Pular para o conteúdo

Spam filtering

Implement intelligent spam detection with keyword analysis, domain reputation, and machine learning techniques

Atualizado em Ver como Markdown

Build spam filtering systems with keyword matching, domain validation, and intelligent detection methods for effective email security.

Basic spam filter

Simple spam detection with keyword matching and domain validation:

interface Env {
	EMAIL: SendEmail;
	EMAIL_ANALYTICS: AnalyticsEngine;
}

interface SpamFilter {
	checkSpam(
		message: any,
	): Promise<{ isSpam: boolean; score: number; reasons: string[] }>;
}

class SimpleSpamFilter implements SpamFilter {
	private spamKeywords = [
		"buy now",
		"limited time",
		"act fast",
		"click here",
		"free money",
		"guaranteed",
		"risk free",
		"urgent",
		"winner",
		"congratulations",
		"inheritance",
		"lottery",
	];

	private trustedDomains = ["example.com", "trusted-partner.com", "vendor.net"];

	async checkSpam(
		message,
	): Promise<{ isSpam: boolean; score: number; reasons: string[] }> {
		let score = 0;
		const reasons = [];

		const sender = message.from;
		const subject = message.headers.get("subject") || "";
		const senderDomain = sender.split("@")[1];

		// Check sender domain
		if (this.trustedDomains.includes(senderDomain)) {
			score -= 2; // Trusted sender
		}

		// Check subject for spam keywords
		const subjectLower = subject.toLowerCase();
		for (const keyword of this.spamKeywords) {
			if (subjectLower.includes(keyword)) {
				score += 1;
				reasons.push(`Spam keyword: ${keyword}`);
			}
		}

		// Check for excessive capitalization
		const capsRatio = (subject.match(/[A-Z]/g) || []).length / subject.length;
		if (capsRatio > 0.7 && subject.length > 10) {
			score += 1;
			reasons.push("Excessive capitalization");
		}

		// Check for suspicious patterns
		if (subject.includes("!!!") || subject.includes("$$$")) {
			score += 1;
			reasons.push("Suspicious punctuation");
		}

		// Check for suspicious sender patterns
		if (
			sender.includes("noreply") &&
			subject.toLowerCase().includes("urgent")
		) {
			score += 2;
			reasons.push("Suspicious noreply + urgent combination");
		}

		return {
			isSpam: score >= 2,
			score,
			reasons,
		};
	}
}

const spamFilter = new SimpleSpamFilter();

export default {
	async email(message, env, ctx): Promise<void> {
		const startTime = Date.now();

		// Check for spam
		const spamCheck = await spamFilter.checkSpam(message);

		// Track spam check metrics
		env.EMAIL_ANALYTICS?.writeDataPoint({
			blobs: [
				"spam_check_completed",
				message.from,
				message.to,
				spamCheck.isSpam ? "spam" : "legitimate",
			],
			doubles: [
				1, // Count
				spamCheck.score,
				Date.now() - startTime,
			],
			indexes: [
				`spam_detected:${spamCheck.isSpam}`,
				`score_range:${getScoreRange(spamCheck.score)}`,
			],
		});

		if (spamCheck.isSpam) {
			console.log(
				`Rejected spam email from ${message.from}: ${spamCheck.reasons.join(", ")}`,
			);
			message.setReject(`Message rejected: ${spamCheck.reasons[0]}`);
			return;
		}

		// Add spam score headers and forward
		const headers = new Headers();
		headers.set("X-Spam-Score", spamCheck.score.toString());
		headers.set("X-Spam-Reasons", spamCheck.reasons.join(", "));
		headers.set("X-Spam-Check-Time", (Date.now() - startTime).toString());

		await message.forward("inbox@example.com", headers);
	},
};

function getScoreRange(score: number): string {
	if (score < 0) return "trusted";
	if (score === 0) return "neutral";
	if (score === 1) return "suspicious";
	return "spam";
}

Advanced spam detection with AI

For more sophisticated spam detection, you can enhance the basic filter using Workers AI to analyze email content with machine learning models. This approach can identify subtle spam patterns that keyword-based filters might miss.

Next steps