359 lines
12 KiB
PHP
359 lines
12 KiB
PHP
<?php
|
||
|
||
namespace app\common;
|
||
|
||
use app\common\service\LLMService;
|
||
use Smalot\PdfParser\Parser as PdfParser;
|
||
use think\Db;
|
||
use think\Env;
|
||
use think\Exception;
|
||
|
||
/**
|
||
* 从用户简历文件提取文本，调用与参考文献校对相同的 LLMService 解析用户基本信息。
|
||
*/
|
||
class UserInfoFromFileService
|
||
{
|
||
const CV_BASE_URL_DEFAULT = 'https://submission.tmrjournals.com/public/reviewer/';
|
||
|
||
/** @var LLMService */
|
||
private $llm;
|
||
|
||
public function __construct(LLMService $llm = null)
|
||
{
|
||
$this->llm = $llm ?: new LLMService();
|
||
}
|
||
|
||
/**
|
||
* 按 user_id 查 user_cv，取最新一条简历解析。
|
||
*
|
||
* @param int $userId
|
||
* @return array
|
||
*/
|
||
public function parseFromUserId($userId)
|
||
{
|
||
$userId = intval($userId);
|
||
if ($userId <= 0) {
|
||
throw new Exception('user_id 无效');
|
||
}
|
||
|
||
$row = Db::name('user_cv')
|
||
->where('user_id', $userId)
|
||
->where('state', 0)
|
||
->order('ctime desc')
|
||
->find();
|
||
if (empty($row) || trim((string) $row['cv']) === '') {
|
||
throw new Exception('未找到该用户的简历（user_cv）');
|
||
}
|
||
|
||
$cvRel = $this->normalizeCvRelativePath($row['cv']);
|
||
$cvUrl = $this->buildCvUrl($cvRel);
|
||
$local = $this->resolveLocalCvPath($cvRel);
|
||
$temp = false;
|
||
$filePath = $local;
|
||
|
||
if ($filePath === '' || !is_file($filePath)) {
|
||
$filePath = $this->downloadCvToTemp($cvUrl, $cvRel);
|
||
$temp = true;
|
||
}
|
||
|
||
try {
|
||
$result = $this->parseFromFile($filePath);
|
||
} finally {
|
||
if ($temp && is_file($filePath)) {
|
||
@unlink($filePath);
|
||
}
|
||
}
|
||
|
||
$result['user_id'] = $userId;
|
||
$result['user_cv_id'] = intval($row['user_cv_id'] ?? 0);
|
||
$result['cv'] = $cvRel;
|
||
$result['cv_url'] = $cvUrl;
|
||
|
||
return $result;
|
||
}
|
||
|
||
/**
|
||
* @param string $filePath 服务器本地绝对路径
|
||
* @return array
|
||
*/
|
||
public function parseFromFile($filePath)
|
||
{
|
||
$filePath = trim((string) $filePath);
|
||
if ($filePath === '' || !is_file($filePath)) {
|
||
throw new Exception('文件不存在');
|
||
}
|
||
|
||
$text = trim($this->extractFileText($filePath));
|
||
if ($text === '') {
|
||
throw new Exception('未能从文件中提取到文本');
|
||
}
|
||
|
||
$text = $this->sanitizeUtf8($text);
|
||
$maxChars = max(1000, (int) Env::get('promotion.promotion_llm_max_chars', 2500));
|
||
if (mb_strlen($text) > $maxChars) {
|
||
$text = mb_substr($text, 0, $maxChars) . "\n...(truncated)";
|
||
}
|
||
|
||
$llmResult = $this->parseUserInfoWithLlm($text);
|
||
|
||
return [
|
||
'file' => basename($filePath),
|
||
'text_length' => mb_strlen($text),
|
||
'text_preview' => mb_substr($text, 0, 600),
|
||
'user_info' => $llmResult['user_info'],
|
||
'llm_raw' => $llmResult['raw'],
|
||
];
|
||
}
|
||
|
||
private function normalizeCvRelativePath($cv)
|
||
{
|
||
$cv = trim(str_replace('\\', '/', (string) $cv));
|
||
if ($cv === '') {
|
||
return '';
|
||
}
|
||
if (preg_match('#^https?://#i', $cv)) {
|
||
return $cv;
|
||
}
|
||
return ltrim($cv, '/');
|
||
}
|
||
|
||
private function buildCvUrl($cvRel)
|
||
{
|
||
if (preg_match('#^https?://#i', $cvRel)) {
|
||
return $cvRel;
|
||
}
|
||
$base = rtrim(trim((string) Env::get('reviewer.cv_base_url', self::CV_BASE_URL_DEFAULT)), '/') . '/';
|
||
return $base . $cvRel;
|
||
}
|
||
|
