--- title: "Encouraging Recycling in Bangi Selatan Through a Content-Based Filtering Web Application" source: "Encouraging Recycling in Bangi Selatan Through a Content-Based Filtering Web Application.pdf" converted: "2026-10-01 10:30:58 +0800" converter: "MarkForge (mdconvert)" engine: "PyMuPDF4LLM (9 page(s), OCR on 9)" conversion_warnings: - "No text layer on page(s) 1, 2, 3, 4, 5, 6, 7, 8, 9: text came from OCR (Tesseract, English) - verify identifiers, numbers and symbols before trusting them." --- Available online at ‘nttps://jcrinn.cora/ Journal of hittps://crinn.conferencehunter.com/ Computing Research and Journal of Computing Research and Innovation 10(1) 2025 Innovation gCRINN hittps://crinn.conferencehunter.com/
a
# Encouraging Recycling in Bangi Selatan Through a ContentBased Filtering Web Application Muhammad Akmal Ahmad Nawawi!, Tajul Rosli Razak! 12 College of Computing, Informatics, and Mathematics, Universiti Teknologi MARA, 40450 Shah Alam, Malaysia ## ARTICLE INFO ## ABSTRACT Articlehistory’ tober 2024 This study addresses the challenges faced by residents of Bangi Selatan Revised 31 December 2004 in adopting 3R (Reduce, Reuse, Recycle) practices, primarily due to a ‘Accepted 10 January 2025 lack of interest in conservation efforts and insufficient awareness of Online first recycling’s importance. To address these challenges, we presented a web Published 1 March 2025 application that enhances recycling adoption by delivering personalized TT content recommendations. The key contributions of this study include Keywords:Content based dati Bites thebased development filtering (CBF) ofa novel with improved recommendation accuracy system b through **a** sedmodified on cont T **e** nt-rm Information retrieval — Frequency-Inverse Document Frequency (TF-IDF) formula. We Recycling awareness compare various recommendation techniques, including collaborative ; " + DOI: and hybrid filtering, and demonstrate how CBF effectively improves 1024191 jcrinn.v10i1 user engagement with recycling content. Our methodology involves advanced text vectorization and cosine similarity for precise content matching. User acceptance testing confirms the system’s effectiveness in increasing user engagement with relevant recycling information. This study highlights the potential of personalized recommendation systems in promoting environmental conservation and provides a foundation for future enhancements in recycling initiatives. ### 1. ## INTRODUCTION Artificial Intelligence (AI) has transformed various sectors with models like ChatGPT and BardAI, showcasing large language models' power to enhance global efficiency (Gill et al., 2024). However, despite these technological strides, the issue of proper solid waste management remains pressing, especially in developing countries. Global waste production is expected to rise from 1.3 billion tonnes to over 2.2 billion by 2025 (David et al., 2019). Despite 80% of this waste being recyclable, much is in landfills due to poor waste management practices (Zaharudin et al., 2023). > Corresponding author. E-mail address: tajulrosli@uitm.edu.my ‘attps://do.org/10.2419 1 jesian v10i1 510 Authors, 2025 219 Ahmad Nawawi & Razak. / Journal of Computing Research and Innovation (2025) Vol. 10, No. 1 In Malaysia, the recycling rate stands at 33.7%, proving that recycling practices remain low (Tiew et al., 2019). Based on the survey conducted in the Bangi Selatan area, 52.9% of respondents do not recycle because they were not encouraged due to a lack of exposure. The residents believed that traditional recycling campaigns were insufficient in promoting recycling awareness because they lacked engagement with recycling education. To address these gaps, leveraging a content-based filtering (CBF) system can be pivotal in encouraging recycling by personalizing recommendations based on user preferences and interests.. Such systems make them ideal for promoting recycling by delivering targeted information on recyclable materials, which can significantly enhance user engagement and foster sustainable recycling habits. This study proposes the development of a CBFweb application designed to enhance recycling exposure in Bangi Selatan. The system utilizes personalized recommendations powered by cosine similarity and the TF-IDF (Term Frequency-Inverse Document Frequency) algorithm. To tailor content to user interests, a vital feature of the application is a search box that enables users to retrieve relevant information from the database. Raising awareness and improving local recycling rates could be a model for integrating technology and environmental sustainability through user-centered solutions. ### 2. LITERATURE This section overviews the factors behind recycling awareness shortfalls, the opinions of Bangi Selatan locals, the development of web applications and the recommendation system. ### 2.1 Factors Behind Recycling Awareness Shortfalls The researcher identifies several factors of recycling awareness shortfalls like insufficient information exposure, inadequate 3R (reduce, reuse, recycle) practices, ineffective campaigns by the government, etc (Kate, 2022; Tiew et al., 2019). This assertion