METHOMIKA: Jurnal Manajemen Informatika & Komputerisasi Akuntansi Vol.
8 No.
2 (Oktober 2.
ISSN: 2598-8565 .
edia ceta.
ISSN: 2620-4339 .
edia onlin.
DESIGNING A WEB-BASED LIGHT NOVEL APPLICATION WITH AN LLMPOWERED CHATBOT RECOMMENDATION SYSTEM USING SCRUM
METHODOLOGY
Yefta ChristianA.
Mangapul Siahaan.
Hansvirgo Jurusan Sistem Informasi.
Universitas Internasional Batam.
Indonesia Email: yefta@uib.
DOI: https://doi.
org/10.
46880/jmika.
Vol8No2.
ABSTRACT
In the era of the internetAos exponential growth, readers are often overwhelmed by the plethora of books available, particularly in the genre of light novels.
This research aims to address this issue by developing a recommendation system for light novels, utilizing a chatbot interface.
The methodology employed follows the Borg and Gall model, with a focus on research, information collection, planning, and development stages.
The research stage involved the use of questionnaires to gather data and analyze the parameters to be used in the recommendation system.
The development stage was carried out using the Scrum methodology and the Retrieval Augmented Generation (RAG) approach for the chatbotAos functionality.
The outcome of this study is a web-based online light novel application and featuring a chatbot conversational recommender system.
Through this system, users can access and read light novels online, while also utilizing the chatbot to request novel recommendations.
The research findings demonstrate the successful integration of Large Language Model (LLM) technology into the web-based light novel application.
The Scrum development approach facilitated efficient system creation, and the RAG-based chatbots are seen as successful in producing recommendations that match user queries based on existing Recommendation results are obtained from semantic search and from the ranking vector with the highest score.
Keyword: Chatbot.
Conversational Recommender System.
Vector Database.
Retrieval Augmented Generation.
Scrum.
INTRODUCTION
The rapid growth of the internet has led to an explosion of digital content, making it challenging for readers to discover new and relevant books to read (Sarma et al.
, 2.
, such as light novels.
Light novels, closely associated with manga, are short prose novels that feature many of the same themes, tropes, and character archetypes often found in manga.
They are quick, pulpy reads meant to entertain and provide a welcome escape from reality.
However, the vast amount of available content can be overwhelming for readers, especially those new to the genre (Hersani et , 2022.
Rosyad et al.
, 2.
This necessitates the development of effective recommendation systems to help readers discover light novels that align with their interests (Intan Hervianda Putri et al.
, 2.
Recommendation systems in general often utilize content-based filtering and collaborative filtering methods to provide suggestions to users (Akbar et al.
, 2023.
Rosita et al.
, 2022.
Rosyad et al.
Content-based filtering recommends items to users based on the similarity of items, while collaborative filtering generates lists of items similar to the user's preferences.
However, these approaches have associated limitations (Arfisko & Wibowo, 2022.
Hakim & Baizal, 2.
For instance, content-based filtering may not offer recommendations to new users, while collaborative filtering can encounter cold start problems for new items without ratings (Hakim & Baizal, 2022.
Hui et al.
, 2022.
Zhang et al.
, 2.
address these problems.
Conversational Recommender System (CRS) has been proposed (Fajari & Baizal.
Recommendations from CRS are based on user preferences, requiring interaction between the user and the system, which can be handled by a chatbot (Fadhlullah et al.
, 2.
Technological developments have played an important role in the evolution of Large Language Model (LLM) and chatbots powered by LLM.
With the Large Language Model (LLM), chatbots are able to recognize, create, translate, or summarize human text with increasingly sophisticated capabilities.
Chatbots built with LLM also perform well and can focus on specific domains (Mansurova et al.
, 2.
This can help in developing a Conversational Recommender Halaman 174 METHOMIKA: Jurnal Manajemen Informatika & Komputerisasi Akuntansi Vol.
8 No.
2 (Oktober 2.
System chatbot that can provide recommendation results that match user preferences (Feng et al.
, 2.
Research conducted by (Hakim & Baizal, 2.
using knowledge-based chatbot for museum recommendation system in Jakarta.
The chatbot was designed to understand user messages and provide museum recommendation results that matched user Further research by (Sari et al.
, 2.
focused on designing a chatbot with supervised learning for tourism recommendations in Central Java using the extreme programming method.
