Piecewise Weighting Function for Collaborative Filtering Recommendation

The assignment of a fixed weight value to an attribute (or variable) is not always considered reasonable, as it may not effectively preserve user similarity, potentially resulting in a decline in the performance of collaborative filtering recommendation algorithms. In this article, we introduce a piecewise weighting method that incorporates hyper-class representation to enhance collaborative filtering recommendations. Our approach begins with applying a kernel function to map the original data into a kernel space, facilitating the learning of attribute weights. Subsequently, we construct a hyper-class representation of the data to derive weights for segmented attribute values (hyper-classes) within each attribute, creating a piecewise weighting function. This piecewise weighting function is then utilized to compute user similarities for collaborative filtering recommendations. Finally, we conduct a series of experiments to assess the performance of the collaborative filtering recommendation algorithm. The results demonstrate that the proposed algorithm, employing the piecewise weighting function, outperforms the compared algorithm that uses fixed weight values, as assessed by RMSE, Mean Absolute Error (MAE), and Precision. The source code for the proposed algorithm is available at https://github.com/Lijy207/HCPW .