Phase 16D.1 - Refine Major Character Discovery

This commit is contained in:
Nick Beckley 2026-06-27 20:16:42 +01:00
parent ac300a94c8
commit 7bee51834b
4 changed files with 219 additions and 15 deletions

View File

@ -338,6 +338,8 @@ public sealed class WordCompanionCharacterCandidateDto
{
public string Text { get; init; } = string.Empty;
public int MentionCount { get; init; }
public string Confidence { get; init; } = "Medium";
public string Reason { get; init; } = string.Empty;
}
public sealed class WordCompanionDiscoverCharactersRequest

View File

@ -35,19 +35,42 @@ public sealed class WordCompanionService(
private const int CharacterDiscoveryMinimumMentions = 5;
private const int CharacterDiscoverySuggestionLimit = 50;
private static readonly Regex CharacterCandidateRegex = new(
@"\b(?:(?:Mr|Mrs|Ms|Miss|Dr|Rev|Fr|Sir|Lady|Lord|DS|DI|DC|PC)\.?\s+)?[A-Z][a-z]+(?:[-'][A-Z]?[a-z]+)?(?:\s+[A-Z][a-z]+(?:[-'][A-Z]?[a-z]+)?){0,2}\b",
@"(?<!['\p{L}])(?:(?:Mr|Mrs|Ms|Miss|Dr|Rev|Fr|Sir|Lady|Lord|DS|DI|DC|PC|Professor|Aunt|Uncle)\.?\s+)?[A-Z][a-z]+(?:[-'][A-Z]?[a-z]+)?(?:\s+[A-Z][a-z]+(?:[-'][A-Z]?[a-z]+)?){0,2}(?!['\p{L}])",
RegexOptions.Compiled);
private static readonly HashSet<string> CharacterDiscoveryStopWords = new(StringComparer.OrdinalIgnoreCase)
{
"A", "An", "And", "As", "At", "But", "By", "For", "From", "He", "Her", "His", "I", "If", "In", "Into",
"It", "Its", "Of", "On", "Or", "Our", "She", "So", "That", "The", "Their", "Then", "There", "They",
"This", "To", "We", "When", "Where", "Who", "With", "You", "Your",
"Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "Saturday", "Sunday",
"January", "February", "March", "April", "May", "June", "July", "August", "September", "October", "November", "December",
"Christmas", "Easter", "Morning", "Afternoon", "Evening", "Night", "Today", "Tomorrow", "Yesterday",
"Chapter", "Scene", "Part", "Prologue", "Epilogue", "Book", "Volume", "Act",
"Mum", "Dad", "Mother", "Father", "Grandma", "Grandad", "Nan", "Nana", "Granny",
"Yes", "No", "Okay", "Ok", "Hello", "Goodbye"
"Yes", "No", "Okay", "Ok", "Hello", "Goodbye",
"What", "Not", "Just", "Yeah", "How", "Well", "Are", "Like", "One", "Can", "Thank", "Why", "Come",
"Let", "Maybe", "Please", "God", "Something", "After", "Did", "Even", "All", "Nothing", "Have",
"Don", "Won", "Could", "Would", "Should", "Do", "Does", "Had", "Has", "Was", "Were", "Is", "Am",
"Right", "Really", "Now", "Look", "Listen", "Sorry", "Thanks", "Fine", "Great", "Sure", "Course",
"Thing", "Things", "Everyone", "Everything", "Anything", "Someone", "Nobody", "People"
};
private static readonly HashSet<string> CharacterDiscoveryTemporalWords = new(StringComparer.OrdinalIgnoreCase)
{
"Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "Saturday", "Sunday",
"January", "February", "March", "April", "May", "June", "July", "August", "September", "October", "November", "December",
"Spring", "Summer", "Autumn", "Fall", "Winter"
};
private static readonly HashSet<string> CharacterDiscoveryNameLikeWords = new(StringComparer.OrdinalIgnoreCase)
{
"Grace", "Jack", "Rose", "May", "Will", "Mark", "Hope", "Faith", "Pat", "Rob", "Bill", "Harry",
"November", "Autumn", "Rain"
};
private static readonly HashSet<string> CharacterDiscoveryHonorifics = new(StringComparer.OrdinalIgnoreCase)
{
"Mr", "Mrs", "Ms", "Miss", "Dr", "Rev", "Fr", "Sir", "Lady", "Lord", "DS", "DI", "DC", "PC", "Professor", "Aunt", "Uncle"
};
private static readonly HashSet<string> CharacterDiscoveryDialogueTags = new(StringComparer.OrdinalIgnoreCase)
{
"said", "asked", "replied", "answered", "whispered", "shouted", "called", "muttered", "cried", "snapped",
"sighed", "laughed", "continued", "began", "told", "yelled"
};
