[论文解读] Symmetric collaborative filtering using the noisy sensor model

Rita Sharma|arXiv (Cornell University)|Jan 1, 2009

Recommender Systems and Techniques参考文献 8被引用 3

一句话总结

该论文提出一种对称协同过滤方法，利用噪声传感器模型通过将其他用户视为噪声传感器来预测用户评分。该方法提出两种变体——针对二值评分的贝叶斯学习方法，以及针对多值评分的线性/同质传感器模型，相较于最先进方法在EachMovie数据集上表现出显著的性能提升，尤其在评分稀疏的情况下，通过联合用户与项目相似性实现更高的预测准确率。

ABSTRACT

Collaborative filtering is the process of making recommendations regarding the potential preference of a user, for example shopping on the Internet, based on the preference ratings of the user and a number of other users for various items. This thesis considers the problem of symmetric collaborative filtering based on explicit ratings. To evaluate the algorithms, we consider only pure collaborative filtering, using given ratings and excluding other information about the people or items. Our approach is to predict an active user's preferences regarding a particular item by using other people's ratings of that item and other items rated by the active user as noisy sensors. The noisy sensor model uses Bayes' theorem to compute the probability distribution for the active user's rating of a new item. We give two variants for learning the noisy sensor model: one, for explicit binary rating data; and the second, for explicit multi-valued rating data. The model for binary rating data is based on Bayesian learning. Its performance motivate us to further explore the use of noisy sensor model for multi-valued rating data. We give two variant models for multi-valued rating data: in one, we learn a linear model of how users rate items; in another, we assume different users rate items identically, but that the accuracy of the sensors must be learned. We compare the two models of multi-valued rating data with stateof- the-art techniques and show how they are significantly better whether a user has rated only two items or many. We report empirical results using the EachMovie database of movie ratings. We also show that by considering the items similarity along with the users similarity the accuracy of the prediction increases.

研究动机与目标

• 开发一种对称协同过滤框架，将用户视为噪声传感器以实现评分预测。
• 利用贝叶斯与线性传感器模型对显式二值和多值用户评分进行建模。
• 在用户仅评分少数项目（低数据场景）下评估模型性能。
• 整合用户与项目相似性以提升预测准确率。
• 将所提模型与最先进协同过滤技术进行对比。

提出的方法

• 噪声传感器模型应用贝叶斯定理，计算活跃用户对未见项目的后验概率分布。
• 针对二值评分，采用贝叶斯学习方法估计传感器可靠性并预测用户偏好。
• 针对多值评分，提出两种模型：一种基于用户评分行为的线性模型，另一种假设用户行为一致且传感器精度可学习。
• 通过利用用户与项目相关性结构，整合用户与项目相似性以提升预测性能。
• 使用EachMovie数据集进行模型训练与评估，仅关注显式评分，不依赖辅助信息。
• 通过在噪声传感器框架下进行概率推理，结合多个用户评分的证据生成预测结果。

实验结果

研究问题

• RQ1 与现有方法相比，噪声传感器模型如何提升协同过滤的预测准确率？
• RQ2 当用户仅评分少量项目时，该模型是否仍能保持高性能？
• RQ3 对用户行为的不同假设（线性 vs. 同质评分模式）如何影响预测质量？
• RQ4 同时引入用户与项目相似性在多大程度上提升了预测准确率？
• RQ5 针对二值评分的贝叶斯学习方法与多值评分模型相比表现如何？

主要发现

• 所提出的噪声传感器模型显著优于最先进协同过滤技术，尤其在用户仅评分两项内容时表现突出。
• 针对二值评分的贝叶斯学习变体表现优异，为后续探索多值评分模型提供了有力基础。
• 多值评分的线性模型性能优于同质传感器模型，尤其在捕捉个体用户评分倾向方面更具优势。
• 同时整合用户与项目相似性可显著提升预测准确率。
• 在EachMovie数据集上的实证结果证实，噪声传感器模型在不同评分稀疏程度下均表现出强鲁棒性与有效性。
• 该模型展现出优异的泛化能力，在用户评分历史极短时仍能保持高准确率。