Match crawled listings to existing facilities (fuzzy) before creating new

When publishing a scraped listing we now look for a facility we already
have that is exactly or closely the same, and only create a new one when
there is no match — avoiding duplicates like «بیمارستان میلاد» vs «میلاد».

- ListingParser: extract a facility name (keyword + distinctive words) from
  the post and surface it in the parser notes.
- FacilityMatcher: Persian-aware normalization (ي/ك, ZWNJ, punctuation),
  type-word stripping for a "core" name, contains + Levenshtein similarity,
  and FindBest (same-city exact → any-city exact → same-city fuzzy → fuzzy).
- Review (manual publish): auto-select a matching facility or prefill the
  new-facility name; resolve-or-create uses fuzzy match; dropdown preselects.
- IngestionService (auto-publish): reuse FacilityMatcher against a run-wide
  facility list (grows as new ones are created) instead of exact-name only.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

src/JobsMedical.Web/Pages/Admin/Review.cshtml

@@ -51,10 +51,10 @@
                 <div class="filter-group">
                     <label>مرکز درمانی</label>
                     <select name="FacilityId">
-                        <option value="0">— انتخاب نشده —</option>
+                        <option value="0" selected="@(Model.FacilityId == 0)">— انتخاب نشده —</option>
                         @foreach (var f in Model.Facilities)
                         {
-                            <option value="@f.Id">@f.Name — @f.City?.Name</option>
+                            <option value="@f.Id" selected="@(Model.FacilityId == f.Id)">@f.Name — @f.City?.Name</option>
                         }
                     </select>
                     <input type="text" name="NewFacilityName" placeholder="یا نام مرکز جدید را وارد کن…" style="margin-top:6px;" />

src/JobsMedical.Web/Pages/Admin/Review.cshtml.cs

@@ -1,6 +1,7 @@
 using JobsMedical.Web.Data;
 using JobsMedical.Web.Models;
 using JobsMedical.Web.Services;
+using JobsMedical.Web.Services.Scraping;
 using Microsoft.AspNetCore.Authorization;
 using Microsoft.AspNetCore.Mvc;
 using Microsoft.AspNetCore.Mvc.RazorPages;
@@ -72,6 +73,25 @@ public class ReviewModel : PageModel
         if (Parsed.PayAmount is not null) { PayAmount = Parsed.PayAmount; SalaryMin = Parsed.PayAmount; }
         Description = Raw.RawText;
         Title = Parsed.RoleName is not null ? $"استخدام {Parsed.RoleName}" : "موقعیت استخدامی";
+
+        // Facility: try to match the listing's facility to one we already have; otherwise
+        // prefill the "new facility" box so publishing creates it.
+        if (!string.IsNullOrWhiteSpace(Parsed.FacilityName))
+        {
+            var cityId = await _db.Cities.Where(c => c.Name == Parsed.CityName)
+                .Select(c => (int?)c.Id).FirstOrDefaultAsync();
+            var match = FacilityMatcher.FindBest(Facilities, Parsed.FacilityName, cityId);
+            if (match is not null)
+            {
+                FacilityId = match.Id;
+                Parsed.Notes.Add($"مرکز منطبق در سیستم: «{match.Name}» — همین انتخاب شد.");
+            }
+            else
+            {
+                NewFacilityName = Parsed.FacilityName;
+                Parsed.Notes.Add($"مرکز جدید پیشنهادی: «{Parsed.FacilityName}» — هنگام انتشار ساخته می‌شود.");
+            }
+        }
         return Page();
     }
 
@@ -181,15 +201,17 @@ public class ReviewModel : PageModel
         if (string.IsNullOrWhiteSpace(NewFacilityName))
             return null;
 
-        // Reuse a same-named facility if one already exists, else create it.
         var name = NewFacilityName.Trim();
-        var existing = await _db.Facilities.FirstOrDefaultAsync(f => f.Name == name);
-        if (existing is not null) return existing.Id;
-
         var cityId = await _db.Cities.OrderByDescending(c => c.IsActive)
             .Select(c => (int?)c.Id).FirstOrDefaultAsync();
         if (cityId is null) return null;   // no cities seeded — cannot create a facility
 
+        // Reuse an existing facility that's exactly or closely the same (Persian-aware fuzzy
+        // match), so we don't create duplicates like «بیمارستان میلاد» vs «میلاد».
+        var all = await _db.Facilities.ToListAsync();
+        var match = FacilityMatcher.FindBest(all, name, cityId);
+        if (match is not null) return match.Id;
+
         var facility = new Facility
         {
             Name = name,

src/JobsMedical.Web/Services/ListingParser.cs

@@ -16,6 +16,7 @@ public class ParsedListing
     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; }
     public List<string> Notes { get; set; } = new();  // what was/wasn't detected (shown to admin)
 }
@@ -107,6 +108,10 @@ public class HeuristicListingParser : IListingParser
             else if (p.SharePercent is null) p.Notes.Add("حقوق: تشخیص داده نشد");
         }
 
