1585 lines
55 KiB
PHP
1585 lines
55 KiB
PHP
<?php
|
||
|
||
namespace app\common;
|
||
|
||
use think\Db;
|
||
use think\Env;
|
||
use think\Queue;
|
||
|
||
/**
|
||
* 正文 <blue>[n]</blue> 引用与 t_production_article_refer（index+1=n）相关性校对。
|
||
* LLM 配置与 PromotionLlmService 相同；单条任务走 ReferenceCheck 队列。
|
||
*/
|
||
class ReferenceCheckService
|
||
{
|
||
const QUEUE_NAME = 'ReferenceCheck';
|
||
|
||
/** t_article_main.ref_check_status */
|
||
const AM_STATUS_NONE = 0;
|
||
const AM_STATUS_PASS = 1;
|
||
const AM_STATUS_FAIL = 2;
|
||
const AM_STATUS_RUNNING = 3;
|
||
|
||
/**
|
||
* 兼容无 ?? 的 PHP 版本
|
||
*/
|
||
private function arrGet($arr, $key, $default = '')
|
||
{
|
||
return isset($arr[$key]) ? $arr[$key] : $default;
|
||
}
|
||
|
||
/**
|
||
* 单条入队（可手工指定正文与文献文本）
|
||
*/
|
||
public function enqueue($contentA, $contentB, array $extra = [])
|
||
{
|
||
$contentA = trim($contentA);
|
||
if ($contentA === '') {
|
||
throw new \InvalidArgumentException('content_a is required');
|
||
}
|
||
|
||
$now = date('Y-m-d H:i:s');
|
||
$checkId = Db::name('article_reference_check_result')->insertGetId([
|
||
'article_id' => intval($this->arrGet($extra, 'article_id', 0)),
|
||
'am_id' => intval($this->arrGet($extra, 'am_id', 0)),
|
||
'p_article_id' => intval($this->arrGet($extra, 'p_article_id', 0)),
|
||
'p_refer_id' => intval($this->arrGet($extra, 'p_refer_id', 0)),
|
||
'refer_index' => intval($this->arrGet($extra, 'refer_index', 0)),
|
||
'reference_no' => intval($this->arrGet($extra, 'reference_no', 0)),
|
||
'reference_raw' => (string)$this->arrGet($extra, 'reference_raw', ''),
|
||
'cite_tag_start' => intval($this->arrGet($extra, 'cite_tag_start', 0)),
|
||
'cite_tag_end' => intval($this->arrGet($extra, 'cite_tag_end', 0)),
|
||
'text_start' => intval($this->arrGet($extra, 'text_start', 0)),
|
||
'text_end' => intval($this->arrGet($extra, 'text_end', 0)),
|
||
'content_a' => $contentA,
|
||
'content_b' => trim($contentB),
|
||
'status' => 0,
|
||
'created_at' => $now,
|
||
'updated_at' => $now,
|
||
]);
|
||
|
||
$amId = intval($this->arrGet($extra, 'am_id', 0));
|
||
if ($amId > 0) {
|
||
$this->setAmRefCheckStatus($amId, self::AM_STATUS_RUNNING);
|
||
}
|
||
|
||
$this->pushJob(intval($checkId), intval($this->arrGet($extra, 'queue_delay', 0)));
|
||
|
||
return ['check_id' => $checkId, 'queued' => 1];
|
||
}
|
||
public function enqueueByArticleMain($main){
|
||
$amId = $main['am_id'];
|
||
// $main = Db::name('article_main')
|
||
// ->field('am_id,content,article_id')
|
||
// ->where('am_id', $amId)
|
||
// ->whereIn('state', [0, 2])
|
||
// ->find();
|
||
$citations = $this->extractReferences((string)$main['content']);
|
||
// return $citations;
|
||
if (empty($citations)) {
|
||
$this->setAmRefCheckStatus($amId, self::AM_STATUS_NONE);
|
||
return;
|
||
}
|
||
$prod = Db::name('production_article')
|
||
->where('article_id', $main['article_id'])
|
||
->where('state', 0)
|
||
->find();
|
||
if (empty($prod)) {
|
||
throw new \RuntimeException('production_article not found for article_id=' . $main['article_id']);
|
||
}
|
||
|
||
$pArticleId = intval($prod['p_article_id']);
|
||
$referMap = $this->loadReferMapByPArticleId($pArticleId);
|
||
|
||
if (empty($citations)) {
|
||
$this->setAmRefCheckStatus($amId, self::AM_STATUS_PASS);
|
||
return;
|
||
}
|
||
|
||
$skipped = 0;
|
||
$delay = 0;
|
||
foreach ($citations as $cite) {
|
||
foreach ($cite['reference_numbers'] as $refNo) {
|
||
$referIndex = $refNo - 1;
|
||
if ($referIndex < 0 || !isset($referMap[$referIndex])) {
|
||
$skipped++;
|
||
continue;
|
||
}
|
||
$refer = $referMap[$referIndex];
|
||
$referText = $this->formatReferForLlm($refer);
|
||
|
||
$now = date('Y-m-d H:i:s');
|
||
// [70-73] 展开为 reference_no=70,71,72,73 共 4 条记录
|
||
$checkId = Db::name('article_reference_check_result')->insertGetId([
|
||
'article_id' => $main['article_id'],
|
||
'p_article_id' => $pArticleId,
|
||
'am_id' => intval($main['am_id']),
|
||
'reference_no' => $refNo,
|
||
'refer_index' => $refNo,
|
||
'origin_text' => $cite['original_text'],
|
||
'refer_text' => $referText,
|
||
'p_refer_id' => $referMap[$referIndex]['p_refer_id'],
|
||
'text_start' => $cite['text_start'],
|
||
'text_end' => $cite['text_end'],
|
||
'created_at' => $now,
|
||
'updated_at' => $now,
|
||
]);
|
||
$this->pushJob(intval($checkId), $delay);
|
||
$checkIds[] = $checkId;
|
||
$delay += 1;
|
||
}
|
||
}
|
||
|
||
$this->setAmRefCheckStatus($amId, self::AM_STATUS_RUNNING);
|
||
}
|
||
/**
|
||
* 手工触发：对已完成且 confidence<=0.65 的记录入队 DOI 第二轮复核
|
||
*/
|
||
public function enqueueSecondPassByArticle($articleId)
|
||
{
|
||
$articleId = intval($articleId);
|
||
if ($articleId <= 0) {
|
||
throw new \InvalidArgumentException('article_id is required');
|
||
}
|
||
|
||
$rows = Db::name('article_reference_check_result')
|
||
->where('article_id', $articleId)
|
||
->where('status', 1)
|
||
->where('confidence', '<=', 0.65)
|
||
->orderRaw('rand()')
|
||
->limit(2)
|
||
->select();
|
||
|
||
$checkIds2 = [];
|
||
$delay2 = 0;
|
||
foreach ($rows as $checkLog) {
|
||
$rowId = $this->resolveCheckRowId($checkLog);
|
||
if ($this->maybeEnqueueSecondPass($rowId, floatval($checkLog['confidence']))) {
|
||
$checkIds2[] = $rowId;
|
||
$delay2 += 1;
|
||
}
|
||
}
|
||
|
||
return [
|
||
'article_id' => $articleId,
|
||
'check_ids2' => $checkIds2,
|
||
'queued' => count($checkIds2),
|
||
];
|
||
}
|
||
public function enqueueByArticle($articleId){
|
||
if ($articleId <= 0) {
|
||
throw new \InvalidArgumentException('article_id is required');
|
||
}
|
||
$prod = Db::name('production_article')
|
||
->where('article_id', $articleId)
|
||
->where('state', [0, 2])
|
||
->find();
|
||
if (empty($prod)) {
|
||
throw new \RuntimeException('production_article not found for article_id=' . $articleId);
|
||
}
|
||
$pArticleId = intval($prod['p_article_id']);
|
||
$referMap = $this->loadReferMapByPArticleId($pArticleId);
