feat: classify emails by importance based on subjects

appinfo/info.xml

@@ -34,7 +34,7 @@ The rating depends on the installed text processing backend. See [the rating ove
 
 Learn more about the Nextcloud Ethical AI Rating [in our blog](https://nextcloud.com/blog/nextcloud-ethical-ai-rating/).
 	]]></description>
-	<version>4.1.0-alpha.2</version>
+	<version>4.1.0-alpha.3</version>
 	<licence>agpl</licence>
 	<author homepage="https://github.com/ChristophWurst">Christoph Wurst</author>
 	<author homepage="https://github.com/GretaD">GretaD</author>
@@ -90,6 +90,8 @@ Learn more about the Nextcloud Ethical AI Rating [in our blog](https://nextcloud
 		<command>OCA\Mail\Command\TrainAccount</command>
 		<command>OCA\Mail\Command\UpdateAccount</command>
 		<command>OCA\Mail\Command\UpdateSystemAutoresponders</command>
+		<command>OCA\Mail\Command\PreprocessAccount</command>
+		<command>OCA\Mail\Command\RunMetaEstimator</command>
 	</commands>
 	<settings>
 		<admin>OCA\Mail\Settings\AdminSettings</admin>

lib/AppInfo/Application.php

@@ -45,7 +45,6 @@
 use OCA\Mail\Listener\MessageCacheUpdaterListener;
 use OCA\Mail\Listener\MessageKnownSinceListener;
 use OCA\Mail\Listener\MoveJunkListener;
-use OCA\Mail\Listener\NewMessageClassificationListener;
 use OCA\Mail\Listener\NewMessagesNotifier;
 use OCA\Mail\Listener\OauthTokenRefreshListener;
 use OCA\Mail\Listener\OptionalIndicesListener;
@@ -124,7 +123,6 @@ public function register(IRegistrationContext $context): void {
 		$context->registerEventListener(MessageDeletedEvent::class, MessageCacheUpdaterListener::class);
 		$context->registerEventListener(MessageSentEvent::class, AddressCollectionListener::class);
 		$context->registerEventListener(MessageSentEvent::class, InteractionListener::class);
-		$context->registerEventListener(NewMessagesSynchronized::class, NewMessageClassificationListener::class);
 		$context->registerEventListener(NewMessagesSynchronized::class, MessageKnownSinceListener::class);
 		$context->registerEventListener(NewMessagesSynchronized::class, NewMessagesNotifier::class);
 		$context->registerEventListener(SynchronizationEvent::class, AccountSynchronizedThreadUpdaterListener::class);

lib/BackgroundJob/TrainImportanceClassifierJob.php

@@ -69,10 +69,7 @@ protected function run($argument) {
 		}
 
 		try {
-			$this->classifier->train(
-				$account,
-				$this->logger
-			);
+			$this->classifier->train($account, $this->logger);
 		} catch (Throwable $e) {
 			$this->logger->error('Cron importance classifier training failed: ' . $e->getMessage(), [
 				'exception' => $e,

lib/Command/PredictImportance.php

@@ -13,6 +13,7 @@
 use OCA\Mail\Db\Message;
 use OCA\Mail\Service\AccountService;
 use OCA\Mail\Service\Classification\ImportanceClassifier;
+use OCA\Mail\Support\ConsoleLoggerDecorator;
 use OCP\AppFramework\Db\DoesNotExistException;
 use OCP\IConfig;
 use Psr\Log\LoggerInterface;
@@ -25,6 +26,7 @@
 class PredictImportance extends Command {
 	public const ARGUMENT_ACCOUNT_ID = 'account-id';
 	public const ARGUMENT_SENDER = 'sender';
+	public const ARGUMENT_SUBJECT = 'subject';
 
 	private AccountService $accountService;
 	private ImportanceClassifier $classifier;
@@ -43,26 +45,27 @@ public function __construct(AccountService $service,
 		$this->config = $config;
 	}
 
