hamkadr/src/JobsMedical.Web/Services/ListingParser.cs

using System.Text.RegularExpressions;
using JobsMedical.Web.Models;

namespace JobsMedical.Web.Services;

/// <summary>Structured guess extracted from a raw channel post. All fields are best-effort.</summary>
public class ParsedListing
{
    public ListingKind Kind { get; set; } = ListingKind.Shift;
    public string? RoleName { get; set; }
    public ShiftType? ShiftType { get; set; }
    public EmploymentType? EmploymentType { get; set; }
    public long? PayAmount { get; set; }        // shift pay or single salary figure
    public int? SharePercent { get; set; }      // profit-share % (درصدی / سهم درآمد)
    public bool PayNegotiable { get; set; }
    public Gender Gender { get; set; } = Gender.Any;   // جنسیت مورد نیاز
    public string? CityName { get; set; }
    public string? DistrictName { get; set; }
    public string? FacilityName { get; set; }   // hospital/clinic name guessed from the text
    public string? Phone { get; set; }

    // «آماده به کار» (talent) extras — populated when Kind == Talent.
    public string? PersonName { get; set; }     // «دکتر سپیده علیزاده»
    public int? YearsExperience { get; set; }   // سابقه (سال)
    public bool IsLicensed { get; set; }        // پروانه‌دار
    public string? AreaNote { get; set; }       // «فقط منطقه ۱»
    public List<string> Notes { get; set; } = new();  // what was/wasn't detected (shown to admin)
}

/// <summary>
/// Turns a messy Persian channel/Divar post into a structured listing guess. This is the
/// Stage-1 implementation: transparent keyword + regex heuristics, no AI dependency (important
/// since LLM APIs are blocked from Iran). A future LlmListingParser can implement the same
/// interface and be swapped in via DI without touching the admin queue.
/// </summary>
public interface IListingParser
{
    ParsedListing Parse(string rawText, IEnumerable<string> knownRoles,
                        IEnumerable<string> knownCities, IEnumerable<string> knownDistricts);
}

public class HeuristicListingParser : IListingParser
{
    public ParsedListing Parse(string raw, IEnumerable<string> knownRoles,
        IEnumerable<string> knownCities, IEnumerable<string> knownDistricts)
    {
        var p = new ParsedListing();
        var text = Normalize(raw);

        // --- Kind: talent (worker offers themselves) vs shift vs hiring ---
        // Talent is checked first: «آماده به کار/همکاری», «جویای کار» mean the *person* is
        // available — distinct from an employer's «دعوت به همکاری».
        bool talentSignals = ContainsAny(text,
            "آماده به کار", "آماده‌به‌کار", "آماده همکاری", "آماده‌ی همکاری", "آماده ی همکاری",
            "آماده فعالیت", "جویای کار", "جویای کار هستم", "متقاضی کار", "نیازمند کار",
            "آماده انجام", "می‌توانم همکاری", "میتوانم همکاری", "حاضر به همکاری");
        bool jobSignals = ContainsAny(text, "استخدام", "جذب", "دعوت به همکاری", "نیازمندیم", "نیازمند است", "حقوق ثابت");
        bool shiftSignals = ContainsAny(text, "شیفت", "آنکال", "انکال", "نوبت", "کشیک");
        if (talentSignals)
        {
            p.Kind = ListingKind.Talent;
            p.Notes.Add("نوع: آماده به کار (تشخیص خودکار)");
        }
        else
        {
            p.Kind = (jobSignals && !shiftSignals) ? ListingKind.Job : ListingKind.Shift;
            p.Notes.Add(p.Kind == ListingKind.Job ? "نوع: استخدام (تشخیص خودکار)" : "نوع: شیفت (تشخیص خودکار)");
        }