|
||
|
||
$mains = Db::name('article_main')
|
||
->field('am_id,content,article_id')
|
||
->where('article_id', $articleId)
|
||
->whereIn('state', [0, 2])
|
||
->order('sort asc')
|
||
->select();
|
||
if (empty($mains)) {
|
||
throw new \RuntimeException('article_main is empty');
|
||
}
|
||
$queued = 0;
|
||
$skipped = 0;
|
||
$checkIds = [];
|
||
$delay = 0;
|
||
$amIdsWithJobs = [];
|
||
|
||
foreach ($mains as $main) {
|
||
$amId = intval($main['am_id']);
|
||
$citations = $this->extractReferences((string)$main['content']);
|
||
if (empty($citations)) {
|
||
$this->setAmRefCheckStatus($amId, self::AM_STATUS_NONE);
|
||
continue;
|
||
}
|
||
foreach ($citations as $cite) {
|
||
foreach ($cite['reference_numbers'] as $refNo) {
|
||
$referIndex = $refNo - 1;
|
||
if ($referIndex < 0 || !isset($referMap[$referIndex])) {
|
||
$skipped++;
|
||
continue;
|
||
}
|
||
$refer = $referMap[$referIndex];
|
||
$referText = $this->formatReferForLlm($refer);
|
||
|
||
$now = date('Y-m-d H:i:s');
|
||
// [70-73] 展开为 reference_no=70,71,72,73 共 4 条记录
|
||
$checkId = Db::name('article_reference_check_result')->insertGetId([
|
||
'article_id' => $main['article_id'],
|
||
'p_article_id' => $pArticleId,
|
||
'am_id' => intval($main['am_id']),
|
||
'reference_no' => $refNo,
|
||
'refer_index' => $refNo,
|
||
'origin_text' => $cite['original_text'],
|
||
'refer_text' => $referText,
|
||
'p_refer_id' => $referMap[$referIndex]['p_refer_id'],
|
||
'text_start' => $cite['text_start'],
|
||
'text_end' => $cite['text_end'],
|
||
'created_at' => $now,
|
||
'updated_at' => $now,
|
||
]);
|
||
|
||
$this->pushJob(intval($checkId), $delay);
|
||
$checkIds[] = $checkId;
|
||
$queued++;
|
||
$delay += 1;
|
||
$amIdsWithJobs[$amId] = true;
|
||
}
|
||
}
|
||
}
|
||
foreach (array_keys($amIdsWithJobs) as $amId) {
|
||
$this->setAmRefCheckStatus($amId, self::AM_STATUS_RUNNING);
|
||
}
|
||
|
||
return [
|
||
'article_id' => $articleId,
|
||
'p_article_id' => $pArticleId,
|
||
'queued' => $queued,
|
||
'skipped' => $skipped,
|
||
'check_ids' => $checkIds,
|
||
'queue' => self::QUEUE_NAME,
|
||
];
|
||
}
|
||
|
||
/**
|
||
* 根据该节全部明细行汇总更新 t_article_main.ref_check_status
|
||
*/
|
||
public function syncAmRefCheckStatus($amId)
|
||
{
|
||
if ($amId <= 0) {
|
||
return self::AM_STATUS_NONE;
|
||
}
|
||
|
||
$rows = Db::name('article_reference_check_result')->where('am_id', $amId)->select();
|
||
if (empty($rows)) {
|
||
$this->setAmRefCheckStatus($amId, self::AM_STATUS_NONE);
|
||
return self::AM_STATUS_NONE;
|
||
}
|
||
|
||
$pending = 0;
|
||
$hasFail = false;
|
||
$done = 0;
|
||
|
||
foreach ($rows as $row) {
|
||
$st = intval($row['status']);
|
||
if ($st === 0) {
|
||
$pending++;
|
||
continue;
|
||
}
|
||
if ($st === 2 || ($st === 1 && intval($row['is_match']) === 0)) {
|
||
$hasFail = true;
|
||
}
|
||
if ($st === 1) {
|
||
$done++;
|
||
}
|
||
}
|
||
|
||
if ($pending > 0) {
|
||
$status = self::AM_STATUS_RUNNING;
|
||
} elseif ($hasFail) {
|
||
$status = self::AM_STATUS_FAIL;
|
||
} elseif ($done === count($rows)) {
|
||
$status = self::AM_STATUS_PASS;
|
||
} else {
|
||
$status = self::AM_STATUS_FAIL;
|
||
}
|
||
|
||
$this->setAmRefCheckStatus($amId, $status);
|
||
return $status;
|
||
}
|
||
|
||
public function setAmRefCheckStatus($amId, $status)
|
||
{
|
||
if ($amId <= 0) {
|
||
return;
|
||
}
|
||
Db::name('article_main')->where('am_id', $amId)->update([
|
||
'ref_check_status' => $status,
|
||
]);
|
||
}
|
||
|
||
public function clearArticleChecks($articleId)
|
||
{
|
||
Db::name('article_reference_check_result')->where('article_id', $articleId)->delete();
|
||
Db::name('article_main')
|
||
->where('article_id', $articleId)
|
||
->whereIn('state', [0, 2])
|
||
->update(['ref_check_status' => self::AM_STATUS_NONE]);
|
||
}
|
||
|
||
public static function amStatusLabel($status)
|
||
{
|
||
$map = [
|
||
self::AM_STATUS_NONE => 'none',
|
||
self::AM_STATUS_PASS => 'pass',
|
||
self::AM_STATUS_FAIL => 'fail',
|
||
self::AM_STATUS_RUNNING => 'running',
|
||
];
|
||
return isset($map[$status]) ? $map[$status] : 'unknown';
|
||
}
|
||
|
||
/**
|
||
* 表主键为 id（对外 API 参数名仍叫 check_id）
|
||
*/
|
||
public function resolveCheckRowId($row)
|
||
{
|
||
if (!is_array($row)) {
|
||
return 0;
|
||
}
|
||
if (isset($row['id']) && intval($row['id']) > 0) {
|
||
return intval($row['id']);
|
||
}
|
||
if (isset($row['check_id']) && intval($row['check_id']) > 0) {
|
||
return intval($row['check_id']);
|
||
}
|
||
return 0;
|
||
}
|
||
|
||
/**
|
||
* 解析 LLM 返回的 is_match（兼容 bool / 0|1 / "true"|"false" 字符串）
|
||
*/
|
||
public function parseLlmIsMatch($value)
|
||
{
|
||
if (is_bool($value)) {
|
||
return $value;
|
||
}
|
||
if (is_int($value) || is_float($value)) {
|
||
return intval($value) === 1;
|
||
}
|
||
$s = strtolower(trim((string)$value));
|
||
return in_array($s, ['1', 'true', 'yes', 'match', 'matched'], true);
|
||
}
|
||
|
||
/**
|
||
* 写入单条校对结果（统一截断 reason/error_msg，避免 varchar(512) 导致 UPDATE 失败）
|
||
*
|
||
* @throws \RuntimeException
|
||
*/
|
||
public function updateCheckResult($checkId, array $fields)
|
||
{
|
||
$checkId = intval($checkId);
|
||
if ($checkId <= 0) {
|
||
throw new \InvalidArgumentException('invalid check id');
|
||
}
|
||
|
||
if (isset($fields['reason'])) {
|
||
$fields['reason'] = mb_substr(trim((string)$fields['reason']), 0, 512);
|
||
}
|
||
if (isset($fields['error_msg'])) {
|
||
$fields['error_msg'] = mb_substr(trim((string)$fields['error_msg']), 0, 512);
|
||
}
|
||
$fields['updated_at'] = date('Y-m-d H:i:s');
|
||
|
||
$exists = Db::name('article_reference_check_result')->where('id', $checkId)->find();
|
||
if (empty($exists)) {
|
||
throw new \RuntimeException('article_reference_check_result not found, id=' . $checkId);
|
||
}
|
||
|
||
$affected = Db::name('article_reference_check_result')->where('id', $checkId)->update($fields);
|
||
if ($affected === false) {