The results indicate that the chatbot runs well and provides tourist recommendations and information to tourists.
another study, (Hersani et al.
, 2.
designed a webbased book recommendation information system using extreme programming methods.
The research resulted in a Share Ur Book website, which can be used to search for books and book recommendations.
Additionally (Mansurova et al.
, 2.
conducted research on designing an LLM-powered answering question chatbot in a special domain, namely The chatbot was developed using the Retrieval Augmented Generation (RAG) method, which utilizes data stored in a vector database to avoid the need for LLM retraining.
The resulting chatbot has promising performance improvements in specialized knowledge-intensive Lastly, (Kurniawati, 2.
focused on implementing scrum in the design of mobile applications for coffee The research states that scrum can facilitate the adoption of any changes and help produce quality products as needed.
Despite the advancements in chatbot technology and the promising results of various studies, there is still a significant gap in the field of recommendation Specifically, there is a lack of development in chatbots for recommendation systems that utilize Large Language Models (LLM).
While LLM have shown great potential in understanding and generating humanlike text, their application in recommendation systems, particularly in the context of light novels, remains largely unexplored.
This presents a unique challenge and opportunity for researchers and developers to innovate and push the boundaries of whatAos possible with LLM-powered recommendation systems.
Based on the description above, the author plans to design a web-based light novel application that incorporates a recommendation system with an LLMpowered chatbot.
The application will be developed using the scrum method, and the chatbot will be ISSN: 2598-8565 .
edia ceta.
ISSN: 2620-4339 .
edia onlin.
integrated with RAG.
This initiative aims to not only increase interest in reading but also provide tailored light novel recommendations to the public.
The research will contribute to the existing literature by proposing a novel approach that leverages advanced technologies to enhance the book discovery process and promote reading habits.
Additionally, the study will fill a gap in the research by exploring the potential of LLM-powered chatbots in the development of conversational recommender systems for light novels, addressing the limitations of existing methods and providing a more interactive and user-friendly experience for readers.
RESEARCH METHODOLOGY
In the research conducted, a multi-method approach was employed, integrating both quantitative and applied research methodologies.
The quantitative aspect centered on identifying parameters for the recommendation system, while the applied research phase entailed utilizing the Scrum method for the development project after parameter.
Quantitative Research At this stage, research will be carried out by distributing questionnaires with the aim of identifying the parameters that influence readers in choosing light novels to read.
A questionnaire is a written statement consisting of a set of questions or written statements used to obtain information from respondents in the form of a report about a person or things they are familiar with.
The questions are as follows:
How much do you consider the author's reputation when selecting a light novel? How much do you rely on reviews and recommendations from others when choosing a light novel? How much do you consider the genre when selecting a light novel to read? How much do you consider the length of the light novel when selecting a light novel to read? How much do you consider the popularity when selecting a light novel to read? The analysis of this study will involve the use of percentages and frequencies, which can be calculated using the formula .
below (Shilfani & Limbongan, yce ycy = ycu y 100 a .
Halaman 175 METHOMIKA: Jurnal Manajemen Informatika & Komputerisasi Akuntansi Vol.
8 No.
2 (Oktober 2.
P = Percentage of respondents f = frequency of respondents who voted n = total respondents Applied Research System Development The light novel web system development method used in this study is Scrum framework.
Scrum is an agile project development method that is used by teams to collaboratively develop projects in short Figure 1.
Scrum Agile The implementation of the Scrum method is divided into several stages:
Scrum Team The following is the scrum team structure that has been prepared to carry out project development.
Table 1.
Scrum Team Responsibility Analyze project Create and compile the Product Owner product backlog Review project development results at each sprint Monitor and provide guidance on the implementation of Scrum Scrum Master to the team Monitor the progress of the product backlog Carrying out the project Development development process Team Testing project development results Role Product backlog:
ISSN: 2598-8565 .
edia ceta.
ISSN: 2620-4339 .
edia onlin.
A prioritized list of tasks that need to be completed by the team during the project development phase.
The list contains a brief description of all the features desired in the product, their sequence, estimated time, and their own value.
The main features to be designed include:
Table 2.
Product Backlog Issue User Story As the owner.
I aim to establish a Use Case Design Use Case Diagram (UCD) and EntityDiagram & Entity Relationship Diagram (ERD) Relationship as the fundamental Diagram framework for system As the owner.
