arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2007.10718cs.CLcs.IR
收藏

基于孟加拉语文本的人体异常检测

Human Abnormality Detection Based on Bengali Text

Bangladesh University of Business & Technology(孟加拉商业技术大学)

M. F. Mridha, Md. Saifur Rahman, Abu Quwsar Ohi

摘要:

在自然语言处理和人机交互领域,人类的态度和情感一直是研究者关注的焦点。然而,在人机交互领域,人体异常检测还没有得到广泛的研究,大多数工作依赖于基于图像的信息。在自然语言处理中,有效的意义可以潜在地通过所有的词来传达。每个词都可能带来困难的遭遇,因为它们与思想或类别的语义联系。本文介绍了一种高效的人体异常检测模型,该模型只使用孟加拉语文本。该模型可以通过分析输入的孟加拉语文本来识别一个人是处于正常状态还是异常状态。据我们所知,这是第一次尝试在开发一个基于文本的人类异常检测系统。我们创建了孟加拉语数据集(包含2000个句子),它是由自愿对话生成的。我们使用朴素贝叶斯和支持向量机作为分类器进行了比较分析。两种不同的特征提取技术计数向量,TF-IDF被用来实验我们构造的数据集。在我们的实验中,我们构建的数据集达到了最高89%的准确率和92%的F1分数。

英文摘要:

In the field of natural language processing and human-computer interaction, human attitudes and sentiments have attracted the researchers. However, in the field of human-computer interaction, human abnormality detection has not been investigated extensively and most works depend on image-based information. In natural language processing, effective meaning can potentially convey by all words. Each word may bring out difficult encounters because of their semantic connection with ideas or categories. In this paper, an efficient and effective human abnormality detection model is introduced, that only uses Bengali text. This proposed model can recognize whether the person is in a normal or abnormal state by analyzing their typed Bengali text. To the best of our knowledge, this is the first attempt in developing a text based human abnormality detection system. We have created our Bengali dataset (contains 2000 sentences) that is generated by voluntary conversations. We have performed the comparative analysis by using Naive Bayes and Support Vector Machine as classifiers. Two different feature extraction techniques count vector, and TF-IDF is used to experiment on our constructed dataset. We have achieved a maximum 89% accuracy and 92% F1-score with our constructed dataset in our experiment.

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。

↑