|
||
throw new \RuntimeException('article_reference_check_result update failed, id=' . $checkId);
|
||
}
|
||
|
||
\think\Log::info('updateCheckResult id=' . $checkId . ' affected=' . intval($affected));
|
||
return intval($affected);
|
||
}
|
||
|
||
public function getResult($checkId)
|
||
{
|
||
if ($checkId <= 0) {
|
||
return null;
|
||
}
|
||
$row = Db::name('article_reference_check_result')->where('id', $checkId)->find();
|
||
return $row ?: null;
|
||
}
|
||
|
||
public function listByArticle($articleId, $status = -1, $onlyMismatch = false)
|
||
{
|
||
$q = Db::name('article_reference_check_result')->where('article_id', $articleId);
|
||
if ($status >= 0) {
|
||
$q->where('status', $status);
|
||
}
|
||
if ($onlyMismatch) {
|
||
$q->where('status', 1)->where('is_match', 0);
|
||
}
|
||
return $q->order('am_id asc, cite_tag_start asc, reference_no asc')->select();
|
||
}
|
||
|
||
/**
|
||
* 稿件预览：在 content 上标记不合理引用序号与引用句
|
||
*
|
||
* @return array{sections: array, issues: array, stats: array}
|
||
*/
|
||
public function buildArticlePreview($articleId, $amId = 0)
|
||
{
|
||
$q = Db::name('article_main')
|
||
->field('am_id,content,sort,ref_check_status')
|
||
->where('article_id', $articleId)
|
||
->whereIn('state', [0, 2]);
|
||
if ($amId > 0) {
|
||
$q->where('am_id', $amId);
|
||
}
|
||
$mains = $q->order('sort asc')->select();
|
||
|
||
$rows = $this->listByArticle($articleId, 1);
|
||
$badByAm = $this->indexBadResults($rows);
|
||
|
||
$sections = [];
|
||
$issues = [];
|
||
$stats = ['total' => 0, 'mismatch' => 0, 'match' => 0, 'pending' => 0];
|
||
|
||
foreach ($this->listByArticle($articleId, -1) as $r) {
|
||
$stats['total']++;
|
||
if (intval($r['status']) === 0) {
|
||
$stats['pending']++;
|
||
} elseif (intval($r['is_match']) === 1) {
|
||
$stats['match']++;
|
||
} else {
|
||
$stats['mismatch']++;
|
||
}
|
||
}
|
||
|
||
foreach ($mains as $main) {
|
||
$id = intval($main['am_id']);
|
||
$content = (string)$main['content'];
|
||
$badIndex = isset($badByAm[$id]) ? $badByAm[$id] : array();
|
||
$marked = $this->markContentForPreview($content, $id, $badIndex);
|
||
$amStatus = intval($this->arrGet($main, 'ref_check_status', 0));
|
||
$sections[] = [
|
||
'am_id' => $id,
|
||
'ref_check_status' => $amStatus,
|
||
'ref_check_pass' => $amStatus === self::AM_STATUS_PASS,
|
||
'ref_check_label' => self::amStatusLabel($amStatus),
|
||
'content' => $content,
|
||
'content_marked' => $marked['html'],
|
||
'issue_count' => $marked['issue_count'],
|
||
];
|
||
foreach ($marked['issues'] as $issue) {
|
||
$issues[] = $issue;
|
||
}
|
||
}
|
||
|
||
$articlePass = $this->resolveArticlePass($sections);
|
||
|
||
return [
|
||
'article_id' => $articleId,
|
||
'article_ref_check_pass' => $articlePass,
|
||
'sections' => $sections,
|
||
'issues' => $issues,
|
||
'stats' => $stats,
|
||
];
|
||
}
|
||
|
||
/**
|
||
* 全文是否通过：各节均为 pass，且无 running/fail（无引用节忽略）
|
||
*/
|
||
private function resolveArticlePass($sections)
|
||
{
|
||
$hasChecked = false;
|
||
foreach ($sections as $sec) {
|
||
$st = intval($this->arrGet($sec, 'ref_check_status', 0));
|
||
if ($st === self::AM_STATUS_NONE) {
|
||
continue;
|
||
}
|
||
$hasChecked = true;
|
||
if ($st !== self::AM_STATUS_PASS) {
|
||
return false;
|
||
}
|
||
}
|
||
return $hasChecked ? true : null;
|
||
}
|
||
|
||
/**
|
||
* @param array $rows status=1 的检测结果
|
||
* @return array<int, array> am_id => indexed bad map
|
||
*/
|
||
private function indexBadResults($rows)
|
||
{
|
||
$byAm = [];
|
||
foreach ($rows as $row) {
|
||
if (intval($row['status']) !== 1 || intval($row['is_match']) === 1) {
|
||
continue;
|
||
}
|
||
$amId = intval($row['am_id']);
|
||
$refNo = intval($row['reference_no']);
|
||
if ($amId <= 0 || $refNo <= 0) {
|
||
continue;
|
||
}
|
||
if (!isset($byAm[$amId])) {
|
||
$byAm[$amId] = ['by_raw' => [], 'contexts' => []];
|
||
}
|
||
$rawKey = $this->normalizeRefRawKey((string)$this->arrGet($row, 'reference_raw', ''));
|
||
if ($rawKey !== '') {
|
||
$byAm[$amId]['by_raw'][$rawKey][$refNo] = $row;
|
||
}
|
||
|
||
$ctxKey = intval($row['text_start']) . '_' . intval($row['text_end']);
|
||
if (!isset($byAm[$amId]['contexts'][$ctxKey])) {
|
||
$byAm[$amId]['contexts'][$ctxKey] = [
|
||
'text_start' => intval($row['text_start']),
|
||
'text_end' => intval($row['text_end']),
|
||
'check_ids' => [],
|
||
'reasons' => [],
|
||
'ref_nos' => [],
|
||
];
|
||
}
|
||
$byAm[$amId]['contexts'][$ctxKey]['check_ids'][] = $this->resolveCheckRowId($row);
|
||
$byAm[$amId]['contexts'][$ctxKey]['ref_nos'][] = $refNo;
|
||
$reason = trim((string)$this->arrGet($row, 'reason', ''));
|
||
if ($reason !== '') {
|
||
$byAm[$amId]['contexts'][$ctxKey]['reasons'][$refNo] = $reason;
|
||
}
|
||
}
|
||
return $byAm;
|
||
}
|
||
|
||
private function normalizeRefRawKey($raw)
|
||
{
|
||
$raw = str_replace(
|
||
['，', '–', '—', '−', '‐', '‑', ' '],
|
||
[',', '-', '-', '-', '-', '-', ''],
|
||
trim($raw)
|
||
);
|
||
return strtolower($raw);
|
||
}
|
||
|
||
/**
|
||
* @param array $badIndex indexBadResults 中单 am 的结构
|
||
*/
|
||
private function markContentForPreview($content, $amId, $badIndex)
|
||
{
|
||
$badByRaw = isset($badIndex['by_raw']) ? $badIndex['by_raw'] : array();
|
||
$contexts = isset($badIndex['contexts']) ? $badIndex['contexts'] : array();
|
||
$issues = array();
|
||
$issueCount = 0;
|
||
|
||
if ($content === '' || (empty($badByRaw) && empty($contexts))) {
|
||
return array('html' => $content, 'issues' => array(), 'issue_count' => 0);
|
||
}
|
||
|
||
$html = $content;
|
||
|
||
// 1) 先标记 blue 内各序号（在原文上操作，[70-73] 仅标不合理者如 70、71）
|
||
preg_match_all(
|
||
'/<blue>\[([\d,\-\s]+)\]<\/blue>/',
|
||
$html,
|
||
$matches,
|
||
PREG_OFFSET_CAPTURE
|
||
);
|
||
$citeDeltas = [];
|
||
if (!empty($matches[0])) {
|
||
$replacements = [];
|
||
foreach ($matches[0] as $idx => $match) {
|
||
$fullTag = $match[0];
|
||
$tagStart = $match[1];
