zm/tougao
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

ai写作辅助检测

Browse Source
This commit is contained in:
wyn
2026-07-15 10:55:08 +08:00
parent c80ff9b99f
commit a46df66028
4 changed files with 695 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

168
application/common/AiTemplateSentenceRuleDetectService.php Normal file
View File

@@ -0,0 +1,168 @@
<?php
namespace app\common;
use think\Db;
/**
* 模板句规则检测（t_ai_template_sentence）
*/
class AiTemplateSentenceRuleDetectService
{
/** @var array<int,array>|null */
private static $ruleCache;
/**
* @param array<string,string> $sectionTexts 键：abstract/introduction/...
* @return array{hits:array,rule_score:int}
*/
public function detect(array $sectionTexts): array
{
$rules = $this->loadActiveRules();
$hits = [];
$ruleScore = 0;
$normalized = [];
foreach ($sectionTexts as $key => $text) {
$section = strtolower(trim((string) $key));
if ($section === '') {
continue;
}
$normalized[$section] = ManuscriptTextCleanService::clean((string) $text);
}
$allText = ManuscriptTextCleanService::clean(implode("\n\n", array_filter($normalized)));
foreach ($rules as $rule) {
$section = strtolower(trim((string) ($rule['section'] ?? '')));
$pattern = trim((string) ($rule['sentence_pattern'] ?? ''));
if ($pattern === '') {
continue;
}
$targetText = '';
if ($section === 'all') {
$targetText = $allText;
} elseif (isset($normalized[$section])) {
$targetText = $normalized[$section];
} else {
continue;
}
if ($targetText === '') {
continue;
}
$matches = $this->matchPattern($targetText, $pattern);
foreach ($matches as $matchedText) {
$sentence = $this->extractSentenceContaining($targetText, $matchedText);
$hitScore = max(1, intval($rule['weight'] ?? 1)) * max(1, intval($rule['risk_level'] ?? 1));
$hits[] = [
'section' => $section,
'type' => '模板化表达',
'sentence' => $sentence !== '' ? $sentence : $matchedText,
'pattern' => $pattern,
'score' => $hitScore,
'remark' => (string) ($rule['remark'] ?? ''),
];
$ruleScore += $hitScore;
}
}
return [
'hits' => $hits,
'rule_score' => min(100, $ruleScore),
];
}
/**
* @return array<int,array>
*/
private function loadActiveRules(): array
{
if (self::$ruleCache !== null) {
return self::$ruleCache;
}
try {
self::$ruleCache = Db::name('ai_template_sentence')
->where('status', 1)
->field('id,sentence_pattern,section,language,risk_level,weight,remark')
->order('section asc,id asc')
->select();
} catch (\Throwable $e) {
self::$ruleCache = [];
}
return self::$ruleCache;
}
/**
* @return array<int,string>
*/
private function matchPattern(string $text, string $pattern): array
{
// 支持 regex: 前缀；否则优先按正则，失败则按字面量匹配
$raw = $pattern;
$forceRegex = false;
if (stripos($pattern, 'regex:') === 0) {
$raw = trim(substr($pattern, 6));
$forceRegex = true;
}
$out = $this->runPregMatchAll($text, '#' . $raw . '#iu');
if (!empty($out) || $forceRegex) {
return $out;
}
// 字面量回退（库内普通句子）
return $this->runPregMatchAll($text, '#' . preg_quote($raw, '#') . '#iu');
}
/**
* @return array<int,string>
*/
private function runPregMatchAll(string $text, string $regex): array
{
$error = null;
set_error_handler(function ($errno, $errstr) use (&$error) {
$error = $errstr;
return true;
});
$ok = preg_match_all($regex, $text, $matches);
restore_error_handler();
if ($ok === false || $error !== null || empty($matches[0])) {
return [];
}
$out = [];
foreach ($matches[0] as $match) {
$match = trim((string) $match);
if ($match !== '') {
$out[] = $match;
}
}
return $out;
}
private function extractSentenceContaining(string $text, string $needle): string
{
$pos = mb_stripos($text, $needle);
if ($pos === false) {
return '';
}
$before = mb_substr($text, 0, $pos);
$after = mb_substr($text, $pos);
$start = $before;
if (preg_match('/.*[.!?。！？]\s*/us', $before, $m)) {
$start = substr($before, strlen($m[0]));
}
$sentence = $start . $after;
if (preg_match('/^(.+?[.!?。！？])/us', $sentence, $m)) {
return trim($m[1]);
}
if (preg_match('/^.{1,500}/us', $sentence, $m)) {
return trim($m[0]);
}
return trim(mb_substr($sentence, 0, 300));
}
}

