<![CDATA[Community Service (PkM) activities constitute a vital pillar of the higher education "Tridharma" (Three Pillars of Higher Education), serving to apply students' technological competencies to address social issues. However, the surge in PkM proposals each semester has created serious administrative challenges, specifically the high frequency of topic repetition and duplication. The manual verification process currently conducted by supervising lecturers faces constraints regarding memory limitations, time inefficiency, and the risk of subjectivity. This research aims to design, implement, and test a web-based automated PkM title selection system utilizing a combination of Natural Language Processing and Information Retrieval. The methodology involves five text preprocessing stages (case folding, cleaning, tokenizing, stopword removal, and stemming using the Sastrawi stemmer), Term Frequency-Inverse Document Frequency (TF-IDF) feature weight extraction, and vector angular distance calculation using Cosine Similarity. The system was developed using the Flask micro-web framework and an SQLite database, implemented in Python 3.11.9. Experiments were conducted on a primary dataset of 839 PkM titles from Informatics Engineering students at Universitas Pamulang (comprising a reference corpus of 641 approved titles and a test set of 198 rejected titles). Threshold tuning established three decision zones: Highly Similar/Saturated (≥0.80), Partially Similar (0.60–0.79), and Unique/Eligible (<0.60). Classification performance testing against expert ground truth yielded an Accuracy of 95.6%, Precision of 92.8%, Recall of 84.5%, and an F1-Score of 88.4%, significantly exceeding the study's minimum key performance indicator (≥85%). The system proved effective as a digital reference for the selection process, reducing verification time and ensuring the originality of student PkM innovation outputs]]>