|
||
$tagEnd = $tagStart + strlen($fullTag);
|
||
$inner = $matches[1][$idx][0];
|
||
$rawKey = $this->normalizeRefRawKey($inner);
|
||
$badNums = isset($badByRaw[$rawKey]) ? $badByRaw[$rawKey] : array();
|
||
|
||
$innerMarked = preg_replace_callback(
|
||
'/\d+/',
|
||
function ($numMatch) use ($badNums, &$issues, &$issueCount, $amId, $inner) {
|
||
$num = intval($numMatch[0]);
|
||
if (!isset($badNums[$num])) {
|
||
return $numMatch[0];
|
||
}
|
||
$row = $badNums[$num];
|
||
$rowReason = isset($row['reason']) ? $row['reason'] : '';
|
||
$issueCount++;
|
||
$issues[] = array(
|
||
'am_id' => $amId,
|
||
'check_id' => $this->resolveCheckRowId($row),
|
||
'reference_no' => $num,
|
||
'reference_raw' => $inner,
|
||
'reason' => $rowReason,
|
||
'confidence' => floatval(isset($row['confidence']) ? $row['confidence'] : 0),
|
||
);
|
||
$title = htmlspecialchars(
|
||
'引用[' . $num . ']不合理: ' . $rowReason,
|
||
ENT_QUOTES,
|
||
'UTF-8'
|
||
);
|
||
return '<span class="ref-no-error" data-check-id="' . $this->resolveCheckRowId($row)
|
||
. '" data-ref-no="' . $num . '" title="' . $title . '">'
|
||
. $numMatch[0] . '</span>';
|
||
},
|
||
$inner
|
||
);
|
||
|
||
$tagClass = !empty($badNums) ? ' ref-cite-error' : '';
|
||
$groupIds = !empty($badNums)
|
||
? implode(',', array_map('intval', array_column($badNums, 'check_id')))
|
||
: '';
|
||
$newHtml = '<blue class="ref-cite-tag' . $tagClass . '" data-ref-raw="' . htmlspecialchars($inner, ENT_QUOTES, 'UTF-8')
|
||
. '" data-check-ids="' . $groupIds . '">[' . $innerMarked . ']</blue>';
|
||
$replacements[] = [
|
||
'start' => $tagStart,
|
||
'end' => $tagEnd,
|
||
'html' => $newHtml,
|
||
'delta' => strlen($newHtml) - ($tagEnd - $tagStart),
|
||
];
|
||
}
|
||
usort($replacements, function ($a, $b) {
|
||
return $b['start'] - $a['start'];
|
||
});
|
||
foreach ($replacements as $rep) {
|
||
$html = substr($html, 0, $rep['start']) . $rep['html'] . substr($html, $rep['end']);
|
||
$citeDeltas[] = ['start' => $rep['start'], 'delta' => $rep['delta']];
|
||
}
|
||
}
|
||
|
||
$shiftByCite = function ($pos) use ($citeDeltas) {
|
||
$d = 0;
|
||
foreach ($citeDeltas as $cd) {
|
||
if ($cd['start'] < $pos) {
|
||
$d += $cd['delta'];
|
||
}
|
||
}
|
||
return $pos + $d;
|
||
};
|
||
|
||
// 2) 再标记引用句（从后往前）
|
||
if (!empty($contexts)) {
|
||
$spans = array_values($contexts);
|
||
usort($spans, function ($a, $b) {
|
||
return $b['text_start'] - $a['text_start'];
|
||
});
|
||
foreach ($spans as $span) {
|
||
$start = $span['text_start'];
|
||
$end = $span['text_end'];
|
||
if ($start < 0 || $end <= $start) {
|
||
continue;
|
||
}
|
||
$s = $shiftByCite($start);
|
||
$e = $shiftByCite($end);
|
||
if ($e > strlen($html)) {
|
||
$e = strlen($html);
|
||
}
|
||
$checkIds = array_values(array_unique($span['check_ids']));
|
||
$refNos = array_values(array_unique($span['ref_nos']));
|
||
sort($refNos);
|
||
$reasonParts = [];
|
||
foreach ($refNos as $rn) {
|
||
if (!empty($span['reasons'][$rn])) {
|
||
$reasonParts[] = '[' . $rn . '] ' . $span['reasons'][$rn];
|
||
}
|
||
}
|
||
$title = htmlspecialchars(
|
||
'引用句可能不合理: ' . implode('; ', $reasonParts),
|
||
ENT_QUOTES,
|
||
'UTF-8'
|
||
);
|
||
$open = '<span class="ref-context-error" data-check-ids="' . implode(',', $checkIds)
|
||
. '" data-ref-nos="' . implode(',', $refNos) . '" title="' . $title . '">';
|
||
$close = '</span>';
|
||
$html = substr($html, 0, $s) . $open . substr($html, $s, $e - $s) . $close . substr($html, $e);
|
||
}
|
||
}
|
||
|
||
return ['html' => $html, 'issues' => $issues, 'issue_count' => $issueCount];
|
||
}
|
||
|
||
/**
|
||
* @return array<int, array> refer_index => row
|
||
*/
|
||
public function loadReferMapByPArticleId($pArticleId)
|
||
{
|
||
$map = [];
|
||
if ($pArticleId <= 0) {
|
||
return $map;
|
||
}
|
||
$rows = Db::name('production_article_refer')
|
||
->where('p_article_id', $pArticleId)
|
||
->where('state', 0)
|
||
->order('index asc')
|
||
->select();
|
||
foreach ($rows as $row) {
|
||
$map[intval($row['index'])] = $row;
|
||
}
|
||
return $map;
|
||
}
|
||
public function formatReferForLlm($refer)
|
||
{
|
||
$parts = [];
|
||
foreach (['title', 'author', 'joura', 'dateno', 'refer_doi', 'doilink'] as $f) {
|
||
$v = trim((string)$this->arrGet($refer, $f, ''));
|
||
if ($v !== '') {
|
||
$parts[] = ucfirst($f) . ': ' . $v;
|
||
}
|
||
}
|
||
$content = trim((string)$this->arrGet($refer, 'refer_content', ''));
|
||
if ($content !== '') {
|
||
$parts[] = 'Reference: ' . $content;
|
||
}
|
||
return implode("\n", $parts);
|
||
}
|
||
|
||
/**
|
||
* 仅使用 refer_doi 字段（二次 Crossref 摘要用）
|
||
*/
|
||
public function extractReferDoiOnly($refer)
|
||
{
|
||
if (!is_array($refer)) {
|
||
return '';
|
||
}
|
||
$raw = trim((string)$this->arrGet($refer, 'refer_doi', ''));
|
||
if ($raw === '' || stripos($raw, 'not available') !== false) {
|
||
return '';
|
||
}
|
||
$dois = $this->extractDoisFromString($raw);
|
||
return empty($dois) ? '' : $dois[0];
|
||
}
|
||
|
||
/**
|
||
* 根据 refer_doi 调用 Crossref works API 获取摘要（二次校对专用）
|
||
*
|
||
* @return array{text:string, has_abstract:bool, doi:string}
|
||
*/
|
||
public function fetchCrossrefAbstractByReferDoi($refer)
|
||
{
|
||
$doi = $this->extractReferDoiOnly($refer);
|
||
if ($doi === '') {
|
||
return ['text' => '', 'has_abstract' => false, 'doi' => ''];
|
||
}
|
||
|
||
$crossref = new CrossrefService([
|
||
'mailto' => trim((string)Env::get('crossref_mailto', '')),
|
||
]);
|
||
$block = $this->extractCrossrefBlock($doi, $crossref);
|
||
if ($block === null) {
|
||
return ['text' => '', 'has_abstract' => false, 'doi' => $doi];
|
||
}
|
||
|
||
return [
|
||
'text' => $block['text'],
|
||
'has_abstract' => !empty($block['has_abstract']),
|
||
'doi' => $doi,
|
||
];
|
||
}
|
||
|
||
/**
|
||
* 解析 LLM 返回的 can_support
|
||
*/
|
||
public function parseLlmCanSupport($llmResult)
|
||
{
|
||
if (!is_array($llmResult)) {
