A Network Embeddings based Recommendation Model with multi-factor consideration
- Liffey Hall 1
- 10:20 on 14 July 2022
- 30 minutes
Recommendation systems are increasingly in demand to provide a personalized customer experience for diversified product mix offerings. Traditionally we use interaction information based on user preferences and item characteristics. This brings in collaborative filtering-driven recommendations with higher accuracy and relevance. However, such a method has certain limitations in utilizing implicit information like cross-domain specific factors that are equally important for making personalized recommendations. We propose an improvised way of using network embeddings based matrix factorization technique with multi-factors to make a match between both implicit and explicit features resulting in more accurate recommendation.
TalkPyData: Machine Learning, Stats
The method consists of three main steps: First, network embedding formulation performed on each user specific behavior network; Then, embeddings weight distribution estimated through intermediate layers of network with final layer for target (item purchased as labels); Finally, both factors: (a) Learned weights from implicit data (cross-domain) and (b) explicit factors from domain data used by multi-factorization method for recommendations. The proposed method transfers knowledge across implicit and explicit factors and associated dimensions. The suggested approach tested real-world data with evidence of outperforming existing algorithms with significant lift in recommendation accuracy. Empirical experimentation outcomes illustrate the potential of both factors for making effective recommendations.