Research On Microblog Topic Recognition Based On Neuro-semantic Topic

In recent years,Internet-related technologies have experienced rapid development,especially the emergence of social media network platforms such as Weibo and Twitter,making social media an important medium for the media and the public to share news events and life emotions.Sina Weibo has benefited from its wide application platform,simple operation logic,and extremely fast information dissemination.It has quickly become a new social media network platform with huge scale and influence in China.Sina Weibo has tens of millions of daily posts,and the text information contained in it is huge.In response to huge microblog text data,how to carry out reasonable and efficient information integration and extract hot topics from them is the core issue of text topic mining.In this paper,based on the Neural Topic Model(NTM),the topic features of the short text of Weibo are not fully expressed for the traditional text topic model,the Weibo topic mining is not accurate enough,and the mined Weibo topics lack semantic information Problem,fully considering the characteristics of microblog text,a microblog topic model based on neural semantics enhancement(MNTS)is proposed.Firstly,the extracted microblog text corpus is divided into two parts: a microblog summary and a microblog article,and different treatments are given based on the characteristics of the two parts.The microblog summary is a short text containing a large amount of microblog theme information It uses the semantic word vectors of Weibo to construct a single-channel text feature matrix to make up for the shortcomings of the original neural theme model,such as high input vector space dimension,sparse features and ignoring a large amount of text information.The Weibo blog post is an expanded description of the Weibo profile,which contains the weight information of the terms in the Weibo profile,so the weight of the terms in the Weibo profile is determined by the similarity of the terms in the Weibo blog post,and the semantics of the Weibo profile is determined The word vector is combined with the weight information of the Weibo blog post as the text feature of the final input Weibo topic model.Secondly,the model adds additional text feature representations of the bag-of-words model to the output,and achieves the simultaneous training of topics and semantics.Comparative experiments on real Weibo corpus show that the MNTS topic model not only improves the accuracy of identifying topics,but also increases the semantic coherence of mining topics by a maximum of 0.6.In order to make better use of Weibo information to fully explore Weibo topics,analogous to the conditional variational self-encoding network,the category label is introduced as a condition for the MNTS topic model,and more guidance information is added for topic model identification.Due to the large and sparse lexical item dimensions,the lexical item probability value in the topic-term distribution is not obvious,especially for the test microblog text.Although the topic of the test microblog text can be found,the terms under the topic do not The subject of the test Weibo text cannot be fully expressed.In this case,a review text is introduced to check the terms under the subject.The introduction of category labels and comment text further enhances the topic mining ability of the MNTS topic model.