基于双向量模型的自适应微博话题追踪方法

Self-adaptive Method Based on Double-vector Model for Microblog Topic Tracking

针对微博文本篇幅短小、网络新词层出不穷等特点以及在话题发展过程中产生的漂移问题,提出了基于双向量模型的自适应微博话题追踪方法.该方法首先提出双向量模型,将文本用词嵌入和VSM向量空间模型两种方法分别向量化,保留文本语义的同时也解决了微博新词问题.其次,将话题和微博分别用双向量模型表示,计算话题双向量模型和微博双向量模型的余弦相似度作为话题与微博的相似度.接着,将话题与微博的相似度与自适应学习获得的相似度阈值进行比较,判定微博是否为话题相关微博.最后,自适应更新话题模型,能够有效地应对微博话题发展所产生的漂移.实验结果表明,该方法能够实时地跟踪话题并降低了话题相关微博的漏检率和误检率.

In order to handle the characteristics of microblog such as short texts,continuous emergence of network neologisms and topic drifting,an adaptive microblog topic tracking method based on Double-Vector model is proposed. Firstly,a Double-Vector model is proposed to transform texts into vectors with word embedding technology and VSM( Vector Space Model),so that the text semantics is preserved and the problem of microblog neologisms is solved. Secondly,the similarity between a microblog and a topic is represented by the cosine value of the Double-Vector model of the microblog and the Double-Vector model of the topic. Thirdly,the similarity between a microblog and a topic is compares with the similarity threshold that is obtained by self-adaptive learning to determine whether the microblog is topic relevant microblog or not. Finally,through self-adaptive updating the topic model,the topic drift aroused by the development of microblog topics can be effectively overcomed. Experimental results show that the proposed method can effectively track the changes of the topic in real time and reduce the missing rate and false positive rate of the topic related microblog.