176
application/common/AiTemplateSentenceService.php Normal file
View File

@@ -0,0 +1,176 @@
<?php
namespace app\common;
use think\Db;
/**
* AI 写作模版句规则匹配（t_ai_template_sentence）
*/
class AiTemplateSentenceService
{
/** @var array<int,array>|null */
private static $ruleCache;
/**
* 按稿件分节统计模版句出现次数
*
* @param array<string,string> $sectionTexts 键：abstract/introduction/methods/results/discussion/conclusion
* @return array{total_matches:int,matched_rule_count:int,items:array,by_section:array}
*/
public function countInManuscript(array $sectionTexts): array
{
$rules = $this->loadActiveRules();
if (empty($rules)) {
return [
'total_matches' => 0,
'matched_rule_count' => 0,
'items' => [],
'by_section' => [],
];
}
$normalizedSections = [];
foreach ($sectionTexts as $key => $text) {
$sectionKey = strtolower(trim((string) $key));
if ($sectionKey === '') {
continue;
}
$normalizedSections[$sectionKey] = $this->normalizeText((string) $text);
}
$allText = $this->normalizeText(implode("\n\n", array_filter($normalizedSections)));
$items = [];
$bySection = [];
foreach ($rules as $rule) {
$ruleSection = strtolower(trim((string) ($rule['section'] ?? '')));
$pattern = trim((string) ($rule['sentence_pattern'] ?? ''));
if ($pattern === '') {
continue;
}
$targetText = '';
if ($ruleSection === 'all') {
$targetText = $allText;
} elseif (isset($normalizedSections[$ruleSection])) {
$targetText = $normalizedSections[$ruleSection];
} else {
continue;
}
if ($targetText === '') {
continue;
}
$count = $this->countPatternMatches($targetText, $pattern);
if ($count <= 0) {
continue;
}
$item = [
'id' => intval($rule['id'] ?? 0),
'section' => $ruleSection,
'sentence_pattern' => $pattern,
'count' => $count,
'risk_level' => intval($rule['risk_level'] ?? 0),
'weight' => intval($rule['weight'] ?? 0),
'remark' => (string) ($rule['remark'] ?? ''),
];
$items[] = $item;
if (!isset($bySection[$ruleSection])) {
$bySection[$ruleSection] = [
'total_matches' => 0,
'matched_rule_count' => 0,
'items' => [],
];
}
$bySection[$ruleSection]['items'][] = $item;
$bySection[$ruleSection]['total_matches'] += $count;
$bySection[$ruleSection]['matched_rule_count']++;
}
usort($items, function ($a, $b) {
if ($a['count'] !== $b['count']) {
return $b['count'] - $a['count'];
}
if ($a['weight'] !== $b['weight']) {
return $b['weight'] - $a['weight'];
}
return $a['id'] - $b['id'];
});
foreach ($bySection as &$sectionStats) {
usort($sectionStats['items'], function ($a, $b) {
if ($a['count'] !== $b['count']) {
return $b['count'] - $a['count'];
}
return $b['weight'] - $a['weight'];
});
}
unset($sectionStats);
$totalMatches = 0;
foreach ($items as $item) {
$totalMatches += intval($item['count']);
}
return [
'total_matches' => $totalMatches,
'matched_rule_count' => count($items),
'items' => $items,
'by_section' => $bySection,
];
}
/**
* @return array<int,array>
*/
private function loadActiveRules(): array
{
if (self::$ruleCache !== null) {
return self::$ruleCache;
}
try {
self::$ruleCache = Db::name('ai_template_sentence')
->where('status', 1)
->field('id,sentence_pattern,section,language,risk_level,weight,remark,article_type,version')
->order('section asc,id asc')
->select();
} catch (\Throwable $e) {
self::$ruleCache = [];
}
return self::$ruleCache;
}
private function normalizeText(string $text): string
{
return ManuscriptTextCleanService::clean($text);
}
private function countPatternMatches(string $text, string $pattern): int
{
if ($text === '' || $pattern === '') {
return 0;
}
$regex = '#' . $pattern . '#iu';
$error = null;
set_error_handler(function ($errno, $errstr) use (&$error) {
$error = $errstr;
return true;
});
$count = preg_match_all($regex, $text, $matches);
restore_error_handler();
if ($count === false || $error !== null) {
return 0;
}
return intval($count);
}
}

