基于模拟评分的服装推荐改进算法

针对传统服装推荐算法中缺乏对消费者与服装特性的关注,以及预测结果缺乏针对性和有效性的问题,利用服装编码、时间间隔和欧氏距离等参数构建了消费者购物兴趣衰减模型,提出基于模拟评分的服装推荐改进算法。对比了模拟评分算法与基于奇异值分解的改进算法的预测值和真实值之间的平均绝对误差。结果表明:模拟评分算法预测评分的平均绝对误差为0.808,相对于基于奇异值分解的改进算法,误差降低了0.024,其中25%的个案的误差大于1,排除这部分个案后的平均绝对误差为0.632;通过对消费者进行回访分析发现,90%消费者的推荐准确率大于96%,只有10%的消费者的推荐准确率为60%~64%;导致误差较大的原因是这部分消费者的喜好发生变化,或是长期没有购买服装。

Improved clothing recommendation algorithm based on simulation scoring

The traditional clothing recommendation algorithms do not pay enough attention to consumers and clothing characteristics, hence the prediction results are short in pertinence and effectiveness. To improve on these, a model of consumers' interest attenuation in shopping was constructed by using clothing coding, time interval and Euclidean distance, and an improved clothing recommendation algorithm based on simulation scoring was proposed. By comparing the average absolute error between the true values and the predicted values of the simulation scoring algorithm and the improved algorithm based on singular value decomposition, it is found that the average absolute error of the simulation scoring algorithm is 0.808, which is 0.024 lower than that of the improved algorithm based on singular value decomposition. The error of 25% of all cases is bigger than 1, and the average error after excluding this part of cases is 0.632. Through such case analysis, it is found that the average absolute accuracy of 90% recommendation is greater than 96%, and the accuracy of 10% recommendation is between 60% and 64%. The reason for big error is either because of the preference changes of the targeted consumer groups, or the targeted consumer group have not purchased clothes for a long time.