        // --- Role (longest match first so «پزشک متخصص» beats «پزشک») ---
        foreach (var role in knownRoles.OrderByDescending(r => r.Length))
        {
            if (text.Contains(Normalize(role))) { p.RoleName = role; break; }
        }
        // Synonyms common on Divar/Medjobs → canonical seeded role names.
        if (p.RoleName is null)
        {
            p.RoleName =
                  ContainsAny(text, "اتاق عمل", "اسکراب") ? "تکنسین اتاق عمل"
                : ContainsAny(text, "فوریت", "اورژانس پیش بیمارستانی", "آمبولانس") ? "تکنسین فوریت‌های پزشکی"
                : ContainsAny(text, "آزمایشگاه", "علوم آزمایشگاهی", "نمونه گیر") ? "کارشناس آزمایشگاه"
                : ContainsAny(text, "بهیار", "کمک بهیار", "کمک پرستار", "بیماربر", "مراقب", "سالمند", "همراه بیمار", "تزریقات", "پانسمان") ? "پرستار"
                : ContainsAny(text, "ماما", "مامایی") ? "ماما"
                : ContainsAny(text, "فوق تخصص", "متخصص") ? "پزشک متخصص"
                : ContainsAny(text, "پزشک", "دکتر", "طبیب") ? "پزشک عمومی"
                : null;
        }
        p.Notes.Add(p.RoleName is null ? "نقش: تشخیص داده نشد" : $"نقش: {p.RoleName}");

        // --- Shift type ---
        if (ContainsAny(text, "آنکال", "انکال")) p.ShiftType = Models.ShiftType.OnCall;
        else if (text.Contains("شب")) p.ShiftType = Models.ShiftType.Night;
        else if (text.Contains("عصر")) p.ShiftType = Models.ShiftType.Evening;
        else if (ContainsAny(text, "صبح", "روز")) p.ShiftType = Models.ShiftType.Day;

        // --- Employment type ---
        if (ContainsAny(text, "پاره وقت", "پاره‌وقت", "پارت تایم")) p.EmploymentType = Models.EmploymentType.PartTime;
        else if (text.Contains("طرح")) p.EmploymentType = Models.EmploymentType.Plan;
        else if (text.Contains("قرارداد")) p.EmploymentType = Models.EmploymentType.Contract;
        else if (ContainsAny(text, "تمام وقت", "تمام‌وقت")) p.EmploymentType = Models.EmploymentType.FullTime;

        // --- Gender requirement ---
        if (ContainsAny(text, "خانم", "خانوم", "بانو", "زن ", "مامای")) p.Gender = Gender.Female;
        else if (ContainsAny(text, "آقا", "اقا", "مرد ", "مرد،", "پسر")) p.Gender = Gender.Male;
        if (p.Gender != Gender.Any)
            p.Notes.Add($"جنسیت: {(p.Gender == Gender.Female ? "خانم" : "آقا")}");

        // --- City / district ---
        p.CityName = knownCities.FirstOrDefault(c => text.Contains(Normalize(c)));
        p.DistrictName = knownDistricts.OrderByDescending(d => d.Length)
            .FirstOrDefault(d => text.Contains(Normalize(d)));

        // --- Profit share (درصدی / سهم) ---
        var latinForShare = ToLatinDigits(text);
        var share = Regex.Match(latinForShare, @"(\d{1,3})\s*(?:٪|%|درصد)");
        if (!share.Success) share = Regex.Match(latinForShare, @"(?:٪|%)\s*(\d{1,3})");
        if (share.Success && int.TryParse(share.Groups[1].Value, out var pct) && pct is > 0 and <= 100)
        { p.SharePercent = pct; p.Notes.Add($"سهم درآمد: {pct}٪"); }
        else if (ContainsAny(text, "درصدی", "سهم درآمد", "شراکت", "پورسانت"))
        { p.Notes.Add("پرداخت درصدی/سهمی (درصد نامشخص)"); }