230
application/common/AiWritingRiskAnalysisService.php Normal file
View File

@@ -0,0 +1,230 @@
<?php
namespace app\common;
/**
* 编辑辅助：AI 辅助写作风险综合分析
*/
class AiWritingRiskAnalysisService
{
/**
* @param array $parsedData parseManuscriptStructure 的 data 字段
*/
public function buildReport(array $parsedData): array
{
$templateStats = (array) ($parsedData['template_sentence_stats'] ?? []);
$repeatedStats = (array) ($parsedData['repeated_phrases'] ?? []);
$sentenceStats = (array) ($parsedData['sentence_stats'] ?? []);
$dimensions = [
$this->analyzeTemplateSentenceRisk($templateStats),
$this->analyzeRepeatedPhraseRisk($repeatedStats),
$this->analyzeSentenceLengthRisk($sentenceStats),
];
$overallScore = 0;
$levelRank = ['low' => 1, 'medium' => 2, 'high' => 3];
$maxRank = 1;
$flags = [];
foreach ($dimensions as $dimension) {
$overallScore += intval($dimension['score'] ?? 0);
$rank = $levelRank[$dimension['level'] ?? 'low'] ?? 1;
if ($rank > $maxRank) {
$maxRank = $rank;
}
foreach ((array) ($dimension['flags'] ?? []) as $flag) {
if ($flag !== '') {
$flags[] = $flag;
}
}
}
$overallLevel = array_search($maxRank, $levelRank, true) ?: 'low';
return [
'overall_level' => $overallLevel,
'overall_score' => min(100, $overallScore),
'summary' => $this->buildSummary($overallLevel, $dimensions, $flags),
'dimensions' => $dimensions,
'flags' => array_values(array_unique($flags)),
];
}
private function analyzeTemplateSentenceRisk(array $stats): array
{
$items = (array) ($stats['items'] ?? []);
$totalMatches = intval($stats['total_matches'] ?? 0);
$matchedRuleCount = intval($stats['matched_rule_count'] ?? 0);
$score = 0;
$flags = [];
foreach ($items as $item) {
$count = intval($item['count'] ?? 0);
$riskLevel = max(1, intval($item['risk_level'] ?? 1));
$weight = max(1, intval($item['weight'] ?? 1));
$score += $count * $riskLevel * $weight;
if ($count > 0) {
$section = (string) ($item['section'] ?? '');
$remark = trim((string) ($item['remark'] ?? ''));
$label = $remark !== '' ? $remark : (string) ($item['sentence_pattern'] ?? '');
$flags[] = sprintf('%s段命中模版句：%s（%d次）', $section, $label, $count);
}
}
$level = $this->scoreToLevel($score, 15, 40);
return [
'key' => 'template_sentence',
'name' => '模版句特征',
'level' => $level,
'score' => min(100, $score),
'summary' => $totalMatches > 0
? sprintf('命中 %d 条模版句规则，共 %d 处', $matchedRuleCount, $totalMatches)
: '未检测到典型 AI 模版句',
'metrics' => [
'total_matches' => $totalMatches,
'matched_rule_count' => $matchedRuleCount,
],
'details' => $stats,
'flags' => array_slice($flags, 0, 8),
];
}
private function analyzeRepeatedPhraseRisk(array $stats): array
{
$items = (array) ($stats['suspicious_items'] ?? []);
if (empty($items)) {
$items = array_values(array_filter((array) ($stats['items'] ?? []), function ($item) {
return ($item['category'] ?? '') === 'suspicious';