|
||
return false;
|
||
}
|
||
if (array_key_exists('can_support', $llmResult)) {
|
||
return $this->parseLlmIsMatch($llmResult['can_support']);
|
||
}
|
||
return $this->parseLlmIsMatch(isset($llmResult['is_match']) ? $llmResult['is_match'] : false);
|
||
}
|
||
|
||
/**
|
||
* 第一次校对：取 article_main.content（整节正文）
|
||
*/
|
||
public function resolveMainContentForJob(array $row, $maxChars = 8000)
|
||
{
|
||
$amId = intval($this->arrGet($row, 'am_id', 0));
|
||
if ($amId <= 0) {
|
||
return '';
|
||
}
|
||
$main = Db::name('article_main')
|
||
->field('content')
|
||
->where('am_id', $amId)
|
||
->find();
|
||
if (empty($main)) {
|
||
return '';
|
||
}
|
||
|
||
$text = trim((string)$this->arrGet($main, 'content', ''));
|
||
if ($text === '') {
|
||
return '';
|
||
}
|
||
|
||
$text = preg_replace('/<blue>\[([\d,\-\s]+)\]<\/blue>/', '[$1]', $text);
|
||
$text = strip_tags($text);
|
||
$text = html_entity_decode($text, ENT_QUOTES | ENT_HTML5, 'UTF-8');
|
||
$text = preg_replace('/\s+/u', ' ', $text);
|
||
$text = trim($text);
|
||
|
||
$maxChars = max(500, intval($maxChars));
|
||
if (mb_strlen($text) > $maxChars) {
|
||
$text = mb_substr($text, 0, $maxChars) . '...';
|
||
}
|
||
|
||
return $text;
|
||
}
|
||
|
||
/**
|
||
* 引用处局部上下文（origin_text），供其它场景使用
|
||
*/
|
||
public function resolveCitationContextForJob(array $row)
|
||
{
|
||
$text = trim((string)$this->arrGet($row, 'origin_text', ''));
|
||
if ($text === '') {
|
||
$text = trim((string)$this->arrGet($row, 'content_a', ''));
|
||
}
|
||
return $text;
|
||
}
|
||
|
||
/**
|
||
* 从 refer 行提取标准 DOI（10.xxxx/...）
|
||
*
|
||
* 优先级：refer_content（原始引用文本里的 DOI 最贴近实际被引用的文献）
|
||
* > refer_doi > doi > doilink
|
||
*/
|
||
public function extractDoiFromRefer($refer)
|
||
{
|
||
$list = $this->extractAllDoiCandidatesFromRefer($refer);
|
||
return empty($list) ? '' : $list[0];
|
||
}
|
||
|
||
/**
|
||
* 返回 refer 行可能对应的全部 DOI 候选（去重，按优先级排序）
|
||
*
|
||
* 用于第二轮 DOI 复核场景：当 metadata 的 refer_doi 与原始引用文本里的 DOI
|
||
* 不一致时（数据漂移），优先尝试原始引用文本里的 DOI 抓真实摘要。
|
||
*
|
||
* @return string[]
|
||
*/
|
||
public function extractAllDoiCandidatesFromRefer($refer)
|
||
{
|
||
if (!is_array($refer)) {
|
||
return [];
|
||
}
|
||
$ordered = [
|
||
(string)$this->arrGet($refer, 'refer_content', ''),
|
||
(string)$this->arrGet($refer, 'refer_doi', ''),
|
||
(string)$this->arrGet($refer, 'doi', ''),
|
||
(string)$this->arrGet($refer, 'doilink', ''),
|
||
];
|
||
|
||
$result = [];
|
||
foreach ($ordered as $raw) {
|
||
foreach ($this->extractDoisFromString($raw) as $doi) {
|
||
if (!in_array($doi, $result, true)) {
|
||
$result[] = $doi;
|
||
}
|
||
}
|
||
}
|
||
return $result;
|
||
}
|
||
|
||
/**
|
||
* 从任意文本里抽取所有形如 10.xxxx/yyy 的 DOI
|
||
* @return string[]
|
||
*/
|
||
private function extractDoisFromString($text)
|
||
{
|
||
$text = trim((string)$text);
|
||
if ($text === '' || stripos($text, 'not available') !== false) {
|
||
return [];
|
||
}
|
||
|
||
$dois = [];
|
||
|
||
if (preg_match_all('~doi\.org/([^\s?#"\'<>]+)~i', $text, $m)) {
|
||
foreach ($m[1] as $cand) {
|
||
$cand = $this->trimDoiTail(trim($cand));
|
||
if ($this->isValidDoi($cand)) {
|
||
$dois[] = $cand;
|
||
}
|
||
}
|
||
}
|
||
|
||
if (preg_match_all('~\b(10\.\d{3,9}/[^\s?#"\'<>]+)~i', $text, $m)) {
|
||
foreach ($m[1] as $cand) {
|
||
$cand = $this->trimDoiTail(trim($cand));
|
||
if ($this->isValidDoi($cand)) {
|
||
$dois[] = $cand;
|
||
}
|
||
}
|
||
}
|
||
|
||
if ($dois === [] && strpos($text, '10.') === 0) {
|
||
$cand = $this->trimDoiTail($text);
|
||
if ($this->isValidDoi($cand)) {
|
||
$dois[] = $cand;
|
||
}
|
||
}
|
||
|
||
return array_values(array_unique($dois));
|
||
}
|
||
|
||
private function trimDoiTail($doi)
|
||
{
|
||
return rtrim($doi, ".,;:)]}>\"'\\ \t\n\r");
|
||
}
|
||
|
||
private function isValidDoi($doi)
|
||
{
|
||
return (bool)preg_match('~^10\.\d{3,9}/[^\s]+$~i', (string)$doi);
|
||
}
|
||
|
||
/**
|
||
* 通过 PubMed / Crossref 拉取 DOI 对应文献内容（本地 LLM 无法打开网页，须预先抓取）
|
||
*
|
||
* 行为：
|
||
* - 尝试 refer 行内所有 DOI 候选（refer_content > refer_doi > doi > doilink）
|
||
* - 优先采用第一个能拿到 abstract 的 DOI
|
||
* - PubMed 无摘要时回落到 Crossref raw 解析摘要（清理 JATS 标签）
|
||
* - 全部失败则返回空字符串（调用方据此跳过二次复核）
|
||
*/
|
||
public function fetchDoiLiteratureBlock($refer)
|
||
{
|
||
$candidates = $this->extractAllDoiCandidatesFromRefer($refer);
|
||
if (empty($candidates)) {
|
||
return '';
|
||
}
|
||
|
||
$pubmed = new PubmedService([
|
||
'email' => trim((string)Env::get('pubmed_email', '')),
|
||
'tool' => trim((string)Env::get('pubmed_tool', 'tmrjournals')),
|
||
]);
|
||
$crossref = new CrossrefService([
|
||
'mailto' => trim((string)Env::get('crossref_mailto', '')),
|
||
]);
|
||
|
||
$best = null;
|
||
$fallback = null;
|
||
|
||
foreach ($candidates as $doi) {
|
||
$block = $this->buildDoiBlockFromSources($doi, $pubmed, $crossref);
|
||
if ($block === null) {
|
||
continue;
|
||
}
|
||
if (!empty($block['has_abstract'])) {
|
||
$best = $block;
|
||
break;
|
||
}
|
||
if ($fallback === null) {
|
||
$fallback = $block;
|
||
}
|
||
}
|
||
|
||
$chosen = $best ?: $fallback;
|
||
if ($chosen === null) {
|
||
return '';
|
||
}
|
||
return $chosen['text'];
|
||
}
|
||
|
||
/**
|
||
* 拉单个 DOI 的真实内容，返回 ['text' => string, 'has_abstract' => bool] 或 null
|
||
*/
|
||
private function buildDoiBlockFromSources($doi, PubmedService $pubmed, CrossrefService $crossref)
|
||
{
|
||
$doi = trim((string)$doi);
|
||
if ($doi === '') {
|
||
return null;
|
||
}
|
||
|
||
$pub = $pubmed->fetchByDoi($doi);
|
||
$pubAbstract = is_array($pub) ? trim((string)$this->arrGet($pub, 'abstract', '')) : '';
|
||
|
||
if (is_array($pub) && ($pubAbstract !== '' || trim((string)$this->arrGet($pub, 'title', '')) !== '')) {
|
||
$lines = ['Source: PubMed (DOI ' . $doi . ')'];
|
||
if (!empty($pub['title'])) {
|
||
$lines[] = 'Actual Title: ' . trim((string)$pub['title']);
|
||
}
|
||
if (!empty($pub['journal'])) {