is supported by Gilani et al. (2023) findings, where their research identified a lack of awareness of environmental care as the leading cause of why plastic waste is the most waste disposal in landfills. When the importance of recycling is not taken seriously, it will lead to diminished public commitment to 3R practices over time. # 2.2 Opinion Of Bangi Selatan Locals issueAforsurvey neglecting of Bangi3R practices. Selatan localsFurthermore, revealed respondentsthat a need forfirmly morebelieve environmentalthat educationawarenessprogramsis theplay maina significant role in understanding recycling benefits and addressing the shortfalls. While the impact of the low resale value of recyclables is acknowledged, it is considered a less significant factor, with only 41.2% of respondents citing it as a reason for reduced recycling activity. Besides that, respondents state that the distance between their homes and the recycling centres causes their efforts not to recycle. ## 2.3. Type of Web Application Web applications are server-based programs accessible via browsers such as Google Chrome and Microsoft Edge (Contributor, 2023). They are built using HTML, CSS, and JavaScript, forming the framework for their development together. These applications operate through client-server communication: when a user (the client) requests data via an HTTP message over TCP/IP, the server responds by sending the requested data in packets (MDN, 2023). Among various types ofweb applications, there are eight groups of web applications, including static websites, dynamic web applications, e- commerce, single-page web applications, animated web applications, content management systems (CMS) web applications and rich internet applications. Choosing the correct type of web application is crucial to ensure it is relevant to the project. Table 1 below compares three types of web applications that have been chosen. battps://doi.org/10.2419 jerina v10i1 510 Authors, 2025 220 Ahmad Nawawi & Razak. / Journal of Computing Research and Innovation (2025) Vol. 10, No. 1 Table 1. Comparison of the types of web applications |Characteristics
|StaticWebApplication
|DynamicWebApplication
|CMSWebApplication
| |---|---|---|---| |Technology

|
Client-sideonly
|
Client-sideandserver-side
|
Dependsonpluginsand

| ||||| ||Limited
|R al-timecontent
|D pendsonpluginsand
template
| |CodingSkills
|Moderate
|High
|Low
| |
DevelopmentCost|Cheap|Expensive|Expensivelicense| |
Pros|SpeedPerformance and
reliability|Can add database andmany
features|
SEO-friendly andtime-
efficient| |Cons
|Non-dynamic content
|Higherdevelopmenttime
|Limitedcustomization
| |Technology|Client-sideonly|Client-sideandserver-side|Dependsonpluginsand
template| Recommendation systems (RS) are tools that retrieve information based on user interactions and interests from a large pool of data, suggesting content aligned with their activities and preferences (Liling, 2019; Maphosa & Maphosa, 2023). RS is used widely around the globe, especially in entertainment streaming like Netflix and Amazon Prime Video. According to Maphosa & Maphosa (2023), RS are innovative tools that predict what users might do next by looking at things they like and have done before from historical data. # 2.4 Overview Of Recommendation System Content-based Filtering (CBF) is a fundamental RS that suggests the content by analysing relationships between user interests and item representation (Liling, 2019). CBF depends on the similarity of the user profile and items to suggest the user's potential content (Liling, 2019; Maphosa & Maphosa, 2023; Pujahari & Sisodia, 2022). Both past behaviours from the user profile and item representation are used to build a user preference profile; later, the user preference will be used as a dataset to compare new items with the user's historical behaviour (Pujahari & Sisodia, 2022). One advantage of CBF is its capability to recommend relevant information to a user even when there has not been any prior interaction between the user and the system (Pérez-Almaguer et al., 2021). Next, a collaborative filtering (CF)-based recommendation system relies on gathering user behavior data to determine item relevance based on user preferences or similarities, ultimately generating recommendations by finding associations between users with comparable interests (Liling, 2019; Maphosa & Maphosa, 2023; Pujahari & Sisodia, 2022). In CF, one widely employed method is the rating system, which gauges user satisfaction or preference regarding a recommended item (Fkih,2023). Unlike CBF, CF requires interactions among users to establish connections between similar user profiles, making it highly effective in environments with rich user interaction data. Additionally, hybrid recommendation systems aim to overcome the limitations of both CBF and CF by integrating multiple recommendation technologies. This approach leverages the strengths of each method, enhancing recommendation accuracy and compensating for data limitations. By doing so, hybrid systems effectively address challenges like the cold start problem commonly encountered in CF systems (Liu et al., 2022). For instance, in a movie recommendation