I desire to implement basic authentication within the User Login Authentication Register Logout View User Profile Edit User Profile Change Password As an Admin.
I want to have light novel management at the system, so I can:
See all list novels.
Add new novel.
Light Novel Edit novel Management Delete novel See all list chapters of Add chapter Edit chapter Delete chapter As a user.
I want to find the Light Novels Page novel and choose which & Chapter Page chapter to read.
As a user.
I would like to bookmark light novels that I find interesting, enabling me to merely open my Bookmark Page bookmarks whenever using the system again to continue reading the desired novels without having to search Halaman 176 METHOMIKA: Jurnal Manajemen Informatika & Komputerisasi Akuntansi Vol.
8 No.
2 (Oktober 2.
Light Novel Recommendation through the novel list page As a user.
I desire a recommendation system that can suggest light novels to Sprint planning:
During sprint planning, the team collaborated to define what could be delivered and how it would be The product backlog was reviewed and refined for clarity, and the team decided on the sprint goal and how to achieve it.
The event also included breaking down the product backlog into smaller tasks to be delivered during the sprint.
At this point, the author sets one sprint for 2 weeks, and the project design will span over 5 Here is a summary of the sprint backlog that has been worked on during these 5 sprints.
Table 3.
Sprint Planning Jan
Feb
Mar
Sprint Backlog
1 2 3 4 1 2 3 4 1 2
Sprint 1 Designing UCD & ERD Designing Mindmap Sprint 2 Authentication Light Novel Management Chapter Management Sprint 3 Home Page Light Novels List Page Light Novel Detail Page Chapter Page Sprint 4 Edit Profile Change Password Bookmark Novel Sprint 5 Chatbot (CRS) Daily scrum:
ISSN: 2598-8565 .
edia ceta.
ISSN: 2620-4339 .
edia onlin.
A brief meeting where team members report on their progress, discuss plans, and identify obstacles to ensure they are on track to meet the sprint goal.
Sprint review:
In this stage, a review of the completed or remaining tasks is conducted with the supervising lecturer after each sprint.
Blackbox testing is also performed during this stage to test the project.
After the review, the product backlog is revised, and the priority of the tasks to be completed in the next sprint is determined.
Sprint retrospective:
In this stage, the author will ask for the supervising lecturer's assistance in checking what has been achieved in one sprint.
The purpose of this stage is to get feedback and suggestions from the supervising lecturer and to evaluate oneself in order to develop a work plan for the next sprint.
Increment:
The result of completing all the tasks in the product backlog.
Light Novel Recommendation Chatbot Development The development method used in this study is RAG (Retrieval Augmented Generatio.
vector database.
This method allows LLMs to dynamically provide context and reduce the need for manual updates, making the chatbot more efficient and effective in providing accurate information.
The LLM model used in this research is the model provided by OpenAI.
GPT3.
5 (Mansurova et al.
, 2.
The development stage involves the following Data Collection The researcher will aggregate light novel data from diverse sources using web scraping techniques.
The acquired data will be stored in an unstructured format, such as a PDF, containing information from light novels based on parameters obtained through a questionnaire.
Data Preprocessing The collected book data is processed and divided into chunks to create embeddings and store Halaman 177 METHOMIKA: Jurnal Manajemen Informatika & Komputerisasi Akuntansi Vol.
8 No.
2 (Oktober 2.
embeddings to a vector database using LangChain, a framework built around Large Language Models.
Figure 2.
Data Processing and Embedding Creation Integrating with RAG At this stage, the LLM model is integrated with the RAG vector database to enable the chatbot to search and filter relevant information.
Figure 3.
Integration Building the Chatbot with LLM After integrating the LLM model with the RAG vector database, the next step involved building the chatbot using LangChain's tools and integrations for web-based deployment.
This included using LLM to generate responses based on the retrieved The interface is implemented in Next.
js and uses the GPT-3.
5 model from the OpenAI API (Huang et al.
, 2023.
Mansurova et al.
Situmeang et al.
, 2.
Testing At this stage, the model is tested to check whether the resulting book recommendations match Black-box testing, a method of software testing that examines the functionality of an application without peering into its internal structures or workings, can be used for this purpose (Wijaya et al.
, 2.
HASIL DAN PEMBAHASAN
Quantitative Research ISSN: 2598-8565 .
edia ceta.
