An Auto-ethnographic Observation:Hashtag Activism in Chinese Post-Feminist Age

This paper provides an auto-ethnographic observation of hashtag feminist activism on Weibo, setting in a context of post-feminism age in China. Two subjects, the Hot Search List and its Public Discussion Forum, were chose to examine the complexity of the current situation of this hashtag activism. An auto-ethnographic methodology was used to interrogate the states quo of Chinese online feminist movement, revealing gender-centric discussions reinforcing stereotypes under the guise of equality. Misogynistic narratives, algorithmic constraints, censorship, and official opposition pose significant barriers to feminist discourse. Nonetheless, the study identifies a potential for hashtag activism within Weibo’s discourse, offering a space for resistance. By acknowledging these challenges, this paper seeks to empower Chinese feminists to challenge dominant narratives and advocate for their rights.

中文微博的Hashtag话题相关性分析

Hashtag(微博话题词)是发布者为微博信息创建的话题标签,能帮助用户在海量微博数据中高效发现热点话题。Hashtag由用户创建的特性使得不同的Hashtag可能代表着同一个话题,挖掘Hashtag之间的话题相关性将有助于热点话题发现和聚合展示。研究了Hashtag之间相关性分析问题,抽取了Hashtag文本特征、微博内容、Hashtag的出现次数-时间分布以及Hashtag共现等一系列特征,以分析Hashtag之间的话题相关性。在新浪微博数据上的实验结果显示,这一系列特征组合能较好地帮助Hashtag相关性分析。

文本表示方法对微博Hashtag推荐影响研究——以Twitter上H7N9微博为例

在总结国内外Hashtag推荐方法和短文本表示方法的基础上,文章利用基于K最近邻(KNN)的Hashtag推荐方法,将微博文本表示为向量然后计算相似度,从语料中选出与目标微博最相似的微博文本,然后抽取候选Hashtag。文章比较了向量空间模型(VSM)、潜在语义分析模型(LSA)、隐含狄利克雷分布模型(LDA)、深度学习(DL)等四种文本表示方法对基于KNN的Hashtag推荐效果的影响。以Twitter上H7N9微博为测试数据,实验结果表明深度学习的文本表示方法在基于KNN的Hashtag推荐中取得最好的效果。

基于小样本学习的个性化Hashtag推荐

近年来,Hashtag推荐任务吸引了很多研究者的关注。目前,大部分深度学习方法把这个任务看作是一个多标签分类问题,将Hashtag看作为微博的类别。但是这些方法的输出空间固定,在没有进行重新训练的情况下,不能处理训练不可见的Hashtag。然而,实际上Hashtag会随着时事热点不断快速更新。为了解决这一问题,该文提出将Hashtag推荐任务建模成小样本学习任务。同时,结合用户使用Hashtag的偏好降低推荐的复杂度。在真实的推特数据集上的实验表明,与目前最优方法相比,该模型不仅可以取得更好的推荐结果,而且表现得更为鲁棒。

Hashtag Recommendation Using LSTM Networks with Self-Attention

On Twitter,people often use hashtags to mark the subject of a tweet.Tweets have specific themes or content that are easy for people to manage.With the increase in the number of tweets,how to automatically recommend hashtags for tweets has received wide attention.The previous hashtag recommendation methods were to convert the task into a multi-class classification problem.However,these methods can only recommend hashtags that appeared in historical information,and cannot recommend the new ones.In this work,we extend the self-attention mechanism to turn the hashtag recommendation task into a sequence labeling task.To train and evaluate the proposed method,we used the real tweet data which is collected from Twitter.Experimental results show that the proposed method can be significantly better than the most advanced method.Compared with the state-of-the-art methods,the accuracy of our method has been increased 4%.

Hashtags as Crowdsourcing: A Case Study of Arabic Hashtags on Twitter

This mixed study aims to highlight the impact of social media in the Arab world, specifically Twitter’s impact on translators’ communities. For this purpose, the role of hashtags among translators will be examined by investigating one particular Arabic hashtag, its purpose, target users, and the classification of content. The hashtag is , #translator_serving_translator. 1) An online survey of six closed questions was employed and posted on Twitter, and 249 responses show that users are from fourteen Arab countries, and the majority is from Saudi Arabia. Hashtag users are translators, freelancers, or TS students. Some are active users who post tweets and answer questions, others only ask questions, and the rest only read tweets. The general attitude toward employing hashtags among translators’ communities was positive. 2) Employing a content analysis approach, the content is classified into two main categories of sharing information and seeking assistance with seven subcategories of each.