||
private function resolveLocalCvPath($cvRel)
|
||
{
|
||
if (preg_match('#^https?://#i', $cvRel)) {
|
||
return '';
|
||
}
|
||
$path = ROOT_PATH . 'public' . DS . 'reviewer' . DS . str_replace('/', DS, $cvRel);
|
||
return is_file($path) ? $path : '';
|
||
}
|
||
|
||
private function downloadCvToTemp($url, $cvRel)
|
||
{
|
||
$ext = strtolower(pathinfo(parse_url($cvRel, PHP_URL_PATH) ?: $cvRel, PATHINFO_EXTENSION));
|
||
if ($ext === '') {
|
||
$ext = 'dat';
|
||
}
|
||
|
||
$saveDir = ROOT_PATH . 'runtime' . DS . 'user_file_parse';
|
||
if (!is_dir($saveDir)) {
|
||
@mkdir($saveDir, 0755, true);
|
||
}
|
||
$path = $saveDir . DS . date('YmdHis') . '_' . uniqid('', true) . '.' . $ext;
|
||
|
||
$ch = curl_init($url);
|
||
curl_setopt_array($ch, [
|
||
CURLOPT_RETURNTRANSFER => true,
|
||
CURLOPT_FOLLOWLOCATION => true,
|
||
CURLOPT_CONNECTTIMEOUT => 15,
|
||
CURLOPT_TIMEOUT => 90,
|
||
CURLOPT_SSL_VERIFYPEER => false,
|
||
CURLOPT_SSL_VERIFYHOST => 0,
|
||
]);
|
||
$body = curl_exec($ch);
|
||
$code = (int) curl_getinfo($ch, CURLINFO_HTTP_CODE);
|
||
$err = curl_error($ch);
|
||
curl_close($ch);
|
||
|
||
if ($body === false || $code < 200 || $code >= 300) {
|
||
throw new Exception('下载简历失败（HTTP ' . $code . '）: ' . $url . ($err !== '' ? ' ' . $err : ''));
|
||
}
|
||
if (@file_put_contents($path, $body) === false) {
|
||
throw new Exception('保存简历临时文件失败');
|
||
}
|
||
return $path;
|
||
}
|
||
|
||
private function extractFileText($filePath)
|
||
{
|
||
$ext = strtolower(pathinfo($filePath, PATHINFO_EXTENSION));
|
||
|
||
if (in_array($ext, ['txt', 'text', 'md', 'csv', 'log'], true)) {
|
||
$raw = @file_get_contents($filePath);
|
||
return is_string($raw) ? $raw : '';
|
||
}
|
||
|
||
if ($ext === 'pdf') {
|
||
return $this->extractPdfText($filePath);
|
||
}
|
||
|
||
if (in_array($ext, ['doc', 'docx'], true)) {
|
||
return $this->extractDocxText($filePath);
|
||
}
|
||
|
||
// 未识别扩展名：先尝试 Word，再尝试 PDF，最后按纯文本读
|
||
$text = $this->tryExtractDocxText($filePath);
|
||
if ($text !== '') {
|
||
return $text;
|
||
}
|
||
$text = $this->tryExtractPdfText($filePath);
|
||
if ($text !== '') {
|
||
return $text;
|
||
}
|
||
|
||
$raw = @file_get_contents($filePath);
|
||
if (!is_string($raw) || $raw === '') {
|
||
return '';
|
||
}
|
||
if ($this->looksLikeBinary($raw)) {
|
||
throw new Exception('不支持的简历文件格式: .' . ($ext !== '' ? $ext : 'unknown'));
|
||
}
|
||
return $raw;
|
||
}
|
||
|
||
private function extractPdfText($filePath)
|
||
{
|
||
$text = $this->tryExtractPdfText($filePath);
|
||
if ($text === '') {
|
||
throw new Exception('PDF 文本提取失败，请确认已安装 smalot/pdfparser：composer require smalot/pdfparser');
|
||
}
|
||
return $text;
|
||
}
|
||
|
||
private function tryExtractPdfText($filePath)
|
||
{
|
||
if (!class_exists(PdfParser::class)) {
|
||
\think\Log::warning('UserInfoFromFile: smalot/pdfparser not installed');
|
||
return '';
|
||
}
|
||
try {
|
||
$parser = new PdfParser();
|
||
$pdf = $parser->parseFile($filePath);
|
||
$text = $pdf->getText();
|
||
return is_string($text) ? $text : '';
|
||
} catch (\Throwable $e) {
|
||
\think\Log::warning('UserInfoFromFile PDF extract: ' . $e->getMessage());
|
||
return '';
|
||
}
|
||
}
|
||
|
||
private function extractDocxText($filePath)
|
||
{
|
||
$text = $this->tryExtractDocxText($filePath);
|
||
if ($text === '') {
|
||
