A Hybrid Examination Questions Generator and Recommender System Using Rule-Based and Collaborative Filtering Algorithms
2026 IEEE International Conference on Automatic Control and Intelligent Systems (I2CACIS), (2026), pp. 342-346
Fanny C. Almeniana
a
,
Ace C. Lagman
b
,
Isagani M. Tano
c
,
Jayson M. Victoriano
d
,
Joseph D. Espino
e
,
Jonilo C. Mababa
a
,
Jovy Jay D. Cabrera
f
a La Consolacion University Philippines, City of Malolos, Philippines
b FEU Institute of Technology, Manila, Philippines
c Quezon City University, Quezon City, Philippines
d Bulacan State University, City of Malolos, Philippines
e National University Baliwag, Baliwag, Philippines
f Immaculate Conception I (ICI), College of Arts and Technology, Philippines
Abstract: Traditional manual exam generation is time-consuming, prone to question redundancy, and lacks analytics for evaluating exam effectiveness. This paper presents a hybrid examination question generator and recommender system that integrates rule-based filtering with collaborative filtering algorithms to automate and optimize the exam creation process. The rule-based engine selects questions from a centralized item bank using educator-defined parameters aligned with Bloom's Taxonomy cognitive levels, question types, and topic coverage, employing stratified sampling with Linear Congruential Generator (LCG) randomization. The collaborative filtering engine analyzes student exam responses through Pearson point-biserial correlation to compute discrimination indices, generating item-level recommendations (retain, revise, or replace) that refine subsequent exam iterations. Question randomization is implemented via the Fisher-Yates Shuffle algorithm to ensure unique exam versions per student. The system was developed using Agile Scrum methodology and evaluated by 23 respondents (faculty members and IT professionals) using the ISO/IEC 25010 Software Product Quality Model. Results yielded a grand mean of 4.33 (Acceptable), with Performance Efficiency achieving the highest rating of 4.68 (Highly Acceptable). The system supports both automatic and manual exam generation modes, automated item analysis with collaborative recommendations, and generates a Table of Specifications (TOS) and answer keys. The hybrid approach demonstrates that combining pedagogical rule structures with data-driven collaborative insights produces higher-quality, fairer, and more diverse examinations compared to either method alone.