|
||
$lines[] = 'Journal: ' . trim((string)$pub['journal']);
|
||
}
|
||
if (!empty($pub['year'])) {
|
||
$lines[] = 'Year: ' . trim((string)$pub['year']);
|
||
}
|
||
if (!empty($pub['publication_types'])) {
|
||
$lines[] = 'Publication Types: ' . implode('; ', (array)$pub['publication_types']);
|
||
}
|
||
if (!empty($pub['mesh_terms'])) {
|
||
$lines[] = 'MeSH: ' . implode('; ', (array)$pub['mesh_terms']);
|
||
}
|
||
if ($pubAbstract !== '') {
|
||
$lines[] = 'Abstract: ' . $this->truncate($pubAbstract, 3500);
|
||
}
|
||
|
||
if ($pubAbstract === '') {
|
||
$cr = $this->extractCrossrefBlock($doi, $crossref);
|
||
if ($cr !== null && $cr['has_abstract']) {
|
||
$lines[] = "\n--- Crossref 补充 ---\n" . $cr['text'];
|
||
return ['text' => implode("\n", $lines), 'has_abstract' => true];
|
||
}
|
||
}
|
||
|
||
return ['text' => implode("\n", $lines), 'has_abstract' => $pubAbstract !== ''];
|
||
}
|
||
|
||
return $this->extractCrossrefBlock($doi, $crossref);
|
||
}
|
||
|
||
/**
|
||
* 从 Crossref 拉取标题/期刊/作者/摘要（abstract 通常包裹 JATS XML，需清洗）
|
||
* @return array|null ['text' => string, 'has_abstract' => bool]
|
||
*/
|
||
private function extractCrossrefBlock($doi, CrossrefService $crossref)
|
||
{
|
||
$msg = $crossref->fetchWork($doi);
|
||
if (!is_array($msg)) {
|
||
return null;
|
||
}
|
||
|
||
$summary = $crossref->fetchWorkSummary($doi);
|
||
if (!is_array($summary)) {
|
||
$summary = [];
|
||
}
|
||
|
||
$lines = ['Source: Crossref api.crossref.org/works/' . rawurlencode($doi)];
|
||
$title = isset($msg['title'][0]) ? trim((string)$msg['title'][0]) : trim((string)$this->arrGet($summary, 'title', ''));
|
||
if ($title !== '') {
|
||
$lines[] = 'Actual Title: ' . $title;
|
||
}
|
||
if (!empty($summary['joura'])) {
|
||
$lines[] = 'Journal: ' . trim((string)$summary['joura']);
|
||
}
|
||
if (!empty($summary['author_str'])) {
|
||
$lines[] = 'Authors: ' . trim((string)$summary['author_str']);
|
||
}
|
||
if (!empty($summary['dateno'])) {
|
||
$lines[] = 'Publication: ' . trim((string)$summary['dateno']);
|
||
}
|
||
if (!empty($summary['doilink'])) {
|
||
$lines[] = 'DOI Link: ' . trim((string)$summary['doilink']);
|
||
}
|
||
if (!empty($summary['is_retracted'])) {
|
||
$lines[] = 'Retraction: yes - ' . trim((string)$this->arrGet($summary, 'retract_reason', ''));
|
||
}
|
||
|
||
$abstract = $this->cleanCrossrefAbstract((string)$this->arrGet($msg, 'abstract', ''));
|
||
$hasAbstract = $abstract !== '';
|
||
if ($hasAbstract) {
|
||
$lines[] = 'Abstract: ' . $this->truncate($abstract, 3500);
|
||
} else {
|
||
$lines[] = 'Note: Crossref 未返回摘要，请结合标题/期刊/作者与正文谨慎判断。';
|
||
}
|
||
|
||
return ['text' => implode("\n", $lines), 'has_abstract' => $hasAbstract];
|
||
}
|
||
|
||
private function cleanCrossrefAbstract($raw)
|
||
{
|
||
$raw = trim((string)$raw);
|
||
if ($raw === '') {
|
||
return '';
|
||
}
|
||
$raw = preg_replace('~<jats:title[^>]*>.*?</jats:title>~is', '', $raw);
|
||
$raw = preg_replace('~<jats:p[^>]*>~i', "\n", $raw);
|
||
$raw = preg_replace('~</jats:p>~i', '', $raw);
|
||
$raw = preg_replace('~</?jats:[^>]+>~i', '', $raw);
|
||
$raw = strip_tags($raw);
|
||
$raw = preg_replace('/[ \t]+/u', ' ', $raw);
|
||
$raw = preg_replace("/\r\n|\r/u", "\n", $raw);
|
||
$raw = preg_replace("/\n{2,}/u", "\n", $raw);
|
||
return trim($raw);
|
||
}
|
||
|
||
private function truncate($text, $max)
|
||
{
|
||
$text = (string)$text;
|
||
if (mb_strlen($text) <= $max) {
|
||
return $text;
|
||
}
|
||
return mb_substr($text, 0, $max) . '...';
|
||
}
|
||
|
||
/**
|
||
* 第二次 DOI 复核数据准备：返回书目信息 + 真实抓取内容
|
||
*
|
||
* @return array{refer_text:string, doi_block:string, has_abstract:bool, doi_used:string}
|
||
*/
|
||
public function prepareRecheckPayload($refer, $referText = '')
|
||
{
|
||
$base = trim($referText) !== '' ? trim($referText) : $this->formatReferForLlm($refer);
|
||
$cr = $this->fetchCrossrefAbstractByReferDoi($refer);
|
||
return [
|
||
'refer_text' => $base,
|
||
'doi_block' => $cr['text'],
|
||
'has_abstract' => $cr['has_abstract'],
|
||
'doi_used' => $cr['doi'],
|
||
];
|
||
}
|
||
|
||
/**
|
||
* 旧接口：拼接成单块文本（向后兼容，建议调用方改用 prepareRecheckPayload）
|
||
*/
|
||
public function formatReferForDoiRecheck($refer, $referText = '')
|
||
{
|
||
$payload = $this->prepareRecheckPayload($refer, $referText);
|
||
if ($payload['doi_block'] === '') {
|
||
return $payload['refer_text']
|
||
. "\n\n【DOI 文献真实内容】\n未能从 PubMed/Crossref 获取该 DOI 的摘要或元数据，请依据书目条目与正文谨慎判断。";
|
||
}
|
||
return $payload['refer_text']
|
||
. "\n\n【Crossref 摘要（依据 Refer_doi 从 api.crossref.org/works 获取）】\n"
|
||
. $payload['doi_block'];
|
||
}
|
||
|
||
/**
|
||
* 第一轮 confidence<=0.65 且能抓到 DOI 真实内容时，延迟入队第二轮复核
|
||
*
|
||
* 跳过条件（避免无意义重跑得到相同结果）：
|
||
* - check_id 不合法 / 一次置信度高于阈值
|
||
* - refer 行不存在
|
||
* - refer_doi 为空或 Crossref 未返回摘要
|
||
*/
|
||
public function maybeEnqueueSecondPass($checkId, $confidence)
|
||
{
|
||
$checkId = intval($checkId);
|
||
$confidence = floatval($confidence);
|
||
if ($checkId <= 0 || $confidence > 0.65) {
|
||
return false;
|
||
}
|
||
|
||
$row = Db::name('article_reference_check_result')->where('id', $checkId)->find();
|
||
if (empty($row)) {
|
||
return false;
|
||
}
|
||
|
||
$refer = null;
|
||
if (intval($row['p_refer_id']) > 0) {
|
||
$refer = Db::name('production_article_refer')
|
||
->where('p_refer_id', intval($row['p_refer_id']))
|
||
->where('state', 0)
|
||
->find();
|
||
}
|
||
if (empty($refer) || $this->extractReferDoiOnly($refer) === '') {
|
||
return false;
|
||
}
|
||
|
||
$cr = $this->fetchCrossrefAbstractByReferDoi($refer);
|
||
if (empty($cr['has_abstract'])) {
|
||
return false;
|
||
}
|
||
|
||
$this->pushJob2($checkId, 5);
|
||
return true;
|
||
}
|
||
|
||
/**
|
||
* 从 article_main.content 提取 blue 引用
|
||
*/
|
||
public function extractReferences($content)
|
||
{
|
||
$result = [];
|
||
preg_match_all('/<blue>\[([\d,\-\s]+)\]<\/blue>/', $content, $matches, PREG_OFFSET_CAPTURE);