        // --- Fixed pay ---
        if (ContainsAny(text, "توافقی", "توافق")) { p.PayNegotiable = true; p.Notes.Add("حقوق: توافقی"); }
        else
        {
            var amount = ExtractAmount(text);
            if (amount is not null) { p.PayAmount = amount; p.Notes.Add($"حقوق تخمینی: {amount:#,0} تومان"); }
            else if (p.SharePercent is null) p.Notes.Add("حقوق: تشخیص داده نشد");
        }

        // --- Talent extras (only meaningful for «آماده به کار») ---
        if (p.Kind == ListingKind.Talent)
        {
            var latinT = ToLatinDigits(text);
            var exp = Regex.Match(latinT, @"سابقه[^\d]{0,8}(\d{1,2})\s*سال");
            if (!exp.Success) exp = Regex.Match(latinT, @"(\d{1,2})\s*سال\s*سابقه");
            if (exp.Success && int.TryParse(exp.Groups[1].Value, out var yrs) && yrs is > 0 and <= 60)
            { p.YearsExperience = yrs; p.Notes.Add($"سابقه: {yrs} سال"); }

            p.IsLicensed = ContainsAny(text, "پروانه دار", "پروانه‌دار", "دارای پروانه", "پروانه فعالیت", "پروانه طبابت");
            if (p.IsLicensed) p.Notes.Add("پروانه‌دار");

            p.PersonName = ExtractPersonName(text);
            if (p.PersonName is not null) p.Notes.Add($"نام: {p.PersonName}");

            var area = Regex.Match(text, @"منطقه\s*[۰-۹0-9]{1,2}");
            if (area.Success) { p.AreaNote = area.Value.Trim(); p.Notes.Add($"محدوده: {p.AreaNote}"); }
        }

        // --- Facility name (بیمارستان/درمانگاه/کلینیک ... + the distinctive name) ---
        if (p.Kind != ListingKind.Talent)
        {
            p.FacilityName = ExtractFacilityName(text);
            if (p.FacilityName is not null) p.Notes.Add($"مرکز: {p.FacilityName}");
        }

        // --- Phone (mobile preferred, landline as fallback) ---
        var latinPhone = ToLatinDigits(text);
        var mobile = Regex.Match(latinPhone, @"(?:\+?98|0)?9\d{9}");
        if (mobile.Success)
        {
            var d = Regex.Replace(mobile.Value, @"\D", "");
            if (d.StartsWith("98")) d = "0" + d[2..];
            if (d.Length == 10 && d.StartsWith("9")) d = "0" + d;
            p.Phone = d;
        }
        else
        {
            var land = Regex.Match(latinPhone, @"0\d{2,3}[\s-]?\d{7,8}");
            if (land.Success) p.Phone = Regex.Replace(land.Value, @"\D", "");
        }

        return p;
    }

    // Words that introduce a facility name, longest/most-specific first.
    private static readonly string[] FacilityKeywords =
    {
        "بیمارستان", "زایشگاه", "پلی کلینیک", "پلیکلینیک", "درمانگاه", "کلینیک",
        "مرکز درمانی", "مرکز جراحی", "مجتمع پزشکی", "مجتمع درمانی", "مرکز", "مجتمع",
        "آزمایشگاه", "مطب", "خانه سالمندان", "سرای سالمندان",
    };

    // Words that clearly aren't part of a facility's name — stop collecting here.
    private static readonly string[] NameStops =
    {
        "جهت", "برای", "به", "با", "در", "از", "که", "نیاز", "نیازمند", "استخدام", "جذب",
        "دعوت", "همکاری", "واقع", "آدرس", "تلفن", "شماره", "شیفت", "ساعت", "حقوق", "روز",
        "شب", "صبح", "عصر", "می", "ها", "این", "یک", "محترم",
    };