+        // --- Facility name (بیمارستان/درمانگاه/کلینیک ... + the distinctive name) ---
+        p.FacilityName = ExtractFacilityName(text);
+        if (p.FacilityName is not null) p.Notes.Add($"مرکز: {p.FacilityName}");
+
         // --- Phone ---
         var phone = Regex.Match(ToLatinDigits(text), @"0?9\d{9}");
         if (phone.Success) p.Phone = phone.Value;
@@ -114,6 +119,48 @@ public class HeuristicListingParser : IListingParser
         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;
+    }
+
     /// <summary>Pull a Toman figure out of free text, handling «میلیون» and Persian digits.</summary>
     private static long? ExtractAmount(string text)
     {

src/JobsMedical.Web/Services/Scraping/FacilityMatcher.cs

@@ -0,0 +1,109 @@
+using System.Text;
+using System.Text.RegularExpressions;
+using JobsMedical.Web.Models;
+
+namespace JobsMedical.Web.Services.Scraping;
+
+/// <summary>
+/// Persian-aware fuzzy matching for facility names, so the same hospital written slightly
+/// differently — spacing, ي/ك vs ی/ک, ZWNJ, with or without «بیمارستان» — resolves to one
+/// record instead of creating a duplicate. Used by both the manual review/publish flow and
+/// the auto-publish ingestion pipeline.
+/// </summary>
+public static class FacilityMatcher
+{
+    // Generic type words stripped to compare the distinctive core of a name.
+    private static readonly string[] TypeWords =
+    {
+        "بیمارستان", "زایشگاه", "پلی کلینیک", "پلیکلینیک", "درمانگاه", "کلینیک",
+        "مرکز درمانی", "مرکز جراحی", "مجتمع پزشکی", "مجتمع درمانی", "مرکز", "مجتمع",
+        "آزمایشگاه", "مطب", "تخصصی", "فوق تخصصی", "فوقتخصصی", "عمومی", "دکتر", "دی کلینیک",
+    };
+
+    /// <summary>Lower-cased, Arabic→Persian folded, punctuation-stripped, whitespace-collapsed.</summary>
+    public static string Normalize(string? s)
+    {
+        if (string.IsNullOrWhiteSpace(s)) return "";
+        var t = s.Replace('ي', 'ی').Replace('ك', 'ک').Replace('ۀ', 'ه').Replace('ة', 'ه')
+                 .Replace('أ', 'ا').Replace('إ', 'ا').Replace('آ', 'ا').Replace('ئ', 'ی')
+                 .Replace('‌', ' ').ToLowerInvariant();
+        var sb = new StringBuilder(t.Length);
+        foreach (var ch in t)
+            sb.Append(char.IsLetterOrDigit(ch) || ch == ' ' ? ch : ' ');
+        return Regex.Replace(sb.ToString(), @"\s+", " ").Trim();
+    }
+
+    /// <summary>Normalized name with generic type words removed — the distinctive part.</summary>
+    public static string Core(string? s)
+    {
+        var n = Normalize(s);
+        if (n.Length == 0) return "";
+        foreach (var w in TypeWords)
+        {
+            var nw = Normalize(w);
+            if (nw.Length == 0) continue;
+            n = Regex.Replace(n, $@"(?<![\p{{L}}\p{{N}}]){Regex.Escape(nw)}(?![\p{{L}}\p{{N}}])", " ");
+        }
+        return Regex.Replace(n, @"\s+", " ").Trim();
+    }
+
+    /// <summary>True when two names almost certainly denote the same facility.</summary>
+    public static bool IsSame(string? a, string? b)
+    {
+        var na = Normalize(a);
+        var nb = Normalize(b);
+        if (na.Length == 0 || nb.Length == 0) return false;
+        if (na == nb) return true;
+
+        var ca = Core(a);
+        var cb = Core(b);
+        if (ca.Length >= 2 && ca == cb) return true;
+        // one core fully contains the other (e.g. «میلاد» vs «میلاد ۱»)
+        if (ca.Length >= 3 && cb.Length >= 3 && (ca.Contains(cb) || cb.Contains(ca))) return true;
+
+        // edit-distance similarity on the most informative basis
+        var (x, y) = ca.Length >= 3 && cb.Length >= 3 ? (ca, cb) : (na, nb);
+        return Similarity(x, y) >= 0.86;
+    }
+
+    /// <summary>
+    /// Best existing facility for <paramref name="name"/>: same-city exact match first, then
+    /// any-city exact, then same-city fuzzy, then any-city fuzzy. Null when nothing matches.
+    /// </summary>
+    public static Facility? FindBest(IEnumerable<Facility> facilities, string? name, int? cityId)
+    {
+        if (string.IsNullOrWhiteSpace(name)) return null;
+        var list = facilities as IList<Facility> ?? facilities.ToList();
+        var target = Normalize(name);
+
+        return list.FirstOrDefault(f => cityId.HasValue && f.CityId == cityId && Normalize(f.Name) == target)
+            ?? list.FirstOrDefault(f => Normalize(f.Name) == target)
+            ?? list.FirstOrDefault(f => cityId.HasValue && f.CityId == cityId && IsSame(f.Name, name))
+            ?? list.FirstOrDefault(f => IsSame(f.Name, name));
+    }
+
+    private static double Similarity(string a, string b)
+    {
+        if (a == b) return 1;
+        var max = Math.Max(a.Length, b.Length);
+        return max == 0 ? 1 : 1.0 - (double)Levenshtein(a, b) / max;
+    }
+
+    private static int Levenshtein(string a, string b)
+    {
+        var dp = new int[b.Length + 1];
+        for (var j = 0; j <= b.Length; j++) dp[j] = j;
+        for (var i = 1; i <= a.Length; i++)
+        {
+            var prev = dp[0];
+            dp[0] = i;
+            for (var j = 1; j <= b.Length; j++)
+            {
+                var tmp = dp[j];
+                dp[j] = Math.Min(Math.Min(dp[j] + 1, dp[j - 1] + 1), prev + (a[i - 1] == b[j - 1] ? 0 : 1));
+                prev = tmp;
+            }
+        }
+        return dp[b.Length];
+    }
+}

