基于语义共现匹配的在线食品安全谣言相关文档识别方法研究

Research on the Methods to Identify Online Food Safety Rumor-Related Documents Based on Semantic Co-occurrence Matching

[目的/意义]文章通过设计一种有效的在线食品安全谣言相关文档识别方法,从而提升人工审核的效率,减轻在线食品安全谣言传播带来的不良影响。[方法/过程]基于待分类文档中的词语分布在不同类型的特征向量库(在线食品安全谣言相关库以及非相关库)中的差异性,设计了基于无监督的文档特征相似性计算方法以及基于有监督的回归方法来识别在线食品安全谣言相关文档。[结果/结论]基于无监督的RM-Sort方法能够有效地识别在线食品安全谣言相关文档,并且优于现有的朴素贝叶斯,决策树以及支持向量机方法。进一步地,基于有监督的RM-LR方法效果则更优。[局限]模型只能够判别文档是否和食品谣言相关,但无法确定该文档是辟谣文章还是谣言文章本身。

[Purpose/significance] The methods proposed by this paper can effectively detect online food safety rumor-related documents,and save a lot of review time for the administrators of social platforms to slower or even stop the online food safety rumor propagations. [Method/process] Based on the difference of the distribution of words features at different word vector spaces,including online food safety rumor correlated corpus and uncorrelated corpus,this paper designs an unsupervised documents similarity calculation method,and a supervised regression method to detect online food safety rumor-related documents. [Result/conclusion] The unsupervised RM-Sort method can effectively detect online food safety rumor-related documents,which is better than the methods of Naive Bayesian,Decision Tree,and Support Vector Machines. Moreover,the supervised RM-LR method shows the best results among all baseline methods. [Limitations] The models proposed can only identify whether the document is related to food safety rumor,but can not identify whether the document is a refuting rumor article or rumor article itself.