摘要
推荐系统是个性化服务的重要工具,但现有方法对隐式反馈语义信息利用不足,影响推荐精度。为此,提出一种融合情感分析与主题建模的推荐算法。首先利用RoBERTa(Robustly Optimized BERT Pretraining Approach,鲁棒优化BERT预训练方法)提取文本特征,并结合BTM(Biterm Topic Model,词对主题模型)主题模型构建基于主题相似度的图结构,引入注意力与门控机制更新节点特征,通过GCN(Graph Convolutional Network,图卷积网络)实现特征聚合。同时,将情感分析结果转化为用户隐式偏好评分,并与显式评分加权融合进行评分预测。基于豆瓣电影评论数据集上的实验结果表明,该方法在推荐准确性上优于对比模型。
Recommender systems are important tools for personalized services.however,existing methods make insufficient use of the semantic information contained in implicit feedback,which limits recommendation accuracy.To address this issue,this paper proposes a recommendation algorithm that integrates sentiment analysis with topic modeling.Text features are first extracted using RoBERTa,and a topic-similarity-based graph is constructed with the BTM model.Attention and gating mechanisms are then introduced to update node representations,which are subsequently aggregated through Graph Convolutional Network(GCN).Meanwhile,sentiment analysis results are transformed into implicit user preference scores and combined with explicit ratings in a weighted manner for rating prediction.Experimental results on the Douban movie review dataset demonstrate that the proposed method outperforms comparative models in recommendation accuracy.
作者
章宏远
谢晋
ZHANG Hongyuan;XIE Jin
出处
《安徽职业技术大学学报》
2026年第1期14-20,27,共8页
Journal of Anhui University of Applied Technology
基金
2024年度安徽省高校自然科学重点研究项目“基于NLP多特征融合的安全漏洞属性提取研究”(2024AH050900):2023年度安徽省高校自然科学重点研究项目“基于Spark的并行大数据清洗的研究”(2023AH051450)。
关键词
推荐系统
BTM主题建模
注意力机制
门控机制
特征聚合
评分预测
recommender systems
BTM topic modeling
attention mechanism
gating mechanism
feature aggregation
rating prediction
