The explosive growth of microblogging data makes information screening increasingly difficult.Constructing a reasonable microblog user interest profile helps to achieve accurate service and improve recommendation performance. First, the LDA(latent Dirichlet allocation)model is used to excavate the users content preference characteristics from the users historical content,and the corresponding emotional tendency in the users different content preferences is calculated by the emotional analysis model, so as to obtain the users interest profile containing two dimensions of content preferences and their corresponding emotional tendencies. When evaluating microblog recommendations based on user interest profile, where the users content preferences are used for the initial screening first to see whether the content of the blog post matches with the users content preference. And if they match each other, the algrithm would further filters the blog posts by emotional tendency, compares their emotional similarity under the same content preference, and select the blog posts above the threshold to add into the recommendation set. The experimental results on the real microblog dataset show that the microblog recommendation- DOI:10.3969/j.issn.1000-1565.2025.01.010基于内容偏好和情绪倾向的微博用户兴趣画像构建方法徐建民,王铭宇(河北大学 网络空间安全与计算机学院,河北 保定 071000)摘 要:微博数据的爆炸式增长,使信息筛选变得越来越困难.构建合理的微博用户兴趣画像,有助于实现精准化服务,提高推荐性能.首先,利用LDA(latent Dirichlet allocation)模型从用户历史内容中挖掘用户的内容偏好特征,并通过情绪分析模型计算用户不同内容偏好对应的情绪倾向,得到包含内容偏好及其对应情绪倾向2个维度的用户兴趣画像;在基于用户兴趣画像进行微博推荐评估时,利用用户内容偏好进行初步筛选,比较待评估博文内容与用户的内容偏好是否匹配,若匹配则进一步通过情绪倾向进行过滤,比较同一内容偏好下的情绪相似度,选取高于阈值的博文加入推荐集.真实微博数据集的实验结果表明,与基于标签的推荐模型、基于情感关联规则的推荐模型和基于主题的推荐模型相比,本文微博推荐方法具有更好的性能,在F1值上分别提升了10%、6%和2%.关键词:用户兴趣画像;内容偏好;情绪倾向;微博推荐中图分类号:TP391.3 文献标志码:A 文章编号:1000-1565(2025)01-0091-13Construction method of microblog user interest profile based on content preference and emotional tendencyXU Jianmin,WANG Mingyu(School of Cyber Security and Computer, Hebei University, Baoding 071000, China)Abstract: The explosive growth of microblogging data makes information screening increasingly difficult.Constructing a reasonable microblog user interest profile helps to achieve accurate service and improve recommendation performance. First, the LDA(latent Dirichlet allocation)model is used to excavate the users content preference characteristics from the users historical content,and the corresponding emotional tendency in the users different content preferences is calculated by the emotional analysis model, so as to obtain the users interest profile containing two dimensions of content preferences and their corresponding emotional tendencies. When evaluating microblog recommendations based on user interest profile, where the users content preferences are used for the initial screening first to see whether the content of the blog post matches with the users content preference. And if they match each other, the algrithm would further filters the blog posts by emotional tendency, compares their emotional similarity under the same content preference, and select the blog posts above the threshold to add into the recommendation set. The experimental results on the real microblog dataset show that the microblog recommendation- 收稿日期:2024-09-19;修回日期:2024-11-05 基金项目:国家社会科学基金资助项目(23BTQ092) 第一作者:徐建民(1966—),男,河北大学教授,博士生导师,主要从事信息检索方向研究.E-mail:hbuxjm@hbu.edu.cn 通信作者:王铭宇(1997—),男,河北大学在读硕士研究生,主要从事微博推荐方向研究.E-mail:yu668135@163.com 第1期徐建民等:基于内容偏好和情绪倾向的微博用户兴趣画像构建方法河北大学学报(自然科学版) 第45卷method in this paper has better performance compared with the label-based recommendation model, sentiment association rule-based recommendation model and topic-based recommendation model, with 10%, 6% and 2% improvement in the F1 value, respectively.
[1] AZZAM F, KAYED M, ALI A. A model for generating a user dynamic profile on social media[J]. J King Saud Univ Comput Inf Sci, 2022, 34(10): 9132-9145. DOI: 10.1016/j.jksuci.2022.08.036.
[2] KHALIL M M Y, WANG Q X, CHEN B, et al. Cross-modality representation learning from transformer for hashtag prediction[J]. J Big Data, 2023, 10(1): 140-148. DOI: 10.1186/s40537-023-00824-2.
[3] DJENOURI Y, BELHADI A, SRIVASTAVA G, et al. An efficient and accurate GPU-based deep learning model for multimedia recommendation[J]. ACM Trans Multimedia Comput Commun Appl, 2024, 20(2): 1-18. DOI: 10.1145/3524022.
[4] 邵一博,秦玉华,崔永军,等.融合多粒度信息的用户画像生成方法[J].计算机应用研究, 2024, 41(2): 401-407. DOI: 10.19734/j.issn.1001-3695.2023.05.0234.
[5] 李丹,高建忠.基于用户画像的图书馆推荐服务初探[J].图书馆, 2019(7): 66-71.DOI: 10.3969/j.issn.1002-1558.2019.07.010.
[6] 陈添源,梅鑫.多源数据融合的用户画像识别与推荐实证研究[J].情报理论与实践, 2024, 47(4): 171-180. DOI: 10.16353/j.cnki.1000-7490.2024.04.022.
[7] PUJAHARI A, SISODIA D S. Item feature refinement using matrix factorization and boosted learning based user profile generation for content-based recommender systems[J]. Expert Syst Appl, 2022, 206: 117849. DOI: 10.1016/j.eswa.2022.117849.