ISSN: 2620-4339 .
edia onlin.
In our research, we employed a questionnaire to determine which parameters influence and can be used for light novel recommendations.
The questionnaire was designed to gather data on readers' preferences and reading habits, as well as their opinions on various aspects of light novels.
The results of the questionnaire were analyzed to identify the most significant parameters that influence readers' decisions to read a light novel.
The questionnaire was distributed to a diverse group of 100 readers, including both casual and avid readers, to ensure a representative sample.
The results showed that several parameters significantly influenced readers' decisions to read a light novel.
Results can be seen in the table below.
Table 4.
Questionnaire Result Parameter Percentage Reason/Opinion Most prioritize quality and content over the fame of the author when it comes to Author light novels.
They believe that lesserknown authors can works that deserve Most believe that the genre plays a crucial Genre role in helping them discover light novels that align with their Most over the length of Length However, some find novels less engaging and lose interest in reading them.
Most believe that light Popularity novels with high ratings are indicative of good quality.
Halaman 178 METHOMIKA: Jurnal Manajemen Informatika & Komputerisasi Akuntansi Vol.
8 No.
2 (Oktober 2.
Synopsis These ratings reflect assessment of the public, making them a reliable measure of a novelAos worth.
Most It offers a glimpse into the significantly impacts the readerAos decision to dive into the ISSN: 2598-8565 .
edia ceta.
ISSN: 2620-4339 .
edia onlin.
perform CRUD operations on Light Novels and Chapters, as well as the capability to view Novels and Chapters.
Additionally, administrators can utilize the Chatbot for novel recommendations.
On the other hand, users have access to view Novels, view Chapters, and use the Chatbot for novel recommendations.
Entity Relationship Diagram Figure 5 shows the model used to design the database on the light novel website system.
Considering the findings, the author has opted to construct a dataset for the chatbot using key parameters such as genre, popularity, and synopsis.
This dataset will serve as the foundation for creating an effective conversational recommender system.
First Sprint During the initial sprint planning, the authorAos primary objective was to finalize the design of the Product Backlog, specifically focusing on the creation of the Use Case Diagram (UCD) and the Entity Relationship Diagram (ERD).
Subsequently, the outcomes of the first sprint review are as follows:
Use Case Diagram Figure 5.
Entity Relationship Diagram Sprint Review Within the confines of this iterative development cycle, the outcomes are systematically exhibited to the product proprietor and the collective team.
Subsequently, a consensus is reached to incorporate these outcomes into the forthcoming developmental endeavor.
Sprint Retrospective Table 5.
First Sprint Retrospective Result What went What went How to improve? All tasks are completed on time Figure 4.
Use Case Diagram According to Figure 4, administrators possess comprehensive system access, including the ability to Second Sprint During the second sprint planning, the team aimed to achieve two primary objectives:
Authentication product backlog and light novel management admin.
Subsequently, the outcomes of the second sprint review are as follows:
Login Page Halaman 179 METHOMIKA: Jurnal Manajemen Informatika & Komputerisasi Akuntansi Vol.
8 No.
2 (Oktober 2.
ISSN: 2598-8565 .
edia ceta.
ISSN: 2620-4339 .
edia onlin.
This web page serves as an interface for user or administrator authentication.
Figure 6.
Login Page Register Page This web Figure 9.
Chapter Management Page Figure 7.
Register Page Light Novel Management This web page provides administrative access to novel-related functionalities, including novel registration tracking, addition of new novels, and modification or removal of existing novel records.
Figure 8.
Light Novel Management Page Chapter Management This web page provides administrative access to novel-related functionalities, including tracking the number of chapters, adding new chapters, and modifying or removing existing chapter records.
Sprint Review Within this sprint, the generated backlog outcomes are systematically demonstrated to the product owner and end-users, with the resultant output deemed suitable and acknowledged.
In addition, an empirical method of system testing, known as black box testing, was executed.
The outcomes of this testing methodology are documented in the subsequent tabulation.
Table 6.
Second Sprint Blackbox Testing Result Testing Expected Result Result Users can access the User Login login page and log in Admin can access Admin Login the login page and OK log in Users can access the Register register page and OK create account Logout User can logout Admin can add, edit Light Novel and delete light OK Management Chapter Admin can add, edit Management and delete chapters Sprint Retrospective Table 7.
