基于用户坐标距离与误差修正的推荐算法
Recommendation Algorithm Based on User Coordinate Distance and Error Correction

推荐系统是一种根据用户历史行为、社交关系等信息去判断用户感兴趣的物品的智能系统，它可以有效的解决互联网信息过载的问题，因此具有重要的研究意义和广泛应用价值。为了解决推荐算法存在准确性有待提升和可扩展性不足的问题，本文提出了一种带有相似性选择和误差修正的用户坐标与评分距离推荐算法。该算法首先基于用户间相似性筛选出正相关的邻居，针对筛选后的邻居集合计算评分距离。然后建立用户间评分距离的合成坐标模型，与得到的用户坐标间距离进行评分预测。最后基于训练题设计误差修正算法，进一步提升预测准确度。通过本文算法与其他推荐算法在MovieLens数据集上进行对比，试验结果表明，该算法可以有效提高推荐算法的预测准确性和可扩展性。研究揭示了利用用户坐标距离进行预测的可行性，为进一步研究推荐系统的性能提升和运行机理提供了有效的依据。
The recommendation system is an intelligent system that determines the items that the user is interested in based on the user’s historical behavior, social relationships and other information. It can effectively solve the problem of Internet information overload, so it has important research significance and wide application value. In order to solve the problems of recommendation algorithm that needs to be improved in accuracy and insufficient scalability, this paper proposes a user coordinate and rating distance recommendation algorithm with similarity selection and error correction. The algorithm first selects positively correlated neighbors based on the similarity between users, and calculates the scoring distance for the filtered neighbor set. Then a synthetic coordinate model of the rating distance between users is established, and the rating prediction is performed with the obtained distance between user coordinates. Finally, an error correction algorithm is designed based on the training questions to further improve the prediction accuracy. By comparing the algorithm in this article with other recommendation algorithms on the MovieLens data set, the experimental results show that the algorithm can effectively improve the prediction accuracy and scalability of the recommendation algorithm. The research reveals the feasibility of using user coordinate distance for prediction, and provides an effective basis for further research on the performance improvement and operating mechanism of the recommendation system.