/
划分小节，但原文其实是一个比较紧凑的摘要，可能不需要分太多小节，但可以看看有没有自然的部分。不过再仔细看，原文的内容是连贯的，可能先处理成几个
段落？不对，原文是一个整体的摘要，但需要看有没有可以拆分的地方。等一下，原文的内容是：1. MoMo的地位和增长；2. 伴随的欺诈问题；3. 提出的机器学习流水线；4. 预处理和向量化；5. 模型训练和评估指标；6. 评估结果；7. 关键特征；8. 结论和背景。 但规则里说每个自然段落用
包裹，所以可能需要把原文拆分成几个自然的
。比如： 第一个
讲MoMo是加纳首选零售支付渠道，2025年12月交易额和活跃账户数，以及伴随的社会工程欺诈情况（SMS和USSD，欺诈者冒充的对象，获取的信息）。 第二个
讲他们提出的监督机器学习流水线，用于分类诈骗和合法的加纳MoMo SMS消息，使用220条消息的标注语料库，涵盖18个诈骗和合法子类别。 第三个
讲领域特定的归一化（货币金额、电话号码、USSD代码），然后进行TF-IDF向量化（unigrams和bigrams）。 第四个
讲使用分层五折交叉验证和20%留出测试集，训练和评估了多项朴素贝叶斯、线性SVM、随机森林分类器，使用的评估指标（准确率、精确率、召回率、F1、MCC、AUC-ROC）。 第五个
讲留出测试集的准确率、F1、AUC都是1.00（朴素贝叶斯和线性SVM）；交叉验证的准确率平均值。 第六个
讲有区分度的n-gram（紧迫性命令词、索要PIN的词），合法消息的特征（确认和谨慎语言）。 第七个
讲研究发现的背景（撒哈拉以南非洲SMS欺诈检测、小模板语料库的上限效应、向更大自然多语言语料库的路径）。 第八个
讲可解释的经典分类器在结构化MoMo诈骗文本上的高准确率，适合低资源、设备端过滤（运营商或手机层面）。 接下来，需要检查有没有重要的句子需要用。比如关键的结论或重要数据，比如“Held-out accuracy, F1, and AUC were 1.00 for Naive Bayes and linear SVM”或者“Mobile money (MoMo) is Ghana's preferred retail payment channel”可能需要加粗？但规则说重要句子可用，所以可以适当加。 然后，列表内容？原文里没有明显的列表，但比如“Multinomial Naive Bayes, linear support vector machine (SVM), and Random Forest classifiers”是列表吗？可以换成
- 吗？但原文是逗号分隔的，规则里说列表内容用
- ，所以如果原文是并列的内容，比如分类器、评估指标，可以换成列表。比如： “we trained and evaluated
- Multinomial Naive Bayes,
- linear support vector machine (SVM),
- and Random Forest classifiers”这样？但需要看原文的结构，原文中的“Multinomial Naive Bayes, linear support vector machine (SVM), and Random Forest classifiers”是并列的分类器，所以可以换成列表。同样，评估指标“accuracy, precision, recall, F1-score, Matthews correlation coefficient (MCC), and area under the receiver operating characteristic curve (AUC-ROC)”也可以换成列表。 还有，“fraudsters impersonate network operators, agents, relatives, or regulators”这里也是并列的，可以换成列表吗？比如“fraudsters impersonate
- network operators,
- agents,
- relatives,
- or regulators”。 另外，关键的特征词比如“send,” “claim,” “please send”和“PIN,” “pincode”；“balance,” “successful,” “do not share”可以处理成列表吗？比如“imperative urgency
- "send,"
- "claim,"
- "please send"”；“PIN solicitation
- "PIN,"
- "pincode"”；“legitimate messages featured confirmatory and cautionary language
- "balance,"
- "successful,"
- "do not share"”。 现在开始组织HTML： 首先，第一个段落：
Mobile money (MoMo) is Ghana's preferred retail payment channel, recording GHC518.4 billion in transactions in December 2025 and over 26.7 million active accounts (Graphic Online, 2026). This growth has coincided with increased social-engineering fraud over Short Message Service (SMS) and Unstructured Supplementary Service Data (USSD), where fraudsters impersonate
- network operators,
- agents,
- relatives, or
- regulators to obtain personal identification numbers (PINs), identity data, or funds.
不对，列表应该在适当的位置，比如“impersonate”后面接列表：
Mobile money (MoMo) is Ghana's preferred retail payment channel, recording GHC518.4 billion in transactions in December 2025 and over 26.7 million active accounts (Graphic Online, 2026). This growth has coincided with increased social-engineering fraud over Short Message Service (SMS) and Unstructured Supplementary Service Data (USSD), where fraudsters impersonate
- network operators,
- agents,
- relatives, or
- regulators to obtain personal identification numbers (PINs), identity data, or funds.
然后第二个段落：
We present a supervised machine-learning pipeline for classifying scam and legitimate Ghanaian MoMo SMS messages using a labelled corpus of 220 messages across 18 scam and legitimate subcategories.
第三个段落：Domain-specific normalisation of currency amounts, phone numbers, and USSD codes preceded TF-IDF vectorisation of unigrams and bigrams.
第四个段落：Using stratified five-fold cross-validation and a 20% held-out test set, we trained and evaluated the following classifiers:
- Multinomial Naive Bayes,
- linear support vector machine (SVM), and
- Random Forest. The evaluation metrics included
- accuracy,
- precision,
- recall,
- F1-score,
- Matthews correlation coefficient (MCC), and
- area under the receiver operating characteristic curve (AUC-ROC).
不对，应该是：
Using stratified five-fold cross-validation and a 20% held-out test set, we trained and evaluated the following classifiers:
- Multinomial Naive Bayes,
- linear support vector machine (SVM), and
- Random Forest.
The evaluation metrics included
- accuracy,
- precision,
- recall,
- F1-score,
- Matthews correlation coefficient (MCC), and
- area under the receiver operating characteristic curve (AUC-ROC).
或者合并成一段，但列表要正确嵌套？不对，列表应该是独立的，比如：
Using stratified five-fold cross-validation and a 20% held-out test set, we trained and evaluated
- Multinomial Naive Bayes,
- linear support vector machine (SVM), and
- Random Forest classifiers using the following metrics:
- accuracy,
- precision,
- recall,
- F1-score,
- Matthews correlation coefficient (MCC), and
- area under the receiver operating characteristic curve (AUC-ROC).
这样可能更紧凑。 然后第五个段落：
Held-out accuracy, F1, and AUC were 1
- ，所以如果原文是并列的内容，比如分类器、评估指标，可以换成列表。比如： “we trained and evaluated