Merge pull request #2186 from mikiher/Fuzzy-Matching-Continued

Fuzzy matching continued
This commit is contained in:
advplyr 2023-10-08 10:39:11 -05:00 committed by GitHub
commit 5ad9f507ba
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

View File

@ -59,12 +59,17 @@ class BookFinder {
// Remove single quotes (i.e. "Ender's Game" becomes "Enders Game")
cleaned = cleaned.replace(/'/g, '')
return this.replaceAccentedChars(cleaned)
return this.replaceAccentedChars(cleaned).toLowerCase()
}
cleanAuthorForCompares(author) {
if (!author) return ''
return this.replaceAccentedChars(author)
let cleanAuthor = this.replaceAccentedChars(author).toLowerCase()
// separate initials
cleanAuthor = cleanAuthor.replace(/([a-z])\.([a-z])/g, '$1. $2')
// remove middle initials
cleanAuthor = cleanAuthor.replace(/(?<=\w\w)(\s+[a-z]\.?)+(?=\s+\w\w)/g, '')
return cleanAuthor
}
filterSearchResults(books, title, author, maxTitleDistance, maxAuthorDistance) {
@ -136,6 +141,10 @@ class BookFinder {
if (!booksFiltered.length && books.length) {
if (this.verbose) Logger.debug(`Search has ${books.length} matches, but no close title matches`)
}
booksFiltered.sort((a, b) => {
return a.totalDistance - b.totalDistance
})
return booksFiltered
}
@ -179,35 +188,152 @@ class BookFinder {
return books
}
addTitleCandidate(title, candidates) {
// Main variant
const cleanTitle = this.cleanTitleForCompares(title).trim()
if (!cleanTitle) return
candidates.add(cleanTitle)
static TitleCandidates = class {
let candidate = cleanTitle
constructor(bookFinder, cleanAuthor) {
this.bookFinder = bookFinder
this.candidates = new Set()
this.cleanAuthor = cleanAuthor
this.priorities = {}
this.positions = {}
}
// Remove subtitle
candidate = candidate.replace(/([,:;_]| by ).*/g, "").trim()
if (candidate)
candidates.add(candidate)
add(title, position = 0) {
// if title contains the author, remove it
if (this.cleanAuthor) {
const authorRe = new RegExp(`(^| | by |)${this.cleanAuthor}(?= |$)`, "g")
title = this.bookFinder.cleanAuthorForCompares(title).replace(authorRe, '').trim()
}
// Remove preceding/trailing numbers
candidate = candidate.replace(/^\d+ | \d+$/g, "").trim()
if (candidate)
candidates.add(candidate)
const titleTransformers = [
[/([,:;_]| by ).*/g, ''], // Remove subtitle
[/(^| )\d+k(bps)?( |$)/, ' '], // Remove bitrate
[/ (2nd|3rd|\d+th)\s+ed(\.|ition)?/g, ''], // Remove edition
[/(^| |\.)(m4b|m4a|mp3)( |$)/g, ''], // Remove file-type
[/ a novel.*$/g, ''], // Remove "a novel"
[/^\d+ | \d+$/g, ''], // Remove preceding/trailing numbers
]
// Remove bitrate
candidate = candidate.replace(/(^| )\d+k(bps)?( |$)/, " ").trim()
if (candidate)
candidates.add(candidate)
// Main variant
const cleanTitle = this.bookFinder.cleanTitleForCompares(title).trim()
if (!cleanTitle) return
this.candidates.add(cleanTitle)
this.priorities[cleanTitle] = 0
this.positions[cleanTitle] = position
// Remove edition
candidate = candidate.replace(/ (2nd|3rd|\d+th)\s+ed(\.|ition)?/, "").trim()
if (candidate)
candidates.add(candidate)
let candidate = cleanTitle
for (const transformer of titleTransformers)
candidate = candidate.replace(transformer[0], transformer[1]).trim()
if (candidate != cleanTitle) {
if (candidate) {
this.candidates.add(candidate)
this.priorities[candidate] = 0
this.positions[candidate] = position
}
this.priorities[cleanTitle] = 1
}
}
get size() {
return this.candidates.size
}
getCandidates() {
var candidates = [...this.candidates]
candidates.sort((a, b) => {
// Candidates that include the author are likely low quality
const includesAuthorDiff = !b.includes(this.cleanAuthor) - !a.includes(this.cleanAuthor)
if (includesAuthorDiff) return includesAuthorDiff
// Candidates that include only digits are also likely low quality
const onlyDigits = /^\d+$/
const includesOnlyDigitsDiff = !onlyDigits.test(b) - !onlyDigits.test(a)
if (includesOnlyDigitsDiff) return includesOnlyDigitsDiff
// transformed candidates receive higher priority
const priorityDiff = this.priorities[a] - this.priorities[b]
if (priorityDiff) return priorityDiff
// if same priorirty, prefer candidates that are closer to the beginning (e.g. titles before subtitles)
const positionDiff = this.positions[a] - this.positions[b]
if (positionDiff) return positionDiff
// Start with longer candidaets, as they are likely more specific
const lengthDiff = b.length - a.length
if (lengthDiff) return lengthDiff
return b.localeCompare(a)
})
Logger.debug(`[${this.constructor.name}] Found ${candidates.length} fuzzy title candidates`)
Logger.debug(candidates)
return candidates
}
delete(title) {
return this.candidates.delete(title)
}
}
static AuthorCandidates = class {
constructor(bookFinder, cleanAuthor) {
this.bookFinder = bookFinder
this.candidates = new Set()
this.cleanAuthor = cleanAuthor
if (cleanAuthor) this.candidates.add(cleanAuthor)
}
validateAuthor(name, region = '', maxLevenshtein = 2) {
