEEE Access, vol.
8,pp. 2020.

[9] Mohawesh, Rami, et al. "Fake Reviews Detection: A Survey."

IEEE Access 9 (2021): 65771-65802.

[10] Hassan, Rakibul, and MdRabiul Islam. "A Supervised Machine

Learning Approach to Detect Fake Online Reviews." 2020 23rd
International Conference on Computer and Information Technology
(ICCIT). IEEE, 2020.