Reverse Recommendation about Pure Cold-Start Item in Multi-hierarchical Heterogeneous Network

• Jianfeng Li,
• Mingxiao Fu,
• Jun Zhai,

Vol. 20, No. 1, January 31, 2026
10.3837/tiis.2026.01.012, Download Paper (Free):

• Reverse recommendation
• Pure cold-start item
• Dual pure cold-start problem
• Multiple layers
• Heterogeneous Network

Abstract

Pure item cold-start problem has received a lot of attention in recommender system, when more and more new products and services hit the market. However, it is mostly overlooked to evaluate the number of users who choose those new or cold-start items reversely. Especially, that reverse recommendation is a dual pure cold-start recommendation, including both cold-start items and users, because cold-start items can attract some cold-start users as well as existing users who have been recorded with known products. To address the problem, this paper proposes a multi-hierarchical heterogeneous network approach, in which, virtual edges and loops are added to concatenate meta paths crossing different layers, and relative auxiliary coefficients are applied to adjust probabilities on the paths from cold-start item nodes to both existing and cold-start user nodes. Through utilizing Movie Lens ml-100k dataset to compare the proposed approach of adjusting coefficients with the general maximum possibility path approach, six important performance metrics: accuracy, precision, recall, FPR, F1 and lift, all demonstrates better effectiveness. For existing users, those metrics have improved by about 0.6%, 37.7%, 39.2%, -3.9%, 31.5% and 39.2%, and for cold-start users, the improvement percentages of those metrics are about 0.1%, 11.8%, 56.0%, -0.3%, 15.9% and 56.0% respectively.

Statistics

Cite this article