Second Sprint Retrospective Result What went What went How All tasks are The coding Improve the code completed on results are to make it clearer not and Halaman 180 METHOMIKA: Jurnal Manajemen Informatika & Komputerisasi Akuntansi Vol.
8 No.
2 (Oktober 2.
Third Sprint The third sprint planning aims to finalize the development of the product backlog light novel list page, implement the filtering feature, and enhance the light novel detail page.
Below are the outcomes of the review conducted during the third sprint.
Home Page This web page serves as the initial landing page for users upon opening the website.
ISSN: 2598-8565 .
edia ceta.
ISSN: 2620-4339 .
edia onlin.
Light Novel Detail Page This web page serves as a repository for detailed information about a selected light novel.
Additionally, it includes functionalities for bookmarking and rating light novels.
Figure 12.
Light Novel Detail Page Chapter Page This web page serves as a repository for the textual content of specifically chosen chapters within a light novel.
Figure 10.
Home Page Light Novel List Page This web page serves as a comprehensive repository of novels, complete with an array of userfriendly features.
Users can explore a curated list of novels and utilize powerful search and filtering functionalities to enhance their browsing Figure 13.
Chapter Page Sprint Review Within this sprint, the generated backlog outcomes are meticulously demonstrated to the product proprietor and end-users, with the resultant output deemed suitable and acknowledged.
Supplementary feedback, which may pertain to the refinement of coding practices, is also obtained and earmarked for future rectification and enhancement.
Furthermore, the outcomes of the empirical black box testing methodology are documented and can be observed in the subsequent tabular Figure 11.
Light Novel List Page Halaman 181 METHOMIKA: Jurnal Manajemen Informatika & Komputerisasi Akuntansi Vol.
8 No.
2 (Oktober 2.
Table 8.
Third Sprint Blackbox Testing Result Testing Expected Result Result Users can view and access the homepage Home Page as the main display OK when opening the Users can view the Light Novel light novel list and List Page search using the search and filter features Users can view the Light Novel details of a selected Detail Page light novel and see a list of existing chapters Users can view the Chapter content of the selected OK Page Sprint Retrospective Table 9.
Third Sprint Retrospective Result What went What went How All tasks are The coding Improve the code completed on results are to make it clearer not and Fourth Sprint The fourth sprint planning aims to finalize the development of the product backlog pages for user profile editing, password modification, and bookmarked novel management.
Below are the outcomes of the review conducted during the fourth Edit Profile Page This web page provides users with a form to update and modify their profile information.
ISSN: 2598-8565 .
edia ceta.
ISSN: 2620-4339 .
edia onlin.
Change Password This web page provides users with a form to modify their account password.
Figure 15.
Change Password Page Bookmarked Novel Page This web page compiles a catalog of novels that users have marked as bookmarks.
Figure 16.
Bookmarked Novel Page Sprint Review During this sprint, the backlog outcomes that have been produced are methodically showcased to both the product owner and end-users.
The resulting output is considered appropriate and accepted.
Furthermore, an empirical system testing approach, specifically black box testing, was implemented.
The results of this testing methodology have been meticulously recorded in the subsequent tabular Table 10.
Fourth Sprint Blackbox Testing Result Testing Expected Result Result Users can access the edit profile Edit Profile page and change OK Page or update their Users can update Change or change their OK Password Figure 14.
Edit Profile Page Halaman 182 METHOMIKA: Jurnal Manajemen Informatika & Komputerisasi Akuntansi Vol.
8 No.
2 (Oktober 2.
Bookmarked Novel Page Users can see a list light novels ISSN: 2598-8565 .
edia ceta.
ISSN: 2620-4339 .
edia onlin.
Sprint Retrospective Table 11.
Fourth Sprint Retrospective Result What went What went How to improve? All tasks are The coding Improve the code completed on results are to make it clearer not and Fifth Sprint The fifth sprint planning aims to finalize the chatbot product backlog related to the recommendation Below are the outcomes of the review conducted during the fifth sprint.
Recommendation System Design Result Chatbot design for recommendation systems using the RAG method.
The recommendation system flow design is in Figure 21 below.
Figure 18.
Light Novel Raw Data The .
txt file will be processed into numbers for embeddings and divided into chunks to be stored in a vector database.
The database used at this stage is the pinecone vector database.
Figure 19.
Vector Data Preparation Code Snippet Figure 17.