-	/**
-	 * @return void
-	 */
-	protected function configure() {
+	protected function configure(): void {
 		$this->setName('mail:predict-importance');
 		$this->setDescription('Predict importance of an incoming message');
 		$this->addArgument(self::ARGUMENT_ACCOUNT_ID, InputArgument::REQUIRED);
 		$this->addArgument(self::ARGUMENT_SENDER, InputArgument::REQUIRED);
+		$this->addArgument(self::ARGUMENT_SUBJECT, InputArgument::OPTIONAL);
 	}
 
-	public function isEnabled() {
+	public function isEnabled(): bool {
 		return $this->config->getSystemValueBool('debug');
 	}
 
-	/**
-	 * @return int
-	 */
 	protected function execute(InputInterface $input, OutputInterface $output): int {
 		$accountId = (int)$input->getArgument(self::ARGUMENT_ACCOUNT_ID);
 		$sender = $input->getArgument(self::ARGUMENT_SENDER);
+		$subject = $input->getArgument(self::ARGUMENT_SUBJECT) ?? '';
+
+		$consoleLogger = new ConsoleLoggerDecorator(
+			$this->logger,
+			$output
+		);
 
 		try {
 			$account = $this->accountService->findById($accountId);
@@ -73,9 +76,11 @@ protected function execute(InputInterface $input, OutputInterface $output): int
 		$fakeMessage = new Message();
 		$fakeMessage->setUid(0);
 		$fakeMessage->setFrom(AddressList::parse("Name <$sender>"));
+		$fakeMessage->setSubject($subject);
 		[$prediction] = $this->classifier->classifyImportance(
 			$account,
-			[$fakeMessage]
+			[$fakeMessage],
+			$consoleLogger
 		);
 		if ($prediction) {
 			$output->writeln('Message is important');

lib/Command/PreprocessAccount.php

@@ -0,0 +1,65 @@
+<?php
+
+declare(strict_types=1);
+
+/**
+ * SPDX-FileCopyrightText: 2024 Nextcloud GmbH and Nextcloud contributors
+ * SPDX-License-Identifier: AGPL-3.0-or-later
+ */
+
+namespace OCA\Mail\Command;
+
+use OCA\Mail\Service\AccountService;
+use OCA\Mail\Service\PreprocessingService;
+use OCP\AppFramework\Db\DoesNotExistException;
+use Psr\Log\LoggerInterface;
+use Symfony\Component\Console\Command\Command;
+use Symfony\Component\Console\Input\InputArgument;
+use Symfony\Component\Console\Input\InputInterface;
+use Symfony\Component\Console\Output\OutputInterface;
+use function memory_get_peak_usage;
+
+class PreprocessAccount extends Command {
+	public const ARGUMENT_ACCOUNT_ID = 'account-id';
+
+	private AccountService $accountService;
+	private PreprocessingService $preprocessingService;
+	private LoggerInterface $logger;
+
+	public function __construct(AccountService $service,
+		PreprocessingService $preprocessingService,
+		LoggerInterface $logger) {
+		parent::__construct();
+
+		$this->accountService = $service;
+		$this->preprocessingService = $preprocessingService;
+		$this->logger = $logger;
+	}
+
+	/**
+	 * @return void
+	 */
+	protected function configure() {
+		$this->setName('mail:account:preprocess');
+		$this->setDescription('Preprocess all mailboxes of an IMAP account');
+		$this->addArgument(self::ARGUMENT_ACCOUNT_ID, InputArgument::REQUIRED);
+	}
+
+	protected function execute(InputInterface $input, OutputInterface $output): int {
+		$accountId = (int)$input->getArgument(self::ARGUMENT_ACCOUNT_ID);
+
+		try {
+			$account = $this->accountService->findById($accountId);
+		} catch (DoesNotExistException $e) {
+			$output->writeln("<error>Account $accountId does not exist</error>");
+			return 1;
+		}
+
+		$this->preprocessingService->process(4294967296, $account);
+
+		$mbs = (int)(memory_get_peak_usage() / 1024 / 1024);
+		$output->writeln('<info>' . $mbs . 'MB of memory used</info>');
+
+		return 0;
+	}
+}