    /// <summary>Best-effort hospital/clinic name: a facility keyword plus up to three name words.</summary>
    private static string? ExtractFacilityName(string text)
    {
        foreach (var kw in FacilityKeywords)
        {
            var idx = text.IndexOf(kw, StringComparison.Ordinal);
            if (idx < 0) continue;
            var after = text[(idx + kw.Length)..];
            var words = after.Split(
                new[] { ' ', '\n', '\r', '\t', '،', ',', '.', '؛', ':', '(', ')', '-', '/', '«', '»', '"' },
                StringSplitOptions.RemoveEmptyEntries);
            var picked = new List<string>();
            foreach (var w in words)
            {
                if (NameStops.Contains(w)) break;
                if (Regex.IsMatch(w, @"\d")) break;        // numbers/phones aren't names
                if (w.Length == 1) break;                  // stray letters
                picked.Add(w);
                if (picked.Count >= 3) break;
            }
            if (picked.Count == 0) continue;               // bare keyword (e.g. just «بیمارستان») isn't useful
            return (kw + " " + string.Join(" ", picked)).Trim();
        }
        return null;
    }

    // Titles that introduce a person's name in «آماده به کار» posts.
    private static readonly string[] PersonTitles = { "دکتر", "خانم دکتر", "آقای دکتر", "مهندس", "سرکار خانم", "جناب آقای", "خانم", "آقای" };

    /// <summary>Best-effort person name: a title (دکتر/خانم/…) plus up to two following words.</summary>
    private static string? ExtractPersonName(string text)
    {
        foreach (var title in PersonTitles)
        {
            var idx = text.IndexOf(title, StringComparison.Ordinal);
            if (idx < 0) continue;
            var after = text[(idx + title.Length)..];
            var words = after.Split(
                new[] { ' ', '\n', '\r', '\t', '،', ',', '.', '؛', ':', '(', ')', '-', '/' },
                StringSplitOptions.RemoveEmptyEntries);
            var picked = new List<string>();
            foreach (var w in words)
            {
                if (NameStops.Contains(w)) break;
                if (Regex.IsMatch(w, @"[\d]")) break;
                if (w.Length == 1) break;
                picked.Add(w);
                if (picked.Count >= 2) break;
            }
            if (picked.Count == 0) continue;
            return (title + " " + string.Join(" ", picked)).Trim();
        }
        return null;
    }

    /// <summary>Pull a Toman figure out of free text, handling «میلیون» and Persian digits.</summary>
    private static long? ExtractAmount(string text)
    {
        var latin = ToLatinDigits(text);
        // e.g. "۲ میلیون" / "2.5 میلیون"
        var million = Regex.Match(latin, @"(\d+(?:[.,]\d+)?)\s*میلیون");
        if (million.Success && double.TryParse(million.Groups[1].Value.Replace(",", "."),
                System.Globalization.NumberStyles.Any, System.Globalization.CultureInfo.InvariantCulture, out var m))
            return (long)(m * 1_000_000);

        // Otherwise the largest plain number that looks like money (>= 6 digits after removing separators).
        long best = 0;
        foreach (Match num in Regex.Matches(latin, @"[\d٬,،.]{6,}"))
        {
            var digits = Regex.Replace(num.Value, @"[^\d]", "");
            if (digits.Length >= 6 && long.TryParse(digits, out var v) && v > best) best = v;
        }
        return best > 0 ? best : null;
    }

    private static string Normalize(string s) => s
        .Replace('ي', 'ی').Replace('ك', 'ک').Replace('‌', ' ').Trim();

    private static bool ContainsAny(string text, params string[] needles)
        => needles.Any(n => text.Contains(n));

    private static string ToLatinDigits(string s)
    {
        var chars = s.ToCharArray();
        for (var i = 0; i < chars.Length; i++)
        {
            if (chars[i] >= '۰' && chars[i] <= '۹') chars[i] = (char)('0' + (chars[i] - '۰'));
            else if (chars[i] >= '٠' && chars[i] <= '٩') chars[i] = (char)('0' + (chars[i] - '٠'));
        }
        return new string(chars);
    }
}