“避男标签”:小红书女性用户的算法抵抗研究

互联网平台的发展使算法日益成为一种新型社会权力,影响着人们在数字社会的行为决策,也激发了人们对算法的警惕和抵抗。本文聚焦小红书平台的“避男标签”现象,探究小红书女性用户如何基于算法想象,使用特定的话题标签开展算法抵抗。研究发现,女性使用“避男标签”是为躲避网络社区的男性凝视和网络骚扰,在具体实践中主要通过多元化标签内容以实现多维度的算法抵抗。但“避男标签”的效果并不稳定,需要及时调整,开展动态的算法博弈。总体来看,女性通过创造性使用平台功能开展算法抵抗,是为了寻求性别平等、包容友好的网络空间。而解决问题的关键不仅仅在于个体的能动抵抗,还要积极推动网络空间治理。

Hashtag研究综述

【目的】分析当前Hashtag研究思路和技术,归纳和总结当前Hashtag研究中所存在的问题,并提炼Hashtag研究的理论意义与实际意义,为更深入的Hashtag研究提供参考。【文献范围】以2007年至2015年的国际会议和国内外期刊的60篇文献作为主要研究对象。【方法】调研Hashtag研究及其应用的相关文献,对Hashtag研究中各环节涉及的方法进行分析和总结。【结果】Hashtag在用户使用、Hashtag挖掘与基于Hashtag的应用研究三方面存在一些可以深入研究的问题。【结论】未来应侧重于Hashtag的理论研究,如用户标注Hashtag的动机、影响Hashtag标注的因素等。在实际应用中,结合不同学科方法和多个领域的技术改善Hashtag在实际应用中的效果。

Combining long-term and short-term user interest for personalized hashtag recommendation

Hashtags, terms prefixed by a hash-symbol #, are widely used and inserted anywhere within short messages （tweets） on micro-blogging systems as they present rich sen- timent information on topics that people are interested in. In this paper, we focus on the problem of hashtag recommenda- tion considering their personalized and temporal aspects. As far as we know, this is the first work addressing this issue spe- cially to recommend personalized hashtags combining long- term and short-term user interest. We introduce three features to capture personal and temporal user interest： 1） hashtag textual information; 2） user behavior; and 3） time. We of- fer two recommendation models for comparison： a linear- combined model, and an enhanced session-based temporal graph （STG） model, Topic-STG, considering the features to learn user preferences and subsequently recommend person- alized hashtags. Experiments on two real tweet datasets illus- trate the effectiveness of the proposed models and algorithms.

Hashtag Recommendation Based on Multi-Features of Microblogs

Hashtag recommendation for microblogs is a very hot research topic that is useful to many applications involving microblogs. However, since short text in microblogs and low utilization rate of hashtags will lead to the data sparsity problem, it is difficult for typical hashtag recommendation methods to achieve accurate recommendation. In light of this, we propose HRMF, a hashtag recommendation method based on multi-features of microblogs in this article. First, our HRMF expands short text into long text, and then it simultaneously models multi-features （i.e., user, hashtag, text） of microblogs by designing a new topic model. To further alleviate the data sparsity problem, HRMF exploits hashtags of both similar users and similar microblogs as the candidate hashtags. In particular, to find similar users, HRMF combines the designed topic model with typical user-based collaborative filtering method. Finally, we realize hashtag recommendation by calculating the recommended score of each hashtag based on the generated topical representations of multi-features. Experimental results on a real-world dataset crawled from Sina Weibo demonstrate the effectiveness of our HRMF for hashtag recommendation.