return this.bookFinder.audnexus.authorASINsRequest(name, region).then((asins) => {
for (const [i, asin] of asins.entries()) {
if (i > 10) break
let cleanName = this.bookFinder.cleanAuthorForCompares(asin.name)
if (!cleanName) continue
if (cleanName.includes(name)) return name
if (name.includes(cleanName)) return cleanName
if (levenshteinDistance(cleanName, name) <= maxLevenshtein) return cleanName
}
return ''
})
}
add(author) {
const cleanAuthor = this.bookFinder.cleanAuthorForCompares(author).trim()
if (!cleanAuthor) return
this.candidates.add(cleanAuthor)
}
get size() {
return this.candidates.size
}
get agressivelyCleanAuthor() {
if (this.cleanAuthor) {
const agressivelyCleanAuthor = this.cleanAuthor.replace(/[,/-].*$/, '').trim()
return agressivelyCleanAuthor ? agressivelyCleanAuthor : this.cleanAuthor
}
return ''
}
async getCandidates() {
var filteredCandidates = []
var promises = []
for (const candidate of this.candidates) {
promises.push(this.validateAuthor(candidate))
}
const results = [...new Set(await Promise.all(promises))]
filteredCandidates = results.filter(author => author)
// If no valid candidates were found, add back an aggresively cleaned author version
if (!filteredCandidates.length && this.cleanAuthor) filteredCandidates.push(this.agressivelyCleanAuthor)
// Always add an empty author candidate
filteredCandidates.push('')
Logger.debug(`[${this.constructor.name}] Found ${filteredCandidates.length} fuzzy author candidates`)
Logger.debug(filteredCandidates)
return filteredCandidates
}
delete(author) {
return this.candidates.delete(author)
}
}
/**
* Search for books including fuzzy searches
*
@ -232,62 +358,36 @@ class BookFinder {
books = await this.runSearch(title, author, provider, asin, maxTitleDistance, maxAuthorDistance)
if (!books.length && maxFuzzySearches > 0) {
// normalize title and author
// Normalize title and author
title = title.trim().toLowerCase()
author = author?.trim().toLowerCase() || ''
// Now run up to maxFuzzySearches fuzzy searches
let candidates = new Set()
let cleanedAuthor = this.cleanAuthorForCompares(author)
this.addTitleCandidate(title, candidates)
const cleanAuthor = this.cleanAuthorForCompares(author)
// remove parentheses and their contents, and replace with a separator
const cleanTitle = title.replace(/\[.*?\]|\(.*?\)|{.*?}/g, " - ")
// Now run up to maxFuzzySearches fuzzy searches
let authorCandidates = new BookFinder.AuthorCandidates(this, cleanAuthor)
// Remove underscores and parentheses with their contents, and replace with a separator
const cleanTitle = title.replace(/\[.*?\]|\(.*?\)|{.*?}|_/g, " - ")
// Split title into hypen-separated parts
const titleParts = cleanTitle.split(/ - | -|- /)
for (const titlePart of titleParts) {
this.addTitleCandidate(titlePart, candidates)
}
// We already searched for original title
if (author == cleanedAuthor) candidates.delete(title)
if (candidates.size > 0) {
candidates = [...candidates]
candidates.sort((a, b) => {
// Candidates that include the author are likely low quality
const includesAuthorDiff = !b.includes(cleanedAuthor) - !a.includes(cleanedAuthor)
if (includesAuthorDiff) return includesAuthorDiff
// Candidates that include only digits are also likely low quality
const onlyDigits = /^\d+$/
const includesOnlyDigitsDiff = !onlyDigits.test(b) - !onlyDigits.test(a)
if (includesOnlyDigitsDiff) return includesOnlyDigitsDiff
// Start with longer candidaets, as they are likely more specific
const lengthDiff = b.length - a.length
if (lengthDiff) return lengthDiff
return b.localeCompare(a)
})
Logger.debug(`[BookFinder] Found ${candidates.length} fuzzy title candidates`, candidates)
for (const candidate of candidates) {
for (const titlePart of titleParts)
authorCandidates.add(titlePart)
authorCandidates = await authorCandidates.getCandidates()
for (const authorCandidate of authorCandidates) {
let titleCandidates = new BookFinder.TitleCandidates(this, authorCandidate)
for (const [position, titlePart] of titleParts.entries())
titleCandidates.add(titlePart, position)
titleCandidates = titleCandidates.getCandidates()
for (const titleCandidate of titleCandidates) {
if (titleCandidate == title && authorCandidate == author) continue // We already tried this
if (++numFuzzySearches > maxFuzzySearches) return books
books = await this.runSearch(candidate, cleanedAuthor, provider, asin, maxTitleDistance, maxAuthorDistance)
if (books.length) break
}
if (!books.length) {
// Now try searching without the author
for (const candidate of candidates) {
if (++numFuzzySearches > maxFuzzySearches) return books
books = await this.runSearch(candidate, '', provider, asin, maxTitleDistance, maxAuthorDistance)
if (books.length) break
}
books = await this.runSearch(titleCandidate, authorCandidate, provider, asin, maxTitleDistance, maxAuthorDistance)
if (books.length) return books
}
}
}
if (provider === 'openlibrary') {
books.sort((a, b) => {
return a.totalDistance - b.totalDistance
})
}
return books
}