throw new Exception('Word 文档文本提取失败');
|
||
}
|
||
return $text;
|
||
}
|
||
|
||
private function tryExtractDocxText($filePath)
|
||
{
|
||
if (!function_exists('docxReader')) {
|
||
return '';
|
||
}
|
||
try {
|
||
$text = docxReader($filePath);
|
||
return is_string($text) ? $text : '';
|
||
} catch (\Throwable $e) {
|
||
\think\Log::warning('UserInfoFromFile docxReader: ' . $e->getMessage());
|
||
return '';
|
||
}
|
||
}
|
||
|
||
private function looksLikeBinary($raw)
|
||
{
|
||
if (strpos($raw, "\0") !== false) {
|
||
return true;
|
||
}
|
||
$sample = substr($raw, 0, 4096);
|
||
if ($sample === '') {
|
||
return false;
|
||
}
|
||
$nonPrintable = preg_match_all('/[\x00-\x08\x0B\x0C\x0E-\x1F]/', $sample);
|
||
return $nonPrintable > 10;
|
||
}
|
||
|
||
private function parseUserInfoWithLlm($text)
|
||
{
|
||
$system = <<<'PROMPT'
|
||
你是学术人员信息抽取助手。根据用户上传的简历、简介或申请表文本，提取可用于期刊投稿/审稿人档案的基本信息。
|
||
|
||
只输出一个 JSON 对象，不要 markdown，不要解释。字段说明：
|
||
- realname: 中文姓名；若只有拼音可填拼音；无法判断则空字符串
|
||
- email: 邮箱
|
||
- phone: 手机或电话
|
||
- orcid: ORCID 号（仅数字与连字符）
|
||
- technical: 职称/职务（如教授、副主任护师、医学博士）
|
||
- field: 主要研究领域，中文，1～3 个词用顿号分隔
|
||
- company: 当前主要任职机构/工作单位（必填优先提取；填完整机构名，如「中南大学湘雅二医院」「暨南大学」，不要只写科室；多机构取最主要或最近一条）
|
||
- department: 所在科室/部门/学院（如感染科、药学院）
|
||
- country: 国家，默认中国可填「中国」
|
||
- introduction: 100字以内中文简介（职务+机构+方向）
|
||
|
||
缺失字段用空字符串 ""，不要编造。机构信息请从「工作单位」「任职」「所在单位」「affiliation」等段落提取。
|
||
PROMPT;
|
||
|
||
\think\Log::info('UserInfoFromFile user head: ' . mb_substr($text, 0, 600));
|
||
|
||
$raw = $this->llm->requestChat([
|
||
['role' => 'system', 'content' => $system],
|
||
['role' => 'user', 'content' => "请从以下文本提取用户信息：\n\n" . $text],
|
||
], 0.1);
|
||
|
||
if ($raw === null) {
|
||
throw new Exception('大模型请求失败，请检查 .env 中 [promotion] PROMOTION_LLM_URL / PROMOTION_LLM_MODEL（与参考文献校对相同）');
|
||
}
|
||
|
||
$parsed = $this->llm->parseJsonResponse($raw);
|
||
if ($parsed === null) {
|
||
throw new Exception('大模型返回无法解析为 JSON：' . mb_substr($raw, 0, 200));
|
||
}
|
||
|
||
return [
|
||
'user_info' => $this->normalizeUserInfo($parsed),
|
||
'raw' => $raw,
|
||
];
|
||
}
|
||
|
||
private function normalizeUserInfo(array $parsed)
|
||
{
|
||
$keys = ['realname', 'email', 'phone', 'orcid', 'technical', 'field', 'company', 'department', 'country', 'introduction'];
|
||
$out = [];
|
||
foreach ($keys as $k) {
|
||
$out[$k] = trim((string) ($parsed[$k] ?? ''));
|
||
}
|
||
|
||
// 机构：兼容大模型可能返回的别名字段
|
||
if ($out['company'] === '') {
|
||
foreach (['机构', '单位', 'institution', 'affiliation', 'organization', 'org', '工作单位', '任职单位'] as $alias) {
|
||
$val = trim((string) ($parsed[$alias] ?? ''));
|
||
if ($val !== '') {
|
||
$out['company'] = $val;
|
||
break;
|
||
}
|
||
}
|
||
}
|
||
|
||
if ($out['orcid'] !== '') {
|
||
$out['orcid'] = preg_replace('/\s+/', '', $out['orcid']);
|
||
}
|
||
|
||
// 对外同时返回 institution（与 company 同值）
|
||
$out['institution'] = $out['company'];
|
||
|
||
return $out;
|
||
}
|
||
|
||
private function sanitizeUtf8($text)
|
||
{
|
||
$text = (string) $text;
|
||
if (function_exists('mb_convert_encoding')) {
|
||
$text = mb_convert_encoding($text, 'UTF-8', 'UTF-8');
|
||
}
|
||
if (function_exists('iconv')) {
|
||
$fixed = @iconv('UTF-8', 'UTF-8//IGNORE', $text);
|
||
if (is_string($fixed)) {
|
||
$text = $fixed;
|
||
}
|
||
}
|
||
return $text;
|
||
}
|
||
}
|