public async Task<WordCompanionProjectsResponse> ListProjectsAsync()
{
@ -343,7 +366,7 @@ public sealed class WordCompanionService(
return [];
}
var mentions = new Dictionary<string, int>(StringComparer.Ordinal);
var candidates = new Dictionary<string, CharacterCandidateEvidence>(StringComparer.Ordinal);
foreach (Match match in CharacterCandidateRegex.Matches(manuscriptText))
{
var candidate = NormalizeCandidateName(match.Value);
@ -352,15 +375,33 @@ public sealed class WordCompanionService(
continue;
}
mentions[candidate] = mentions.TryGetValue(candidate, out var count) ? count + 1 : 1;
if (!candidates.TryGetValue(candidate, out var evidence))
{
evidence = new CharacterCandidateEvidence(candidate);
candidates[candidate] = evidence;
}
evidence.MentionCount += 1;
if (AppearsAtSentenceStart(manuscriptText, match.Index))
{
evidence.SentenceStartMentions += 1;
}
if (HasDialogueTagEvidence(manuscriptText, match.Index, match.Length))
{
evidence.DialogueEvidenceCount += 1;
}
}
return mentions
.Where(item => item.Value >= CharacterDiscoveryMinimumMentions)
.OrderByDescending(item => item.Value)
.ThenBy(item => item.Key, StringComparer.OrdinalIgnoreCase)
return candidates.Values
.Where(item => item.MentionCount >= CharacterDiscoveryMinimumMentions)
.Select(BuildCandidateDto)
.Where(item => item is not null)
.Select(item => item!)
.OrderByDescending(item => ConfidenceSortValue(item.Confidence))
.ThenByDescending(item => item.MentionCount)
.ThenBy(item => item.Text, StringComparer.OrdinalIgnoreCase)
.Take(CharacterDiscoverySuggestionLimit)
.Select(item => new WordCompanionCharacterCandidateDto { Text = item.Key, MentionCount = item.Value })
.ToList();
}
@ -381,9 +422,159 @@ public sealed class WordCompanionService(
return words.Length > 1 || first.Length > 2;
}
private static WordCompanionCharacterCandidateDto? BuildCandidateDto(CharacterCandidateEvidence evidence)
{
var score = 0;
var words = evidence.Text.Split(' ', StringSplitOptions.RemoveEmptyEntries);
var first = words.Length > 0 ? TrimHonorificPunctuation(words[0]) : string.Empty;
var hasHonorific = words.Length > 1 && CharacterDiscoveryHonorifics.Contains(first);
var hasMultiWordName = words.Length > 1;
var hasNameLikeWord = words.Any(word => CharacterDiscoveryNameLikeWords.Contains(TrimHonorificPunctuation(word)));
var hasTemporalWord = words.Any(word => CharacterDiscoveryTemporalWords.Contains(TrimHonorificPunctuation(word)));
var nonSentenceStartMentions = evidence.MentionCount - evidence.SentenceStartMentions;
var mostlySentenceStart = evidence.MentionCount > 0 && evidence.SentenceStartMentions >= Math.Ceiling(evidence.MentionCount * 0.8m);
if (evidence.MentionCount >= CharacterDiscoveryMinimumMentions)
{
score += 1;
}
if (evidence.MentionCount >= 10)
{
score += 1;
}
if (nonSentenceStartMentions > 0)
{
score += 2;
}
if (evidence.DialogueEvidenceCount > 0)
{
score += 3;
}
if (hasMultiWordName)
{
score += 2;
}
if (hasHonorific)
{
score += 2;
}
if (hasNameLikeWord)
{
score += 3;
}
if (hasTemporalWord)
{
score -= 1;
}
if (mostlySentenceStart)
{
score -= 2;
}
if (!hasMultiWordName && evidence.DialogueEvidenceCount == 0 && nonSentenceStartMentions == 0)
{
score -= 1;
}
if (score < 0)
{
return null;
}
var confidence = score >= 5 ? "High" : score >= 2 ? "Medium" : "Low";
return new WordCompanionCharacterCandidateDto
{
Text = evidence.Text,
MentionCount = evidence.MentionCount,
Confidence = confidence,
Reason = BuildCandidateReason(evidence, hasHonorific, hasMultiWordName, hasTemporalWord, mostlySentenceStart)
};
}
private static string BuildCandidateReason(
CharacterCandidateEvidence evidence,
bool hasHonorific,
bool hasMultiWordName,
bool hasTemporalWord,
bool mostlySentenceStart)
{
if (evidence.DialogueEvidenceCount > 0)