application, the Content-Based Collaborative Filtering (CBCF) model first utilizes CBF to analyse item attributes and generate initial user preferences. These preferences are then refined using CF to identify similarities among users with analogous tastes, resulting in more personalized recommendations (Afoudi et al., 2021). battps://doi.org/10.2419 jerina v10i1 510 Authors, 2025 221 Ahmad Nawawi & Razak. / Journal of Computing Research and Innovation (2025) Vol. 10, No. 1 3. METHODOLOGY This section discusses the specific approach employed to accomplish the research objectives by highlighting several important methods implemented in the CBF algorithm construction. ## 3.1 Data Classification For item representation, the valuable information related to recycling content data is stored in the string, and every string represents the description of the content material. The content material comprises 20 examples, with five focused on e-waste, three on paper, six on plastics, three on metals, and three on general recycling. This collection includes pollution facts, awareness information, and fun facts about recycling, all designed to highlight the environmental importance of recycling. Since the user profile is based on an input query, therefore data sample for the user profile is insignificant. ## 3.2 Text Vectorization In order to determine the similarity between the user query and content item representation, it is essential first to convert the text into numerical vectors. The Term Frequency-Inverse Document Frequency (TF-IDF) technique plays a crucial role in this transformation, making it possible to compare user queries and content descriptions meaningfully. This approach is widely adopted in content-based filtering (CBF) systems. Given that the proposed system is centred around text representation and information retrieval, the content and user queries are represented as numerical vectors, where each term in the content description is assigned, a value reflecting its relevance. Thus, effective for the CBF system relies on user query input as the user profile representation. TF-IDF calculates the significance of each term by considering two key factors: (i) Term frequency (TF) To determine the importance of a term within a content description or user query, TF calculates how often a term appears, with more frequent terms indicating higher relevance. This approach highlights the critical terms for more accurate content recommendations. TF formula is provided as follows: Wheren refers to the number of occurrences of terms in content, i denotes the specific term, j denotes specific content, and k denotes all terms in the content. # (ii) Inverse document frequency (IDF) The standard IDF formula has been modified to incorporate a smoothing technique in this system. This adjustment serves two essential purposes: first, it prevents excessively high IDF values that could arise with the standard formula; second, it ensures that terms that do not appear in any document are still assigned a nonzero and non-negative IDF value. By balancing term frequency with document frequency, this modified IDF improves the system's ability to make accurate content recommendations to its users. The modification theorem is as follows: where N represents the total content and df_i is the number of terms mentioned in the content. battps://doi.org/10.2419 jerina v10i1 510 Authors, 2025 222 Ahmad Nawawi & Razak. / Journal of Computing Research and Innovation (2025) Vol. 10, No. 1 ## (iii) TF-IDF The TF-IDF process results in two key vectors: one for the content item and one for the user query. These vectors are derived using the TF-IDF formula, TF-IDF = TF x IDF, which provides a numerical representation of term relevance in both the content and user input. ## 3.3. Similarity Computation Cosine similarity measures the similarity between the TF-IDF vector representation of the item representation (content description) and the TF-IDF vector representation of user interest (user query). Theoretically, a smaller cosine distance indicates that the similarity between the two nonzero vectors is closer and more relevant to recommend. The cosine similarity theorem is shown below: A and B are nonzero vectors, representing user interests and item representation respectively. ## 3.4 Content-based Filtering Architecture and Algorithm The CBF architecture for this system is illustrated in Figure 1, which presents a model tailored to handle search queries within a user interface. The core functionality of this model is to recommend content that aligns with the user's search input by leveraging a string-based similarity computation. When a user enters a query into the search box, this input is treated as the user's profile. The model then compares this profile against the textual representations of available content, which consists of materials related to recycling awareness and education. By analysing the similarity between the user's query and the content descriptions, the system can suggest relevant materials that closely match the user's interests, thereby providing a personalized and efficient information retrieval experience. Content4
s Content.