src/JobsMedical.Web/Services/Scraping/IngestionService.cs

@@ -52,6 +52,7 @@ public class IngestionService
         var roles = await _db.Roles.ToListAsync(ct);
         var cities = await _db.Cities.ToListAsync(ct);
         var districts = await _db.Districts.ToListAsync(ct);
+        var facilities = await _db.Facilities.ToListAsync(ct);   // fuzzy-matched + grown as we create
         var roleNames = roles.Select(r => r.Name).ToList();
         var cityNames = cities.Select(c => c.Name).ToList();
         var districtNames = districts.Select(d => d.Name).ToList();
@@ -95,7 +96,7 @@ public class IngestionService
 
                 if (status == RawListingStatus.Normalized)
                 {
-                    try { Publish(parsed, ai, raw, roles, cities, districts); published++; }
+                    try { Publish(parsed, ai, raw, roles, cities, districts, facilities); published++; }
                     catch (Exception ex) { _log.LogWarning(ex, "Auto-publish failed; queueing instead"); raw.Status = RawListingStatus.New; queued++; }
                 }
                 else if (status == RawListingStatus.New) queued++;
@@ -157,7 +158,7 @@ public class IngestionService
     }
 
     private void Publish(ParsedListing parsed, AiAuditResult? ai, RawListing raw,
-        List<Role> roles, List<City> cities, List<District> districts)
+        List<Role> roles, List<City> cities, List<District> districts, List<Facility> facilities)
     {
         var d = ai?.Data;
         var roleName = d?.Role ?? parsed.RoleName;
@@ -170,9 +171,10 @@ public class IngestionService
         var district = districts.FirstOrDefault(x => x.Name == districtName && x.CityId == city.Id);
 
         var facilityName = !string.IsNullOrWhiteSpace(d?.FacilityName) ? d!.FacilityName!.Trim()
+            : !string.IsNullOrWhiteSpace(parsed.FacilityName) ? parsed.FacilityName!.Trim()
             : $"مرکز درمانی (از {raw.SourceChannel})";
-        var facility = _db.Facilities.Local.FirstOrDefault(f => f.Name == facilityName && f.CityId == city.Id)
-                       ?? _db.Facilities.FirstOrDefault(f => f.Name == facilityName && f.CityId == city.Id);
+        // Reuse an existing facility (exact or Persian-aware fuzzy match) before creating a new one.
+        var facility = FacilityMatcher.FindBest(facilities, facilityName, city.Id);
         if (facility is null)
         {
             facility = new Facility
@@ -181,6 +183,7 @@ public class IngestionService
                 Phone = parsed.Phone, IsVerified = false,
             };
             _db.Facilities.Add(facility);
+            facilities.Add(facility);   // so later listings in this run match it too
         }
 
         var kind = (d?.Kind ?? parsed.Kind.ToString()).ToLowerInvariant();