|
||
if (empty($matches[0])) {
|
||
return [];
|
||
}
|
||
|
||
$tagSpans = [];
|
||
foreach ($matches[0] as $index => $match) {
|
||
$tagSpans[] = [
|
||
'start' => $match[1],
|
||
'end' => $match[1] + strlen($match[0]),
|
||
'index' => $index,
|
||
];
|
||
}
|
||
|
||
foreach ($matches[0] as $index => $match) {
|
||
$fullTag = $match[0];
|
||
$tagStart = $match[1];
|
||
$tagEnd = $tagStart + strlen($fullTag);
|
||
$rawRef = trim($matches[1][$index][0]);
|
||
$referenceNumbers = $this->expandReferenceNumbers($rawRef);
|
||
|
||
list($localStart, $localEnd, $originalText) = $this->extractLocalCitationContext(
|
||
$content,
|
||
$tagStart,
|
||
$tagEnd,
|
||
$tagSpans
|
||
);
|
||
|
||
if (!$this->isMeaningfulCitationContext($originalText) || empty($referenceNumbers)) {
|
||
continue;
|
||
}
|
||
|
||
$result[] = [
|
||
'reference_raw' => $rawRef,
|
||
'reference_numbers' => $referenceNumbers,
|
||
'original_text' => $originalText,
|
||
'reference_start' => $tagStart,
|
||
'reference_end' => $tagEnd,
|
||
'text_start' => $localStart,
|
||
'text_end' => $localEnd,
|
||
];
|
||
}
|
||
|
||
return $result;
|
||
}
|
||
|
||
/**
|
||
* 按引用位置截取局部上下文：优先取标签前叙述；同句多引时后续引用从上一标签后开始。
|
||
*/
|
||
private function extractLocalCitationContext($content, $tagStart, $tagEnd, array $tagSpans)
|
||
{
|
||
$paragraphStart = $this->findParagraphStart($content, $tagStart);
|
||
$sentenceEnd = $this->findSentenceEnd($content, $tagEnd, $tagEnd);
|
||
|
||
$prevTagEnd = $paragraphStart;
|
||
$nextTagStart = $sentenceEnd;
|
||
foreach ($tagSpans as $span) {
|
||
if ($span['end'] <= $tagStart && $span['end'] > $prevTagEnd) {
|
||
$prevTagEnd = $span['end'];
|
||
}
|
||
if ($span['start'] > $tagEnd && $span['start'] < $nextTagStart) {
|
||
$nextTagStart = $span['start'];
|
||
}
|
||
}
|
||
|
||
$hasPriorCiteInParagraph = ($prevTagEnd > $paragraphStart);
|
||
$sentenceStart = $this->findSentenceStart($content, $tagStart);
|
||
|
||
// 段内首个引用：整段到标签前；后续引用：取「本句」起点（可早于上一标签），避免只剩 “and external environment” 再误用标签后文本
|
||
if ($hasPriorCiteInParagraph) {
|
||
$localStart = max($paragraphStart, $sentenceStart);
|
||
} else {
|
||
$localStart = $this->capContextStartBeforeTag($content, $tagStart, $paragraphStart);
|
||
}
|
||
|
||
// 默认：引用标签前的论述
|
||
$localEnd = $tagStart;
|
||
$originalText = $this->buildCitationContextText($content, $localStart, $localEnd);
|
||
|
||
// 仅段内首个引用、且标签前极短（如句末 ICU nurses [14]）时，才改用标签后片段；同段多引禁止标签后截取（会错取下一句）
|
||
$allowTrailing = !$hasPriorCiteInParagraph;
|
||
if ($allowTrailing && (
|
||
!$this->isMeaningfulCitationContext($originalText)
|
||
|| $this->shouldUseTrailingCitationContext($content, $localStart, $tagStart, $tagEnd)
|
||
)) {
|
||
$trailEnd = ($nextTagStart < $sentenceEnd) ? $nextTagStart : $sentenceEnd;
|
||
$trailText = $this->buildCitationContextText($content, $tagEnd, $trailEnd);
|
||
if ($this->isMeaningfulCitationContext($trailText)) {
|
||
$localStart = $tagEnd;
|
||
$localEnd = $trailEnd;
|
||
$originalText = $trailText;
|
||
}
|
||
}
|
||
|
||
if (!$this->isMeaningfulCitationContext($originalText)) {
|
||
list($localStart, $localEnd) = $this->widenCitationContextBounds(
|
||
$content,
|
||
$tagStart,
|
||
$tagEnd,
|
||
$localStart,
|
||
$localEnd
|
||
);
|
||
$originalText = $this->buildCitationContextText($content, $localStart, $localEnd);
|
||
}
|
||
|
||
return [$localStart, $localEnd, $originalText];
|
||
}
|
||
|
||
/**
|
||
* 标签前仅有作者缩写等极短片段时，改用标签后上下文
|
||
*/
|
||
private function shouldUseTrailingCitationContext($content, $localStart, $tagStart, $tagEnd)
|
||
{
|
||
$before = $this->buildCitationContextText($content, $localStart, $tagStart);
|
||
if (!$this->isMeaningfulCitationContext($before)) {
|
||
return true;
|
||
}
|
||
|
||
return mb_strlen($before) < 25;
|
||
}
|
||
|
||
public function expandReferenceNumbers($refStr)
|
||
{
|
||
$refStr = str_replace(
|
||
['，', '–', '—', '−', '‐', '‑'],
|
||
[',', '-', '-', '-', '-', '-'],
|
||
trim($refStr)
|
||
);
|
||
$numbers = [];
|
||
foreach (explode(',', $refStr) as $part) {
|
||
$part = trim($part);
|
||
if ($part === '') {
|
||
continue;
|
||
}
|
||
if (preg_match('/^(\d+)\s*-\s*(\d+)$/', $part, $m)) {
|
||
$start = intval($m[1]);
|
||
$end = intval($m[2]);
|
||
if ($start <= $end) {
|
||
$numbers = array_merge($numbers, range($start, $end));
|
||
}
|
||
} elseif (ctype_digit($part)) {
|
||
$numbers[] = intval($part);
|
||
}
|
||
}
|
||
return array_values(array_unique($numbers));
|
||
}
|
||
|
||
/**
|
||
* 返回 $bytePos 处 UTF-8 码点占用的最后一字节之后的位置（下一字符起始）
|
||
*/
|
||
private function utf8CharEnd($content, $bytePos)
|
||
{
|
||
$len = strlen($content);
|
||
if ($bytePos < 0 || $bytePos >= $len) {
|
||
return max(0, min($len, $bytePos + 1));
|
||
}
|
||
$next = $bytePos + 1;
|
||
while ($next < $len && (ord($content[$next]) & 0xC0) === 0x80) {
|
||
$next++;
|
||
}
|
||
|
||
return $next;
|
||
}
|
||
|
||
/**
|
||
* 按字节偏移截取（与 strpos/strlen 一致）；勿用 mb_substr，否则遇中文前缀会截断英文词头
|
||
*/
|
||
private function byteSubstr($content, $start, $end)
|
||
{
|
||
$length = max(0, $end - $start);
|
||
if ($length === 0) {
|
||
return '';
|
||
}
|
||
|
||
return (string)mb_strcut($content, $start, $length, 'UTF-8');
|
||
}
|
||
|
||
private function buildCitationContextText($content, $start, $end)
|
||
{
|
||
$text = $this->byteSubstr($content, $start, $end);
|
||
$text = preg_replace('/<blue>\[[\d,\-\s]+\]<\/blue>/', '', $text);
|
||
$text = trim(strip_tags($text));
|
||
$text = preg_replace('/\s+/u', ' ', $text);
|
||
$text = ltrim($text, "\xEF\xBB\xBF");
|
||
|
||
return $text;
|
||
}
|
||
|
||
/**
|
||
* 过滤仅标点、过短或无字母/汉字的上下文（如去掉标签后只剩 "."）
|
||
*/
|
||
private function isMeaningfulCitationContext($text)
|
||
{
|
||
$text = trim($text);
|
||
if ($text === '') {
|
||
return false;
|
||
}
|
||
if ($this->isOnlyPunctuationOrSpace($text)) {
|
||
return false;
|
||
}
|
||
if (!preg_match('/[\p{L}\p{N}]/u', $text)) {