Modeling Chinese Microblogs with Five Ws for Topic Hashtags Extraction

Hashtags are important metadata in microblogs and are used to mark topics or index messages. However,statistics show that hashtags are absent from most microblogs. This poses great challenges for the retrieval and analysis of these tagless microblogs. In this paper, we summarize the similarity between microblogs and shortmessage-style news, and then propose an algorithm, named 5WTAG, for detecting microblog topics based on a model of five Ws（When, Where, Who, What, ho W）. As five-W attributes are the core components in event description, it is guaranteed theoretically that 5WTAG can properly extract semantic topics from microblogs. We introduce the detailed procedure of the algorithm in this paper including spam microblog identification, microblog segmentation, and candidate hashtag construction. In addition, we propose a novel recommendation computing method for ranking candidate hashtags, which combines syntax and semantic analysis and observes the distribution of artificial topic hashtags. Finally, we conduct comprehensive experiments to verify the semantic correctness and completeness of the candidate hashtags, as well as the accuracy of the recommendation method using real data from Sina Weibo.

一种融合聚类和时间信息的微博排序新方法

随着微博的快速发展,微博检索已经成为近年来研究领域的热点之一。微博检索与传统文本检索在两个方面明显不同:一是微博具有自己的特点,表现在文本短和内容中具有主题概括词(称为Hashtag);二是微博排序中除了考虑文本和语义相似度,还需考虑时间信息。根据这两点区别,该文在统计语言模型的基础上,使用聚类进行文本扩展,并将Hashtag信息运用到聚类过程中。同时,因为微博数据集中具有Hashtag的微博个数不超过13%,针对这一现象,该文还提出了一种扩展微博Hashtag的方法,最终提出了基于聚类的三个模型。然后通过定义文档先验将时间信息加入到提出的三个检索模型中,得到融入聚类和时间信息的三个模型。最后基于TREC Microblog数据的实验结果证明,融合聚类信息和时间信息的模型在MAP和P@30上有明显提高,分别提高7.1%和11.6%。

融合社交关系和标签信息的混合新闻推荐算法

针对传统新闻推荐的数据稀疏性和用户的兴趣爱好快速变化问题,提出了一种融合社交关系和标签信息的混合新闻推荐算法。首先,该算法充分利用社交网络中的社交关系和标签信息;然后使用概率主题模型(latent Dirichlet allocation,LDA)对用户兴趣进行建模;最后采用基于内容与协同过滤相结合的混合推荐算法来完成新闻推荐。实验结果表明,所提算法与已有的推荐算法相比较,在精确度上提升了10.7%、平均倒数排名上(mean reciprocal rank,MRR)提升了4.1%,在归一化折损累计增益(normalized discounted cumulative gain,NDCG)上提升了10%。该算法可在一定程度上提高新闻推荐算法的精度及推荐质量。

Library and Information Science Papers Discussed on Twitter: A new Network-based Approach for Measuring Public Attention

Purpose: In recent years, one can witness a trend in research evaluation to measure the impact on society or attention to research by society(beyond science). We address the following question: can Twitter be meaningfully used for the mapping of public and scientific discourses?Design/methodology/approach: Recently, Haunschild et al.(2019) introduced a new network-oriented approach for using Twitter data in research evaluation. Such a procedure can be used to measure the public discussion around a specific field or topic. In this study, we used all papers published in the Web of Science(WoS, Clarivate Analytics) subject category Information Science & Library Science to explore the publicly discussed topics from the area of library and information science(LIS) in comparison to the topics used by scholars in their publications in this area.Findings: The results show that LIS papers are represented rather well on Twitter. Similar topics appear in the networks of author keywords of all LIS papers, not tweeted LIS papers, and tweeted LIS papers. The networks of the author keywords of all LIS papers and not tweeted LIS papers are most similar to each other.Research limitations: Only papers published since 2011 with DOI were analyzed.Practical implications: Although Twitter data do not seem to be useful for quantitative research evaluation, it seems that Twitter data can be used in a more qualitative way for mapping of public and scientific discourses.Originality/value: This study explores a rather new methodology for comparing public and scientific discourses.