lib/Command/RunMetaEstimator.php

@@ -0,0 +1,116 @@
+<?php
+
+declare(strict_types=1);
+
+/**
+ * SPDX-FileCopyrightText: 2024 Nextcloud GmbH and Nextcloud contributors
+ * SPDX-License-Identifier: AGPL-3.0-or-later
+ */
+
+namespace OCA\Mail\Command;
+
+use OCA\Mail\Service\AccountService;
+use OCA\Mail\Service\Classification\ImportanceClassifier;
+use OCA\Mail\Support\ConsoleLoggerDecorator;
+use OCP\AppFramework\Db\DoesNotExistException;
+use OCP\IConfig;
+use Psr\Log\LoggerInterface;
+use Rubix\ML\Backends\Amp;
+use Rubix\ML\Classifiers\KNearestNeighbors;
+use Rubix\ML\CrossValidation\KFold;
+use Rubix\ML\CrossValidation\Metrics\FBeta;
+use Rubix\ML\GridSearch;
+use Rubix\ML\Kernels\Distance\Euclidean;
+use Rubix\ML\Kernels\Distance\Jaccard;
+use Rubix\ML\Kernels\Distance\Manhattan;
+use Symfony\Component\Console\Command\Command;
+use Symfony\Component\Console\Input\InputArgument;
+use Symfony\Component\Console\Input\InputInterface;
+use Symfony\Component\Console\Output\OutputInterface;
+
+class RunMetaEstimator extends Command {
+	public const ARGUMENT_ACCOUNT_ID = 'account-id';
+	public const ARGUMENT_SHUFFLE = 'shuffle';
+
+	private AccountService $accountService;
+	private LoggerInterface $logger;
+	private ImportanceClassifier $classifier;
+	private IConfig $config;
+
+	public function __construct(
+		AccountService $accountService,
+		LoggerInterface $logger,
+		ImportanceClassifier $classifier,
+		IConfig $config,
+	) {
+		parent::__construct();
+
+		$this->accountService = $accountService;
+		$this->logger = $logger;
+		$this->classifier = $classifier;
+		$this->config = $config;
+	}
+
+	protected function configure(): void {
+		$this->setName('mail:account:run-meta-estimator');
+		$this->setDescription('Run the meta estimator for an account');
+		$this->addArgument(self::ARGUMENT_ACCOUNT_ID, InputArgument::REQUIRED);
+		$this->addOption(self::ARGUMENT_SHUFFLE, null, null, 'Shuffle data set before training');
+	}
+
+	public function isEnabled(): bool {
+		return $this->config->getSystemValueBool('debug');
+	}
+
+	protected function execute(InputInterface $input, OutputInterface $output): int {
+		$accountId = (int)$input->getArgument(self::ARGUMENT_ACCOUNT_ID);
+		$shuffle = (bool)$input->getOption(self::ARGUMENT_SHUFFLE);
+
+		try {
+			$account = $this->accountService->findById($accountId);
+		} catch (DoesNotExistException $e) {
+			$output->writeln("<error>Account $accountId does not exist</error>");
+			return 1;
+		}
+
+		$consoleLogger = new ConsoleLoggerDecorator(
+			$this->logger,
+			$output
+		);
+
+		$estimator = static function () use ($consoleLogger) {
+			$params = [
+				[5, 10, 15, 20, 25, 30, 35, 40], // Neighbors
+				[true, false], // Weighted?
+				[new Euclidean(), new Manhattan(), new Jaccard()], // Kernel
+			];
+
+			$estimator = new GridSearch(
+				KNearestNeighbors::class,
+				$params,
+				new FBeta(),
+				new KFold(5)
+			);
+			$estimator->setLogger($consoleLogger);
+			$estimator->setBackend(new Amp());
+			return $estimator;
+		};
+
+		/** @var GridSearch $metaEstimator */
+		$metaEstimator = $this->classifier->train(
+			$account,
+			$consoleLogger,
+			$estimator,
+			$shuffle,
+			false,
+		);
+
+		if ($metaEstimator !== null) {
+			$output->writeln("<info>Best estimator: {$metaEstimator->base()}</info>");
+		}
+
+		$mbs = (int)(memory_get_peak_usage() / 1024 / 1024);
+		$output->writeln('<info>' . $mbs . 'MB of memory used</info>');
+		return 0;
+	}
+}