Content 4
A & wo ‘Fp Similar to
4 Content 2
user SSL Meng fi
~ ve De Content 3
Content4
Fig 1: Proposed Content-based Filtering Model
the queryIn realizingand the thiscontent model,into thenumerical system employsvectors.a featureThese extractionvectors are usingessential the TF-IDFfor measuring methodthe to transformdegree of
similarity between the user's input and the available content using cosine similarity. This technique allows
the system to quantify the relevance of each content item to the user's query, ensuring that the most pertinent
materials are recommended.
# 4. RESULT AND DISCUSSIONS This section shows the experimental evaluation of the modification of the IDF theorem. The test will conduct to determine whether the proposed CBF algorithm manages to recommend to users the recycling battps://doi.org/10.2419 jerina v10i1 510 Authors, 2025 223 Ahmad Nawawi & Razak. / Journal of Computing Research and Innovation (2025) Vol. 10, No. 1 content based on their input query, including its recommendation accuracy. Besides that, acceptance testing was conducted to assess whether the proposed system could effectively promote interest in 3R practices among residents of Bangi Selatan. # 4.1 Evaluation Result The performance of the proposed IDF theorem and standard theorem are summarized in Table 2, highlighting three document terms, which are "awareness', "recycling", and "plastics" from 20 documents (content) as the dataset. Table 2. Comparison of IDF value between different theorems |DocumentTerms|awareness|plastics|recycling| |---|---|---|---| |
Standard Theorem|1.2040|1.3863|0.2231| |ProposedTheorem|2.2528|2.0986|1.2113| Based on Table 2 above, the Standard Theorem yields IDF values of 1.2040, 0.2231, and 1.3869 for these terms, while the Proposed Theorem produces higher values of 2.2528, 1.2113, and 2.0986. Consequently, the Proposed Theorem consistently generates higher IDF values, especially for standard terms, potentially assigning greater weight to frequent terms in subsequent calculations. This adjustment could lead to improved outcomes compared to the Standard Theorem. This highlights how choosing the Smoothing technique can substantially influence the relevance and weighting of terms, potentially enhancing the overall performance of the recommendation system. Moreover, Tables 3 and 4 below present the results, demonstrating the system's ability to recommend recycling content, specifically when incorporating the user query input. Table 3. Example of user input and content material |1D|Vector
DocumentText/UserInput| |---|---| |1|
Itemrepresentation
Greenhouse emissionreductionbyrecyclingpapers| |2
3
ble 4. Simila|Ttemrepresentation
e-Waste shouldberecycledto avoidpollution
User Interest
papers dump effect
rityand content recommendation with ID=3| |Vector
|1
2

| |Similarity|‘Yes
No| |Recommend|to user
Yes
‘No| |Reason|Have apaperterm
Nomatchedterm| |Cosinesimila|rityvalue
Existwithuserquery
Doesnotexistwithuserquery| Table 4. Similarity and content recommendation with ID=3 The experiment results in Table 4 illustrate the effectiveness of the proposed CBF system in recommending relevant content to users based on their input. Using cosine similarity, we compare the user input (ID=3) with two document vectors (ID=1 and ID=2). Document ID=1 contains the term "papers," which also exists in the user input, leading to a positive cosine similarity value. As a result, this document is recommended to the user. In contrast, document ID=2 lacks any matching terms with the user input, resulting in a negative similarity assessment and the document not being recommended. This demonstrates that the system successfully identifies relevant documents when standard terms are present between the input and content, ensuring accurate recommendations. battps://doi.org/10.2419 jerina v10i1 510 Authors, 2025 224 Ahmad Nawawi & Razak. / Journal of Computing Research and Innovation (2025) Vol. 10, No. 1 Fig. 2: Proposed Content-based Filtering Model Next, Figure 