Recommendation System Design Result First, users will ask questions or ask for recommendations from the chatbot.
From user requests, the chatbot recommendation system that is built will filter and embeddings in the vector database to search for related ones and then return them as a response to the user.
Figure 20.
Presentation of The Processed Data.
Pinecone Vector Database The dataset used in the study is collected based on the results of a questionnaire.
The dataset used comes from web scrapping results from various sources which were initially saved in the form of a txt file.
Figure 21.
Transformation of Data into Vector Format.
Halaman 183 METHOMIKA: Jurnal Manajemen Informatika & Komputerisasi Akuntansi Vol.
8 No.
2 (Oktober 2.
ISSN: 2598-8565 .
edia ceta.
ISSN: 2620-4339 .
edia onlin.
Chatbot Integration At this juncture, the chatbot is web-based and seamlessly integrated with LLM (Language Mode.
and vector databases.
The designed chatbot interface is visible below.
Figure 24.
Vector Ranking Result Figure 22.
Chatbot Interface The developed chatbot has been seamlessly integrated with LLM and vector database, enabling it to generate personalized recommendations based on user queries.
The recommendation outcomes are presented below.
Figure 22 delineates the ranking procedure of the recommendation outcomes depicted in Figure The recommendations proffered possess the highest scores, specifically 0.
5520, 0.
5518, and 5447, which closely align with the userAos query.
In the depicted scenario, the conversational recommendation system offers personalized suggestions to users based on their specific requests and individual preferences.
Furthermore, users have the capability to directly inquire about additional information or seek follow-up recommendations in response to the systemAos initial answers.
Figure 25.
Follow-up Questions Figure 23.
Recommendation Result In the current phase of the process, recommendations are generated by leveraging semantic search outcomes.
These outcomes are subsequently organized in a vectorial hierarchy, utilizing vector data procured from the Pinecone The recommendation output comprises vector data that exhibits the maximum value or score, aligning with the userAos query.
This methodology ensures the provision of the most relevant and high-quality recommendations.
Figure 26.
Follow-up Chat History Payload In cases where user inquiries fall outside the domain of the chatbotAos expertise, the system will appropriately respond with AoI donAot knowAo.
Halaman 184 METHOMIKA: Jurnal Manajemen Informatika & Komputerisasi Akuntansi Vol.
8 No.
2 (Oktober 2.
Figure 27.
Chatbot Response When Encountering Unknown Queries This implementation was constructed utilizing a promptTemplate, which facilitates the creation of a more specialized model, specifically tailored for the objective of book recommendations.
Figure 28.
Prompt Template Implementation Sprint Review During this sprint, the LLM-based chatbot was seamlessly integrated with the vector database and successfully deployed on the designed light novel Additional feedback from the sprint review includes addressing code inefficiencies and augmenting the chatbotAos knowledge by incorporating additional light novel data.
The outcomes of the empirical black box testing are documented and can be observed in the subsequent tabular representation.
Table 12.
Fifth Sprint Blackbox Testing Result Testing Expected Result Result Users can chat with Chatbot the chatbot and get Recommendation CONCLUSION In summary, this research aimed to enhance website development by integrating it with large language model (LLM) technology.
The outcome of this study is a web-based online light novel application, developed using the Scrum methodology, and featuring an LLM-based chatbot conversational recommender system built using the RAG method.
Through this system, users can access and read light novels online, while also utilizing the chatbot to request novel The research findings demonstrate the successful integration of LLM technology and vector databases into a web-based light novel application.
The Scrum development approach facilitates the creation of ISSN: 2598-8565 .
edia ceta.
ISSN: 2620-4339 .
edia onlin.
efficient systems, and RAG-based chatbots are seen as successful in producing recommendations that match user queries based on existing knowledge.
Recommendation results are obtained from semantic search and from the ranking vector with the highest Despite the achievements, certain limitations For instance, the chatbotAos knowledge is constrained by its training data, and it may respond with AoI donAot knowAo when faced with queries beyond its Additionally, the systemAos performance may vary based on user interactions and novel availability.
To advance this field, future research efforts could conduct research that combines fine-tuning and RAG (Retrieval-Augmented Generatio.
techniques in producing chatbots with better performance.
Apart from that, research can also be carried out on developing a recommendation chatbot that combines internet live search features which have the potential to prevent problems with limited data.
DAFTAR PUSTAKA