融合情感评论倾向与均衡长尾物品的推荐方法

推荐领域中已有研究较多的考虑属性维度在评分-物品上来提高算法的准确性,对于用户对产品的真实态度判断不够。论文利用LSTM神经网络模型将用户评论情感得分和用户评分进行融合分析,来计算用户对物品真正的兴趣度;运用向量空间对用户情感进行三分类,借鉴经济学中基尼系数的思想,引入惩罚因子通过对热门物品进行调节,发挥系统的长尾能力从而增加推荐结果的覆盖率,从而实现对热门物品和冷门物品被推荐程度的均衡优化。最后,对情感系数和惩罚系数参数的取值进行实验,得出最优模型参数组合,结果表明,评论的情感因素和惩罚系数对于构建性能更优的推荐模型效果明显,该模型在推荐的准确率和覆盖率上相较其他主流模型更加符合用户需求。

Recommendation Method Integrating Emotional Comment Tendency and Balancing Long Tail Items

In the field of recommendation, many studies have considered the attribute dimension to improve the accuracy of the algorithm in terms of score-item, which is not enough to judge the user’s real attitude towards the product. In this paper, the LSTM neural network model is used to fuse the user comment emotion score and user score to calculate the user's real interest in the item. This paper uses vector space to classify user emotion into three categories, draws lessons from the idea of Gini coefficient in economics, introduces penalty factor, and gives play to the long tail ability of the system by adjusting popular items, so as to increase the coverage of recommendation results, and realize the balanced optimization of the recommended degree of popular items and unpopular items. Finally, experiments are carried out on the values of emotion coefficient and punishment coefficient parameters, and the optimal combination of model parameters is obtained. The results show that the accuracy and coverage of the model are more in line with user needs compared with other mainstream models. It is also proved that the commented emotion factors and punishment coefficient are effective for building a better recommendation model.