[8] 夏立新,胡畔,刘坤华,等.融入信息推荐场景要素的在线健康社区用户画像研究[J].图书情报知识, 2023, 40(3): 116-128.DOI: 10.13366/j.dik.2023.03.116.
[9] 金吉琼,居雷,张易,等.基于用户画像的卷烟消费者特征识别和价值评估[J].烟草科技, 2023, 56(1):105-112. DOI:10.16135/j.issn1002-0861.2022.0531.
[10] 费鹏,林鸿飞,杨亮,等.一种用于构建用户画像的多视角融合框架[J].计算机科学, 2018, 45(1): 179-182. DOI: 10.11896/j.issn.1002-137X.2018.01.031.
[11] CUI Y C, YU H L, GUO X X, et al. RAKCR: reviews sentiment-aware based knowledge graph convolutional networks for Personalized Recommendation[J]. Expert Syst Appl, 2024, 248: 123403. DOI: 10.1016/j.eswa.2024.123403.
[12] 于伟杰,杨文忠,任秋如.基于全词BERT的集成用户画像方法[J].东北师大学报(自然科学版), 2022, 54(4): 87-92. DOI: 10.16163/j.cnki.dslkxb202104053.
[13] 杨洋洋.数据驱动下网络辟谣信息画像与治理模式研究——基于引爆点理论[J/OL].情报科学, 1-14[2024-05-07]. http://kns.cnki.net/kcms/detail/22.1264.G2.20240506.1702.014.html.
[14] 吴树芳,吴崇崇,朱杰.基于兴趣转移的微博用户动态画像生成[J].情报科学, 2021, 39(8): 103-111.DOI: 10.13833/j.issn.1007-7634.2021.08.013.
[15] 王志刚,邱长波.基于主题的政务微博评论用户画像研究[J].情报杂志, 2022, 41(3): 159-165.DOI: 10.3969/j.issn.1002-1965.2022.03.022.
[16] XU K, ZHENG X S, CAI Y, et al. Improving user recommendation by extracting social topics and interest topics of users in uni-directional social networks[J]. Knowl Based Syst, 2018, 140: 120-133. DOI: 10.1016/j.knosys.2017.10.031.
[17] 王战平,夏榕.基于社会化标签挖掘的微博内容推荐方法研究[J].情报科学, 2021, 39(5): 91-96.DOI:10.13833/j.issn.1007-7634.2021.05.013.
[18] BAO F, XU W, FENG Y, et al. A topic-rank recommendation model based on microblog topic relevance &user preference analysis[J]. Hum-Cent Comput Info, 2022, 12(10): 1-19.DOI: https://doi.org/10.22967/HCIS.2022.12.010.
[19] 杨永清,孙凯,张媛媛,等.基于信息画像的突发事故灾难舆情传播效果的预测模型研究[J].情报科学, 2024, 42(4), 27-35. DOI: 10.13833/j.issn.1007-7634.2024.04.004.
[20] 王帅,纪雪梅.基于在线健康社区用户画像的情感表达特征研究[J].情报理论与实践, 2022, 45(6): 179-87. DOI: 10.16353/j.cnki.1000-7490.2022.06.024.
[21] 李铁军,颜端武,杨雄飞.基于情感加权关联规则的微博推荐研究[J].数据分析与知识发现, 2020, 4(4): 27-33. DOI: 10.11925/infotech.2096-3467.2019.0765.
[22] 赵又霖,林怡妮,陆颖隽,等.社会感知数据驱动下用户时空行为画像及语义关联研究[J].图书馆学研究, 2024(2): 54-62. DOI: 10.3969/j.issn.1672-0504.2022.01.011.
[23] ROBERTS K, ROACH M A, JOHNSON J, et al. EmpaTweet: annotating and detecting emotions on twitter[C] //Proc 8th Int Conf Lang Resour Eval LREC 2012, 2012: 3806-3813.DOI: 10.1155/2012/678107.
[24] 张柳,王晰巍,黄博,等.基于LDA模型的新冠肺炎疫情微博用户主题聚类图谱及主题传播路径研究[J].情报学报, 2021, 40(3): 234-244. DOI: 10.3772/j.issn.1000-0135.2021.03.002.
[25] LIU X, BURNS A C, HOU Y J. An investigation of brand-related user-generated content on twitter[J]. J Advert, 2017, 46(2): 236-247. DOI: 10.1080/00913367.2017.1297273.
[26] 张国防,王鑫,徐建民.基于主题词共现的文档非对称关系量化研究[J].数据分析与知识发现, 2023, 7(3): 110-120. DOI: 10.11925/infotech.2096-3467.2022.0342.
[27] 曾子明,陈思语.基于LDA与BERT-BiLSTM-Attention模型的突发公共卫生事件网络舆情演化分析[J].情报理论与实践, 2023, 46(9): 158-166. DOI: 10.16353/j.cnki.1000-7490.2023.09.019.
[28] 徐琳宏,林鸿飞,潘宇,等.情感词汇本体的构造[J].情报学报, 2008, 27(2): 180-185. DOI: 10.3969/j.issn.1000-0135.2008.02.004.
[29] 黄山成,韩东红,乔百友,等.基于 ERNIE2. 0-BiLSTM-Attention 的隐式情感分析方法[J].小型微型计算机系统, 2021, 42(12): 2485-2489. DOI:1000-1220(2021)12-2485-05.
[30] 张金柱,孙雯雯,仇蒙蒙.融合异构网络表示学习与注意力机制的引文推荐研究[J/OL]. 数据分析与知识发现, 1-17 [2024-06-19]. http://kns.cnki.net/kcms/detail/10.1478.g2.20240117.1104.012.html. (