|
||
return false;
|
||
}
|
||
|
||
return mb_strlen($text) >= 2;
|
||
}
|
||
|
||
private function isOnlyPunctuationOrSpace($text)
|
||
{
|
||
return preg_match('/^[\s\p{P}\p{S}]+$/u', $text) === 1;
|
||
}
|
||
|
||
/**
|
||
* 首句过短时向前后各扩展一句（上限约 2000 字符）
|
||
*/
|
||
private function widenCitationContextBounds($content, $tagStart, $tagEnd, $start, $end)
|
||
{
|
||
$len = strlen($content);
|
||
$maxSpan = 2000;
|
||
|
||
if ($start > 0) {
|
||
$prevStart = $this->findSentenceStart($content, max(0, $start - 1));
|
||
if ($prevStart < $start) {
|
||
$start = $prevStart;
|
||
}
|
||
}
|
||
|
||
$nextEnd = $this->findSentenceEnd($content, $end, $tagEnd);
|
||
if ($nextEnd > $end && $nextEnd <= $len) {
|
||
$end = $nextEnd;
|
||
}
|
||
|
||
if ($end - $start > $maxSpan) {
|
||
$half = (int)floor($maxSpan / 2);
|
||
$mid = (int)floor(($tagStart + $tagEnd) / 2);
|
||
$start = max(0, $mid - $half);
|
||
$end = min($len, $start + $maxSpan);
|
||
}
|
||
|
||
return [$start, $end];
|
||
}
|
||
|
||
/**
|
||
* 句号是否可作为句界（排除小数点、et al. 等缩写）
|
||
*/
|
||
private function isSentenceDelimiterAt($content, $pos, $delimiter)
|
||
{
|
||
$len = strlen($content);
|
||
if ($delimiter !== '.' || $pos < 0 || $pos >= $len) {
|
||
return true;
|
||
}
|
||
if ($pos > 0 && $pos + 1 < $len
|
||
&& ctype_digit($content[$pos - 1])
|
||
&& ctype_digit($content[$pos + 1])
|
||
) {
|
||
return false;
|
||
}
|
||
|
||
$before = substr($content, max(0, $pos - 12), min(12, $pos));
|
||
if (preg_match('/\b(et\s+al|e\.g|i\.e|vs|etc|fig|no)\s*\.?\s*$/i', $before)) {
|
||
return false;
|
||
}
|
||
|
||
$after = substr($content, $pos + 1, 24);
|
||
if (preg_match('/^\s*<blue>\s*\[/', $after)) {
|
||
return false;
|
||
}
|
||
|
||
return true;
|
||
}
|
||
|
||
/**
|
||
* 段落起始（HTML / 换行），避免英文多句段落只取到最后一个句号后的一句
|
||
*/
|
||
private function findParagraphStart($content, $tagStart)
|
||
{
|
||
$search = substr($content, 0, max(0, $tagStart));
|
||
if ($search === '') {
|
||
return 0;
|
||
}
|
||
|
||
$best = 0;
|
||
|
||
if (preg_match_all('/<p[^>]*>/i', $search, $m, PREG_OFFSET_CAPTURE)) {
|
||
$last = end($m[0]);
|
||
$best = max($best, $last[1] + strlen($last[0]));
|
||
}
|
||
if (preg_match_all('/<\/p>\s*/i', $search, $m, PREG_OFFSET_CAPTURE)) {
|
||
$last = end($m[0]);
|
||
$best = max($best, $last[1] + strlen($last[0]));
|
||
}
|
||
if (preg_match_all('/<br\s*\/?>\s*/i', $search, $m, PREG_OFFSET_CAPTURE)) {
|
||
$last = end($m[0]);
|
||
$best = max($best, $last[1] + strlen($last[0]));
|
||
}
|
||
|
||
$pos = strrpos($search, "\n\n");
|
||
if ($pos !== false) {
|
||
$best = max($best, $pos + 2);
|
||
}
|
||
$pos = strrpos($search, "\n");
|
||
if ($pos !== false) {
|
||
$best = max($best, $pos + 1);
|
||
}
|
||
|
||
return $best;
|
||
}
|
||
|
||
/**
|
||
* 段落过长时从引用处向前截取上限，避免单次 LLM 上下文过大
|
||
*/
|
||
private function capContextStartBeforeTag($content, $tagStart, $paragraphStart, $maxBytes = 2500)
|
||
{
|
||
if ($tagStart - $paragraphStart <= $maxBytes) {
|
||
return $paragraphStart;
|
||
}
|
||
|
||
$start = $tagStart - $maxBytes;
|
||
$slice = substr($content, $start, $tagStart - $start);
|
||
if (preg_match('/[.!?。！？]\s+/u', $slice, $m, PREG_OFFSET_CAPTURE)) {
|
||
$last = end($m[0]);
|
||
$rel = $last[1] + strlen($last[0]);
|
||
return $start + $rel;
|
||
}
|
||
|
||
return max($paragraphStart, $start);
|
||
}
|
||
|
||
private function findSentenceStart($content, $position)
|
||
{
|
||
$start = 0;
|
||
foreach (['.', '。', '!', '?', "\n"] as $delimiter) {
|
||
$pos = strrpos(substr($content, 0, $position), $delimiter);
|
||
if ($pos !== false && $this->isSentenceDelimiterAt($content, $pos, $delimiter)) {
|
||
$start = max($start, $this->utf8CharEnd($content, $pos));
|
||
}
|
||
}
|
||
return $start;
|
||
}
|
||
|
||
/**
|
||
* @param int $searchFrom 从该字节位置起查找句末
|
||
* @param int $tagEnd 引用标签结束位置；用于跳过 </blue> 后紧跟的孤立句号
|
||
*/
|
||
private function findSentenceEnd($content, $searchFrom, $tagEnd = 0)
|
||
{
|
||
$length = strlen($content);
|
||
$minPos = max(0, $searchFrom);
|
||
|
||
while ($minPos < $length) {
|
||
$endPositions = [];
|
||
foreach (['.', '。', '!', '?', "\n"] as $delimiter) {
|
||
$pos = strpos($content, $delimiter, $minPos);
|
||
if ($pos !== false && $this->isSentenceDelimiterAt($content, $pos, $delimiter)) {
|
||
$endPositions[] = $this->utf8CharEnd($content, $pos);
|
||
}
|
||
}
|
||
if (empty($endPositions)) {
|
||
return $length;
|
||
}
|
||
|
||
$end = min($endPositions);
|
||
if ($tagEnd <= 0 || $end <= $tagEnd) {
|
||
return $end;
|
||
}
|
||
|
||
$gap = substr($content, $tagEnd, $end - $tagEnd);
|
||
$gapText = trim(strip_tags(preg_replace('/<blue>\[[\d,\-\s]+\]<\/blue>/', '', $gap)));
|
||
if ($gapText !== '' && !$this->isOnlyPunctuationOrSpace($gapText)) {
|
||
return $end;
|
||
}
|
||
|
||
$minPos = $end;
|
||
}
|
||
|
||
return $length;
|
||
}
|
||
|
||
private function pushJob($checkId, $delaySeconds = 0)
|
||
{
|
||
$jobClass = 'app\api\job\ReferenceCheck@fire';
|
||
$data = ['check_id' => $checkId];
|
||
try {
|
||
if ($delaySeconds > 0) {
|
||
$jobId = Queue::later($delaySeconds, $jobClass, $data, self::QUEUE_NAME);
|
||
} else {
|
||
$jobId = Queue::push($jobClass, $data, self::QUEUE_NAME);
|
||
}
|
||
} catch (\Exception $e) {
|
||
\think\Log::error('ReferenceCheck pushJob failed check_id=' . $checkId . ' ' . $e->getMessage());
|
||
throw $e;
|
||
}
|
||
}
|
||
private function pushJob2($checkId, $delaySeconds = 0)
|
||
{
|
||
$jobClass = 'app\api\job\ReferenceCheckTwo@fire';
|
||
$data = ['check_id' => $checkId];
|
||
try {
|
||
if ($delaySeconds > 0) {
|
||
$jobId = Queue::later($delaySeconds, $jobClass, $data, self::QUEUE_NAME);
|
||
} else {
|
||
$jobId = Queue::push($jobClass, $data, self::QUEUE_NAME);
|
||
}
|
||
} catch (\Exception $e) {
|
||
\think\Log::error('ReferenceCheckTwo pushJob failed check_id=' . $checkId . ' ' . $e->getMessage());
|
||
throw $e;
|
||
}
|
||
}
|
||
}
|