}));
}
$uniqueCount = count($items);
$expectedCount = count((array) ($stats['expected_topic_items'] ?? []));
$topCount = 0;
$flags = [];
foreach (array_slice($items, 0, 5) as $item) {
$count = intval($item['count'] ?? 0);
if ($count > $topCount) {
$topCount = $count;
}
if ($count >= 4) {
$flags[] = sprintf('可疑重复短语「%s」出现 %d 次', (string) ($item['phrase'] ?? ''), $count);
}
}
$score = min(100, $uniqueCount * 3 + max(0, $topCount - 3) * 6);
$level = 'low';
if ($uniqueCount >= 10 || $topCount >= 8) {
$level = 'high';
} elseif ($uniqueCount >= 4 || $topCount >= 5) {
$level = 'medium';
}
$summary = '未发现可疑机械重复短语';
if ($uniqueCount > 0) {
$summary = sprintf('发现 %d 组可疑重复短语，最高重复 %d 次', $uniqueCount, $topCount);
} elseif ($expectedCount > 0) {
$summary = sprintf('主题词正常重复 %d 组，未触发可疑短语', $expectedCount);
}
return [
'key' => 'repeated_phrase',
'name' => '重复短语',
'level' => $level,
'score' => $score,
'summary' => $summary,
'metrics' => [
'suspicious_count' => $uniqueCount,
'expected_topic_count' => $expectedCount,
'top_count' => $topCount,
],
'details' => $stats,
'flags' => $flags,
];
}
private function analyzeSentenceLengthRisk(array $stats): array
{
$avg = floatval($stats['avg'] ?? 0);
$std = floatval($stats['std'] ?? 0);
$max = intval($stats['max'] ?? 0);
$sentenceCount = intval($stats['sentence_count'] ?? 0);
$score = 0;
$flags = [];
$level = 'low';
if ($sentenceCount >= 8) {
if ($avg >= 32) {
$score += 20;
$flags[] = sprintf('平均句长偏高（%.1f 词/句）', $avg);
}
if ($std > 0 && $std <= 5 && $avg >= 12 && $avg <= 28) {
$score += 25;
$flags[] = sprintf('句长分布过于均匀（均值 %.1f，标准差 %.1f）', $avg, $std);
}
if ($max >= 50) {
$score += 10;
$flags[] = sprintf('存在超长句（最长 %d 词）', $max);
}
}
$level = $this->scoreToLevel($score, 10, 25);
return [
'key' => 'sentence_length',
'name' => '句长特征',
'level' => $level,
'score' => min(100, $score),
'summary' => $sentenceCount > 0
? sprintf('平均句长 %.1f，标准差 %.1f，共 %d 句', $avg, $std, $sentenceCount)
: '正文句长数据不足',
'metrics' => [
'avg' => $avg,
'std' => $std,
'max' => $max,
'min' => intval($stats['min'] ?? 0),
'sentence_count' => $sentenceCount,
],
'details' => $stats,
'flags' => $flags,
];
}
private function scoreToLevel(int $score, int $mediumThreshold, int $highThreshold): string
{
if ($score >= $highThreshold) {
return 'high';
}
if ($score >= $mediumThreshold) {
return 'medium';
}
return 'low';
}
private function buildSummary(string $overallLevel, array $dimensions, array $flags): string
{
$levelText = [
'low' => '低风险',
'medium' => '中等风险',
'high' => '较高风险',
];
$parts = [$levelText[$overallLevel] ?? '低风险'];
foreach ($dimensions as $dimension) {
if (($dimension['level'] ?? 'low') === 'low') {
continue;
}
$parts[] = $dimension['name'] . '需关注';
}
if (!empty($flags)) {
$parts[] = '重点：' . $flags[0];
}
return implode('；', $parts);
}
}