lib/Command/TrainAccount.php

@@ -3,7 +3,7 @@
 declare(strict_types=1);
 
 /**
- * SPDX-FileCopyrightText: 2019 Nextcloud GmbH and Nextcloud contributors
+ * SPDX-FileCopyrightText: 2019-2024 Nextcloud GmbH and Nextcloud contributors
  * SPDX-License-Identifier: AGPL-3.0-or-later
  */
 
@@ -23,6 +23,9 @@
 
 class TrainAccount extends Command {
 	public const ARGUMENT_ACCOUNT_ID = 'account-id';
+	public const ARGUMENT_SHUFFLE = 'shuffle';
+	public const ARGUMENT_DRY_RUN = 'dry-run';
+	public const ARGUMENT_FORCE = 'force';
 
 	private AccountService $accountService;
 	private ImportanceClassifier $classifier;
@@ -41,28 +44,39 @@ public function __construct(AccountService $service,
 		$this->classificationSettingsService = $classificationSettingsService;
 	}
 
-	/**
-	 * @return void
-	 */
-	protected function configure() {
+	protected function configure(): void {
 		$this->setName('mail:account:train');
 		$this->setDescription('Train the classifier of new messages');
 		$this->addArgument(self::ARGUMENT_ACCOUNT_ID, InputArgument::REQUIRED);
+		$this->addOption(self::ARGUMENT_SHUFFLE, null, null, 'Shuffle data set before training');
+		$this->addOption(
+			self::ARGUMENT_DRY_RUN,
+			null,
+			null,
+			'Don\'t persist classifier after training'
+		);
+		$this->addOption(
+			self::ARGUMENT_FORCE,
+			null,
+			null,
+			'Train an estimator even if the classification is disabled by the user'
+		);
 	}
 
-	/**
-	 * @return int
-	 */
 	protected function execute(InputInterface $input, OutputInterface $output): int {
 		$accountId = (int)$input->getArgument(self::ARGUMENT_ACCOUNT_ID);
+		$shuffle = (bool)$input->getOption(self::ARGUMENT_SHUFFLE);
+		$dryRun = (bool)$input->getOption(self::ARGUMENT_DRY_RUN);
+		$force = (bool)$input->getOption(self::ARGUMENT_FORCE);
 
 		try {
 			$account = $this->accountService->findById($accountId);
 		} catch (DoesNotExistException $e) {
 			$output->writeln("<error>account $accountId does not exist</error>");
 			return 1;
 		}
-		if (!$this->classificationSettingsService->isClassificationEnabled($account->getUserId())) {
+
+		if (!$force && !$this->classificationSettingsService->isClassificationEnabled($account->getUserId())) {
 			$output->writeln("<info>classification is turned off for account $accountId</info>");
 			return 2;
 		}
@@ -71,9 +85,13 @@ protected function execute(InputInterface $input, OutputInterface $output): int
 			$this->logger,
 			$output
 		);
+
 		$this->classifier->train(
 			$account,
-			$consoleLogger
+			$consoleLogger,
+			null,
+			$shuffle,
+			!$dryRun
 		);
 
 		$mbs = (int)(memory_get_peak_usage() / 1024 / 1024);