2 presents the results of the CBF recommendation algorithm when the user query includes the topics "e-waste" and "recycling." The system successfully ranks content items based on their cosine similarity scores, with the most relevant content appearing at the top. The scores range from this maximum value down to 0, clearly indicating content relevance. Items with a score of 0 were deemed irrelevant to the user's query. This distribution of similarity scores indicates that the system can effectively distinguish between highly relevant, moderately relevant, and irrelevant content. As a result, users are presented with a well-curated selection of information tailored to their specific interests, encouraging them to practice recycling through reading. ## 5. USER ACCEPTANCE TESTING The user acceptance survey conducted in Bangi Selatan, involving 30 respondents, showed positive feedback regarding the CBF web system’s impact on recycling behaviour. Users reported feeling more engaged in recycling after reading the recommended content tailored to their interests based on their input queries. The CBF system, powered by TF-IDF and cosine similarity, successfully matched most user inputs to relevant content. Additionally, respondents indicated that the web application was impactful in encouraging recycling, showing overall positive user acceptance. However, issues were raised regarding the system's inability to recognise synonymous terms, such as "recycling" and "recycled," which occasionally hindered its effectiveness in delivering relevant recommendations. Besides that, some users expressed frustration with the system's handling ofunknown inputs, noting that when it could not recognise their interests, it provided alternative content that was often unrelated. Addressing this limitation could enhance the system's usability and accuracy in future iterations. # 6. CONCLUSION AND RECOMMENDATION In conclusion, the outcome of User Acceptance Testing (UAT) with residents in Bangi Selatan successfully met the project’s objectives. Based on the current evidence, the online survey results indicate that the content-based filtering recommendation system effectively promotes recycling practices by delivering tailored recycling-related content through the search box. This approach has significantly increased user awareness and understanding of recycling’s importance. battps://doi.org/10.2419 jerina v10i1 510 Authors, 2025 225 Ahmad Nawawi & Razak. / Journal of Computing Research and Innovation (2025) Vol. 10, No. 1 However, the implementation of TF-IDF with cosine similarity presents some limitations. TF-IDF’s focus on word frequency rather than semantic meaning can lead to missed connections between synonyms, such as “paper” and “papers.” Additionally, the system may face issues with sparsity as vocabulary size grows, diminishing the effectiveness of vectorization and leading to decreased cosine similarity scores. To enhance the system’s accuracy and usability, future iterations should consider incorporating semantic analysis to address synonym recognition and reduce sparsity challenges. # 7. ACKNOWLEDGEMENTS I want to extend my most profound appreciation to my supervisor, Dr Tajul Rosli bin Razak, for their invaluable guidance, unwavering support, and insightful feedback throughout this journal writing. I also thank God for granting me the opportunity to pursue my bachelor’s in computer science and for the strength to complete this challenging endeavor. I am thankful to the academic staff of Universiti Teknologi MARA (UiTM) for their assistance and resources and to my family and friends for their patience, understanding, and constant motivation during difficult times. Special thanks to the Rakan Alam Sekitar UiTM Student Chapter for their invitation, which enriched my experience in environmental conservation. Finally, I am deeply grateful to all the participants who shared their time and experiences, making this study possible. ## 8. REFERENCES - Afoudi, Y., Lazaar, M., & Al Achhab, M. (2021). Hybrid recommendation system combined content-based filtering and collaborative prediction using artificial neural network. Simulation Modelling Practice and Theory, 113. https://doi.org/10.1016/j.simpat.2021.102375 - Contributor, T. (2023b, January 20). web application (web app). Software Quality. https://www.techtarget.com/searchsoftwarequality/definition/Web-application-Web-app - David, A., Thangavel, Y. D., & Sankriti, R. (2019). Recover, recycle and reuse: An efficient way to reduce the waste. International Journal of Mechanical and Production Engineering Research and Development, 9(3), 31-42. https://doi.org/10.24247/ijmperdjun20194 - Fkih, F. (2023). Enhancing item-based collaborative filtering by users’ similarities injection and lowquality data handling. Data & Knowledge Engineering, 144, 102126. https://doi.org/10.1016/j.datak.2022. 102126 - Gilani, I. E., Sayadi, S., Zouari, N., & Al-Ghouti, M. A. (2023). Plastic waste impact and biotechnology: Exploring polymer degradation, microbial role, and sustainable development implications. In Bioresource Technology Reports (Vol. 24). Elsevier Ltd. https://doi.org/10.1016/j.biteb.2023.101606 - Gill, S. S., Xu, M., Patros, P., Wu, H.. Kaur, R., Kaur, K., Fuller, S., Singh, M., Arora, P., Parlikad, A. K., Stankovski, V., Abraham, A., Ghosh, S. K., Lutfiyya, H., Kanhere, S. S., Bahsoon, R., Rana, O., Dustdar, S., Sakellariou, R., ... Buyya, R. (2024). Transformative effects of ChatGPT on modern education: Emerging Era of AI Chatbots. Internet of Things and Cyber-Physical Systems, 4, 19-23. https://doi.org/10.1016/j.iotcps.2023.06.002 - How the web works - Learn web development | MDN. (2023, November 17). MDN Web Docs. https://developer.mozilla.org/enUS/docs/Learn/Getting_started_withthe _web/How_the_Web_works - Liling. L. (2019). Summary of recommendation system development. Journal of Physics: Conference Series, 1187(5). https://doi.org/10.1088/1742-6596/1187/5/052044 Liu, J., Shi, C., Yang, C., Lu, Z., & Yu, P. S. (2022). A survey on heterogeneous information network based battps://doi.org/10.2419 jerina v10i1 510 Authors, 2025 226 Ahmad Nawawi & Razak. / Journal of Computing Research and Innovation (2025) Vol. 10, No. 1 recommender systems: Concepts, methods, applications and resources. AI Open, 3, 40-57. https://doi.org/10.1016/j.aiopen.2022.03.002 - Maphosa, V., & Maphosa, M. (2023). Fifteen Years of Recommender Systems Research in Higher Education: Current Trends and Future Direction. In Applied Artificial Intelligence (Vol. 37, Issue 1). Taylor and Francis Ltd. https://doi.org/10.1080/08839514.2023.2175106 - Pérez-Almaguer, Y., Yera, R., Alzahrani, A. A., & Martinez, L. (2021). Content-based group recommender systems: A general taxonomy and further improvements. Expert Systems with Applications, 184, 115444. https://doi.org/10.1016/j.eswa.2021.115444 - Pujahari, A., & Sisodia, D. S. (2022). Item feature refinement using matrix factorization and boosted learning based user profile generation for content-based recommender systems. Expert Systems with Applications, 206. https://doi-org/10.1016/j.eswa.2022.117849 - Tiew, K. G., Basri, N. E. A., Deng, H., Watanabe, K., Zain, S. M., & Wang, S. (2019). Comparative study on recycling behaviours between regular recyclers and non regular recyclers in Malaysia. In Journal of Environmental Management (Vol. 237, pp. 255-263). Academic Press. https://doi.org/10.1016/j.jenvman.2019.02.033 - Zaharudin, Z. A., Shuib, A., Hadianti, R., & Rodzi, Z. M. (2023). Towards Sustainable City: A Covering Model for Recycling Facility Location-allocation in Nilai, Malaysia. Science and Technology Indonesia, 8(4), 570-578. https://doi.org/10.26554/sti.2023.8.4.570-578 - © 2025 by the authors. Submitted for possible open access publication under the terms and - BY conditions of the Creative Commons Attribution (CC BY) _ license (http://creativecommons.org/licenses/by/4.0/). battps://doi.org/10.2419 jerina v10i1 510 Authors, 2025