121
application/common/ManuscriptTextCleanService.php Normal file
View File

@@ -0,0 +1,121 @@
<?php
namespace app\common;
/**
* 稿件文本简单清洗（检测前预处理，不改变接口返回的原始正文）
*/
class ManuscriptTextCleanService
{
/**
* 检测前文本清洗：解码、去噪、去引用标记、统一空白
*/
public static function clean(string $text): string
{
$text = trim($text);
if ($text === '') {
return '';
}
$text = html_entity_decode($text, ENT_QUOTES | ENT_HTML5, 'UTF-8');
$text = self::normalizeEncoding($text);
$text = self::normalizeQuotesAndDashes($text);
$text = self::stripInvisibleChars($text);
$text = self::stripUrlsAndEmails($text);
$text = self::stripCitationMarkers($text);
$text = self::stripFigureTableTags($text);
$text = self::stripFootnoteMarkers($text);
$text = self::stripSearchStrategyNoise($text);
$text = preg_replace('/\s+/u', ' ', $text);
$text = trim((string) preg_replace('/\s+([,.;:!?])/u', '$1', $text));
return trim($text);
}
/**
* @param array<string,string> $sections
* @return array<string,string>
*/
public static function cleanSections(array $sections): array
{
$cleaned = [];
foreach ($sections as $key => $text) {
$cleaned[$key] = self::clean((string) $text);
}
return $cleaned;
}
private static function normalizeEncoding(string $text): string
{
if (!mb_check_encoding($text, 'UTF-8')) {
$text = mb_convert_encoding($text, 'UTF-8', 'GBK,GB2312,GB18030,ISO-8859-1');
}
return $text;
}
private static function normalizeQuotesAndDashes(string $text): string
{
return strtr($text, [
'‘' => "'",
'’' => "'",
'“' => '"',
'”' => '"',
'„' => '"',
'‟' => '"',
'—' => '-',
'–' => '-',
'‑' => '-',
'　' => ' ',
]);
}
private static function stripInvisibleChars(string $text): string
{
$text = preg_replace('/[\x00-\x08\x0B\x0C\x0E-\x1F\x7F]/u', '', $text);
$text = preg_replace('/[\x{200B}-\x{200F}\x{202A}-\x{202E}\x{2060}\x{FEFF}]/u', '', $text);
$text = str_replace([chr(0xC2) . chr(0xA0), '&nbsp;'], ' ', $text);
return $text;
}
private static function stripUrlsAndEmails(string $text): string
{
$text = preg_replace('/https?:\/\/\S+/iu', ' ', $text);
$text = preg_replace('/\b[\w.+-]+@[\w.-]+\.[a-z]{2,}\b/iu', ' ', $text);
return $text;
}
private static function stripCitationMarkers(string $text): string
{
// [1] [1,2] [1-3] [1, 2-4]
$text = preg_replace('/\[\s*\d+(?:\s*[,，\-–—]\s*\d+)*\s*\]/u', ' ', $text);
// 上标残留数字（如 word 提取后的 ^1^）
$text = preg_replace('/\b\^\d+\^/u', ' ', $text);
return $text;
}
private static function stripFigureTableTags(string $text): string
{
$text = preg_replace('/\((?:Fig(?:ure)?|Table|Tab\.)\s*[A-Za-z0-9.-]+\)/iu', ' ', $text);
$text = preg_replace('/\b(?:Fig(?:ure)?|Table|Tab\.)\s*[A-Za-z0-9.-]+\b/iu', ' ', $text);
return $text;
}
private static function stripFootnoteMarkers(string $text): string
{
$text = preg_replace('/[*#†‡§]+\d*/u', ' ', $text);
return $text;
}
/**
* 去除文献检索式、数据库字段等 Methods 常见残留
*/
private static function stripSearchStrategyNoise(string $text): string
{
$text = preg_replace('/\btitle\s*[\/\-]\s*abstract\b/iu', ' ', $text);
$text = preg_replace('/\babstract\s*[\/\-]\s*title\b/iu', ' ', $text);
$text = preg_replace('/\bti\s*[\-\/]?\s*ab\b/iu', ' ', $text);
$text = preg_replace('/\bmesh\s+terms?\b/iu', ' ', $text);
$text = preg_replace('/\b(search\s+terms?|search\s+strategy)\b/iu', ' ', $text);
return $text;
}
}