基于话题标签的微博主题挖掘

随着互联网的发展,微博已成为人们获取信息的主要平台,为从海量微博中挖掘出有价值的主题信息,结合微博中的会话、转发和话题标签,将微博划分为用户兴趣、用户互动和话题微博3类,提出基于作者主题模型(ATM)的话题标签主题模型HC-ATM,使用Gibbs抽样法对模型进行推导,获取微博主题结构。在Twitter数据集上的实验结果表明,与ATM模型和基于潜在狄利克雷分布的微博生成模型相比,HC-ATM模型的主题困惑度更小、差异度更大,并且能有效挖掘出不同微博类型的主题分布。

基于标签的微博关键词抽取排序方法

针对微博关键词抽取准确率不高的问题,提出一种基于标签优先的抽取排序方法。该方法利用微博本身具有的社交特征——标签,从微博内容集中抽取关键词。该方法首先根据微博自身建立初始词与微博之间的加权图,再将基于标签的随机游走方法应用于图中,随机游走反复跳跃到标签词节点上,经过一系列迭代得出每个词的平稳概率,并通过概率决定词的最终排序。该抽取方法根据真实的新浪微博内容进行测验,结果显示,与通过词与词的加权图来抽取关键词相比,基于标签的微博关键词抽取方法准确率提高了50%,在实际应用中能够有效提高关键词抽取的正确率。

基于图文注意力融合的主题标签推荐

为了解决社交媒体平台上的信息超载问题,帮助用户快速捕捉所需信息,对基于多模态内容的标签推荐问题进行研究。针对不同模态间的异质性差异,采用共注意力机制进行跨模态内容的特征建模与融合;针对多标签分类方法只能推荐出数据集标签空间中标签的不足,采用Seq2Seq框架生成新的标签序列,并通过一种聚合策略将分类方法的推荐结果聚合到生成的标签序列中,得到2种方法的统一推荐模型。在大规模数据集上的实验结果表明:多模态方法比单模态方法更具优势,所提出的统一推荐模型的F1值比仅使用单模态的对比模型高9.44百分点;生成新标签序列的方法也优于传统的分类方法,所提出的标签序列生成模型的F1值比对比模型COA高3.41百分点;所提出的统一推荐模型UNIFIED-CO-ATT的F1值比GEN-CO-ATT模型高1.25百分点,其效果优于其他对比模型。所提出的模型综合了分类方法和生成方法的特点,可以使推荐的标签同时具有准确性和新颖性。

面向社交媒体的分级注意力表情符预测模型

将预测社交媒体表情符的任务作为文本分类问题,将输入文本映射到最有可能的伴随表情符号。首先,通过研究帖子中出现的表情符与标签之间的关系,提出一个基于标签、发帖用户、发帖时间、发帖地点的注意力机制;其次,添加表情符位置特征;最后,探讨注意力机制、分级模型对于表情符预测任务的作用,训练多种模型并比较其预测效果。实验结果表明,模型对于不同使用频率的表情符的预测效果均有显著提升,模型是可行的、高效的。

面向微博短文本的社交与概念化语义扩展搜索方法

充分挖掘微博短文本的语义以实现精准搜索是一项重要任务.由于微博文本内容具有稀疏性和语义局限性的特点,使得仅通过分析字面语义来进行短文本理解和相似性匹配的传统搜索方法受到了一定的限制.因此提出了一种社交与概念化语义结合的扩展搜索方法,通过挖掘社交网络独特的社交属性如#标签#、"@"和链接信息URL,对微博短文本实现进一步的社交语义扩展.该方法将文本字面分析获取的概念词语和社交关系中潜在的关联标签信息相结合,对短文本进行2种角度下的语义特征表示,实现了基于微博短文本语义充分理解的精准搜索.在微博数据集上的对比实验表明,与已有的扩展搜索方法相比所提方法能捕捉更多的语义特征,微博搜索的性能也得到了显著的提升.

网络用户自描述标签向量生成及标签层次体系构建方法

通过用户自描述标签可以快速、高效地对网络用户进行检索、分类和智能推荐。为更好地研究网络用户自描述标签的内在关系,提出了一种网络用户自描述标签层次体系的构建方法。通过对低频标签的理解方法定义了标签向量生成规则。利用搜索引擎语料扩充的方法提取上下位关系的关键模式,实现了一种基于多特征的标签对上下位关系检测方法。以标签频率作为权重,介绍了构建网络用户自描述标签层次树的思路。实验结果表明,本文提出的上下位关系检测方法在精确度、查准率、查全率和F1等指标上较以往的相关工作有大幅度提升。