Bundle recommendation systems play an essential role in enhancing user experience on
e-commerce platforms, particularly in facilitating personalized and contextual product ex-
ploration. However, most existing bundle recommendation systems still face limitations,
such as low personalization capability, limited diversity of recommendation results, and
the absence of interaction mechanisms that allow users to directly adjust recommenda-
tions according to their preferences. These shortcomings often result in monotonous rec-
ommendations that are less relevant to the dynamic needs of users.This study proposes
a hybrid approach to generate relevant and adaptive book bundles by integrating the
Modified FP-Growth (MFP-Growth) algorithm with the Compound Critiquing technique.
MFP-Growth is employed to extract association rules by considering relevance metrics
such as confidence, lift, and metadata scores (e.g., similarity in genre, author, and pub-
lication year). The resulting rules are then validated to form initial bundles with strong
semantic quality and topical coherence. Compound Critiquing further enables users to
provide multidimensional feedback on the recommended bundles, such as preferences for
price, di!erent genres, authors, or publishers. This mechanism enriches system interaction
by iteratively refining recommendations based on user input.Implementation and analysis
results show that the proposed approach improves MAP@10 by 0.181556, nDCG@10 by
0.327216, and increases diversity and novelty to 0.949578 and 5.75, respectively, compared
to the FP-Growth baseline. Further analysis demonstrates that Compound Critiquing ef-
fectively shifts recommendation focus in line with user preferences, with a slight decrease
in similarity score that contributes to broader item exploration. Moreover, MFP-Growth
achieves computational e”ciency with approximately 30% faster execution time (151.50
seconds) and lower memory usage (37 MB) compared to conventional FP-Growth. These
findings confirm that the proposed system o!ers advantages in interpretability, e”ciency,
and recommendation quality, making it a viable alternative to deep learning-based ap-
proaches. With a transparent architecture and direct user feedback integration, this study
provides a practical contribution to the development of adaptive bundle recommendation
systems in the book e-commerce domain.
Keywords: Bundle Recommendation System, MFP-Growth, Compound Critiquing