Two-Phase Weibo Short Text Summarization Method For Psychological Crisis

Nowadays,with the rapid development of the national economy,the pressure that individuals face in society is increasing,and it is easy to cause psychological crisis problems in this state for a long time.But most people who are in a psychological crisis will not seek the help of relevant medical personnel or society,but they will inadvertently show negative or even suicidal thoughts when talking with people.Although there are many kinds of automatic summarization,including news,microblog,emotion and other data,there is no literature about the automatic summary of psychological crisis.This paper proposes a two-phase short text summarization method for psychological crisis problems,which includes two phases: recognition of psychological crisis microblog based on multi feature fusion and extraction of psychological crisis problems.In the first phase,microblogs with psychological crisis problems are identified.Firstly,the influence of various features on the recognition of microblog of psychological crisis is analyzed,and then the micro blog including psychological crisis is selected by quantifying the attribute features,language features and word vector features in the text of microblog.It is found that word vector features have better performance in single category features,and it is more conducive to filter out more unrelated microblogs and find out candidate microblogs containing psychological crisis problems after feature selection.In the second phase,we extract the summary of microblog with psychological crisis.First of all,we use TextRank algorithm to generate the undirected graph of microblog,and give different weights to each microblog.But the classic TextRank method considers the overall information of microblog,and ignores the semantic features of microblog itself.In order to express the semantics of microblog more accurately,a text vector representation model combining word2 vec and TF-IWF is proposed.In order to strengthen the impact of psychological crisis information on the results of the summary,using the features of words such as emotional words contained in microblog,combining degree adverbs and negative words,the author analyzes the degree of psychological crisis,makes corresponding weight adjustment strategies,and adjusts the weight of microblog appropriately.Finally,the score of microblog is calculated according to the weight.According to the ranking from high to low,choose K microblogs with the highest score are selected in order,and the target summary is generated according to the time of microblog publishing.In the first phase of the recognition of psychological crisis microblog,word vector word2 vec is used to fuse other single features with obvious differentiation.When identifying psychological crisis micro blog,the accuracy and F value can reach 0.17 and 0.32,which is better than single features.In the second phase,the weight adjustment strategy of "TFIWF_w2v" makes the average accuracy of the summary extraction reach 0.433,which is higher than 0.267 of the classic TextRank.At the same time,the AVG_precision,AVG_recall and AVG_F values of the two-phase algorithm are higher than those of the classical TextRank.It shows that the algorithm proposed in this paper is effective.