{"title":"使用深度学习算法识别facebook上可恶的阿姆哈拉语表情包","authors":"Mequanent Degu Belete, Girma Kassa Alitasb","doi":"10.1016/j.sasc.2025.200258","DOIUrl":null,"url":null,"abstract":"<div><div>Hate speech has been disseminated more frequently on social media sites like Facebook in recent years. On Facebook, hate speech can proliferate through text, image, or video. We suggested a deep learning approach to identify offensive memes posted on Facebook in case of Amharic language'. The research process commenced by manually gathering memes posted by Facebook users. Next came textual data extraction, annotation, preprocessing, splitting, feature extraction, model development and assessment Amharic OCRs were employed to extract textual data. Character normalization, stop word removal, and unnecessary character removal make up the text-preprocessing step. Using Stratified KFold the textual dataset is split into the train set (80 %), the validation set (10 %) and the test set (10 %). Vectors are created from the preprocessed texts using the Bog of words (BOW), TFIDF and word embeddings. Following that, the vectors are fed into Machine learning algorithms: NB, DT, RF, KNN, LSVM and LR, and deep learning models that are based on Dense, BiGRU, and BiLSTM algorithms. The model with the optimal parameters is chosen after numerous experiments. With an accuracy rate of 94 %, the BiLSTM + Dense model, the suggested technique identified nasty meme posts on Facebook written in Amharic.</div></div>","PeriodicalId":101205,"journal":{"name":"Systems and Soft Computing","volume":"7 ","pages":"Article 200258"},"PeriodicalIF":3.6000,"publicationDate":"2025-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Identification of hateful amharic language memes on facebook using deep learning algorithms\",\"authors\":\"Mequanent Degu Belete, Girma Kassa Alitasb\",\"doi\":\"10.1016/j.sasc.2025.200258\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><div>Hate speech has been disseminated more frequently on social media sites like Facebook in recent years. On Facebook, hate speech can proliferate through text, image, or video. We suggested a deep learning approach to identify offensive memes posted on Facebook in case of Amharic language'. The research process commenced by manually gathering memes posted by Facebook users. Next came textual data extraction, annotation, preprocessing, splitting, feature extraction, model development and assessment Amharic OCRs were employed to extract textual data. Character normalization, stop word removal, and unnecessary character removal make up the text-preprocessing step. Using Stratified KFold the textual dataset is split into the train set (80 %), the validation set (10 %) and the test set (10 %). Vectors are created from the preprocessed texts using the Bog of words (BOW), TFIDF and word embeddings. Following that, the vectors are fed into Machine learning algorithms: NB, DT, RF, KNN, LSVM and LR, and deep learning models that are based on Dense, BiGRU, and BiLSTM algorithms. The model with the optimal parameters is chosen after numerous experiments. With an accuracy rate of 94 %, the BiLSTM + Dense model, the suggested technique identified nasty meme posts on Facebook written in Amharic.</div></div>\",\"PeriodicalId\":101205,\"journal\":{\"name\":\"Systems and Soft Computing\",\"volume\":\"7 \",\"pages\":\"Article 200258\"},\"PeriodicalIF\":3.6000,\"publicationDate\":\"2025-04-24\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Systems and Soft Computing\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://www.sciencedirect.com/science/article/pii/S2772941925000766\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Systems and Soft Computing","FirstCategoryId":"1085","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2772941925000766","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
摘要
近年来,仇恨言论在Facebook等社交媒体网站上的传播更为频繁。在Facebook上,仇恨言论可以通过文字、图片或视频传播。我们建议采用深度学习方法来识别Facebook上发布的冒犯性表情包,以阿姆哈拉语为例。”研究过程始于手动收集Facebook用户发布的表情包。接下来是文本数据提取、标注、预处理、分割、特征提取、模型开发和评价,使用Amharic ocr提取文本数据。字符规范化、停止词删除和不必要的字符删除组成了文本预处理步骤。使用分层KFold将文本数据集分为训练集(80%)、验证集(10%)和测试集(10%)。向量是从使用Bog of words (BOW)、TFIDF和单词嵌入的预处理文本中创建的。然后,将向量输入机器学习算法:NB、DT、RF、KNN、LSVM和LR,以及基于Dense、BiGRU和BiLSTM算法的深度学习模型。经过多次试验,选择了参数最优的模型。BiLSTM + Dense模型的准确率为94%,该技术可以识别出Facebook上用阿姆哈拉语写的令人讨厌的表情包。
Identification of hateful amharic language memes on facebook using deep learning algorithms
Hate speech has been disseminated more frequently on social media sites like Facebook in recent years. On Facebook, hate speech can proliferate through text, image, or video. We suggested a deep learning approach to identify offensive memes posted on Facebook in case of Amharic language'. The research process commenced by manually gathering memes posted by Facebook users. Next came textual data extraction, annotation, preprocessing, splitting, feature extraction, model development and assessment Amharic OCRs were employed to extract textual data. Character normalization, stop word removal, and unnecessary character removal make up the text-preprocessing step. Using Stratified KFold the textual dataset is split into the train set (80 %), the validation set (10 %) and the test set (10 %). Vectors are created from the preprocessed texts using the Bog of words (BOW), TFIDF and word embeddings. Following that, the vectors are fed into Machine learning algorithms: NB, DT, RF, KNN, LSVM and LR, and deep learning models that are based on Dense, BiGRU, and BiLSTM algorithms. The model with the optimal parameters is chosen after numerous experiments. With an accuracy rate of 94 %, the BiLSTM + Dense model, the suggested technique identified nasty meme posts on Facebook written in Amharic.