{
return "Dialogue evidence";
}
if (hasHonorific)
{
return "Title or honorific";
}
if (hasMultiWordName)
{
return "Proper-name phrase";
}
if (hasTemporalWord)
{
return "Possible name";
}
if (mostlySentenceStart)
{
return "Mostly sentence starts";
}
return evidence.MentionCount >= 10 ? "Frequent possible name" : "Possible name";
}
private static int ConfidenceSortValue(string confidence)
{
return string.Equals(confidence, "High", StringComparison.OrdinalIgnoreCase) ? 3
: string.Equals(confidence, "Medium", StringComparison.OrdinalIgnoreCase) ? 2
: 1;
}
private static bool AppearsAtSentenceStart(string manuscriptText, int matchIndex)
{
for (var i = matchIndex - 1; i >= 0; i--)
{
var character = manuscriptText[i];
if (char.IsWhiteSpace(character))
{
if (character is '\n' or '\r')
{
return true;
}
continue;
}
if (character is '"' or '\'' or '' or '“' or '”' or '(' or '[')
{
continue;
}
return character is '.' or '!' or '?' or ':' or ';';
}
return true;
}
private static bool HasDialogueTagEvidence(string manuscriptText, int matchIndex, int matchLength)
{
var before = manuscriptText[Math.Max(0, matchIndex - 48)..matchIndex];
var afterStart = matchIndex + matchLength;
var after = manuscriptText[afterStart..Math.Min(manuscriptText.Length, afterStart + 48)];
return HasTagAfterCandidate(after) || HasTagBeforeCandidate(before);
}
private static bool HasTagAfterCandidate(string textAfterCandidate)
{
var match = Regex.Match(textAfterCandidate, @"^\s*,?\s*(?<tag>[a-z]+)\b", RegexOptions.IgnoreCase);
return match.Success && CharacterDiscoveryDialogueTags.Contains(match.Groups["tag"].Value);
}
private static bool HasTagBeforeCandidate(string textBeforeCandidate)
{
var match = Regex.Match(textBeforeCandidate, @"\b(?<tag>[a-z]+)\s*,?\s*[""'\u2019\u201d]?\s*$", RegexOptions.IgnoreCase);
return match.Success && CharacterDiscoveryDialogueTags.Contains(match.Groups["tag"].Value);
}
private static string NormalizeCandidateName(string candidate)
{
return Regex.Replace(candidate ?? string.Empty, @"\s+", " ").Trim().Trim(',', '.', ';', ':', '!', '?', '"', '\'');
return Regex.Replace(candidate ?? string.Empty, @"\s+", " ").Trim().Trim(',', '.', ';', ':', '!', '?', '"', '\'', '');
}
private static string TrimHonorificPunctuation(string word)
@ -391,6 +582,14 @@ public sealed class WordCompanionService(
return (word ?? string.Empty).Trim().TrimEnd('.');
}
private sealed class CharacterCandidateEvidence(string text)
{
public string Text { get; } = text;
public int MentionCount { get; set; }
public int SentenceStartMentions { get; set; }
public int DialogueEvidenceCount { get; set; }
}
private static WordCompanionManuscriptDocumentDto ToDto(ManuscriptDocumentModel document) => new()
{
ManuscriptDocumentId = document.ManuscriptDocumentID,

View File

@ -1102,14 +1102,17 @@
const checkbox = document.createElement("input");
checkbox.type = "checkbox";
checkbox.value = candidate.text || "";
checkbox.checked = true;
checkbox.checked = String(candidate.confidence || "").toLowerCase() !== "very low";
checkbox.addEventListener("change", updateFirstRunButtons);
const name = document.createElement("strong");
name.textContent = candidate.text || "";
const mentions = document.createElement("em");
mentions.textContent = `${Number(candidate.mentionCount || 0).toLocaleString()} mentions`;
const reason = String(candidate.reason || "").trim();
mentions.textContent = reason
? `${Number(candidate.mentionCount || 0).toLocaleString()} mentions, ${reason.toLowerCase()}`
: `${Number(candidate.mentionCount || 0).toLocaleString()} mentions`;
label.append(checkbox, name, mentions);
firstRunCharacterCandidates.append(label);

File diff suppressed because one or more lines are too long