arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2012.02104cs.SIcs.AIcs.CLcs.LG
收藏

在短文档中产生可解释模型的判别表达式

Discriminatory Expressions to Produce Interpretable Models in Short Documents

University of Granada(格拉纳达大学)

Manuel Francisco, Juan Luis Castro

摘要:

社交网站(SNS)是最重要的沟通方式之一。特别是,微博网站由于其特殊性(简洁,短文本.)而被用作分析途径。有无数的研究以新颖的方式使用SNS,但机器学习主要集中在分类性能,而不是可解释性和/或其他良好的指标。因此,最先进的模型是黑箱,不应该用来解决可能产生社会影响的问题。当问题需要透明性时,有必要构建可解释的管道。虽然分类器可能是可解释的,但产生的模型太复杂而无法理解,使得人类无法理解实际的决策。本文提出了一种特征选择机制,该机制能够通过使用更少但更有意义的特征来提高可理解性,同时在可解释性是强制性的微博上下文中实现良好的性能。此外,我们提出了一个排名方法来评估功能的统计相关性和偏见。我们用五个不同的数据集进行了详尽的测试,以评估分类性能,泛化能力和模型的复杂性。结果表明,我们的建议是更好的和最稳定的准确性,概括性和可理解性。

英文摘要:

Social Networking Sites (SNS) are one of the most important ways of communication. In particular, microblogging sites are being used as analysis avenues due to their peculiarities (promptness, short texts...). There are countless researches that use SNS in novel manners, but machine learning has focused mainly in classification performance rather than interpretability and/or other goodness metrics. Thus, state-of-the-art models are black boxes that should not be used to solve problems that may have a social impact. When the problem requires transparency, it is necessary to build interpretable pipelines. Although the classifier may be interpretable, resulting models are too complex to be considered comprehensible, making it impossible for humans to understand the actual decisions. This paper presents a feature selection mechanism that is able to improve comprehensibility by using less but more meaningful features while achieving good performance in microblogging contexts where interpretability is mandatory. Moreover, we present a ranking method to evaluate features in terms of statistical relevance and bias. We conducted exhaustive tests with five different datasets in order to evaluate classification performance, generalisation capacity and complexity of the model. Results show that our proposal is better and the most stable one in terms of accuracy, generalisation and comprehensibility.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