X-Git-Url: https://git.mdrn.pl/wolnelektury.git/blobdiff_plain/ec93d5082ab137ec690f310eb6eba3c54d10b825..57e98a5807465f9494e5006e16860f16abbf4136:/apps/search/index.py diff --git a/apps/search/index.py b/apps/search/index.py index 25c16a96a..0f69d0725 100644 --- a/apps/search/index.py +++ b/apps/search/index.py @@ -6,10 +6,10 @@ from lucene import SimpleFSDirectory, IndexWriter, CheckIndex, \ NumericField, Version, Document, JavaError, IndexSearcher, \ QueryParser, PerFieldAnalyzerWrapper, \ SimpleAnalyzer, PolishAnalyzer, ArrayList, \ - KeywordAnalyzer, NumericRangeQuery, BooleanQuery, \ - BlockJoinQuery, BlockJoinCollector, TermsFilter, \ + KeywordAnalyzer, NumericRangeQuery, NumericRangeFilter, BooleanQuery, \ + BlockJoinQuery, BlockJoinCollector, Filter, TermsFilter, ChainedFilter, \ HashSet, BooleanClause, Term, CharTermAttribute, \ - PhraseQuery, MultiPhraseQuery, StringReader, TermQuery, BlockJoinQuery, \ + PhraseQuery, MultiPhraseQuery, StringReader, TermQuery, \ FuzzyQuery, FuzzyTermEnum, PrefixTermEnum, Sort, Integer, \ SimpleHTMLFormatter, Highlighter, QueryScorer, TokenSources, TextFragment, \ BooleanFilter, TermsFilter, FilterClause, QueryWrapperFilter, \ @@ -25,7 +25,8 @@ import re import errno from librarian import dcparser from librarian.parser import WLDocument -from catalogue import models +from lxml import etree +import catalogue.models from multiprocessing.pool import ThreadPool from threading import current_thread import atexit @@ -35,11 +36,9 @@ import traceback class WLAnalyzer(PerFieldAnalyzerWrapper): def __init__(self): polish = PolishAnalyzer(Version.LUCENE_34) - polish_gap = PolishAnalyzer(Version.LUCENE_34) # polish_gap.setPositionIncrementGap(999) simple = SimpleAnalyzer(Version.LUCENE_34) - simple_gap = SimpleAnalyzer(Version.LUCENE_34) # simple_gap.setPositionIncrementGap(999) keyword = KeywordAnalyzer(Version.LUCENE_34) @@ -48,21 +47,26 @@ class WLAnalyzer(PerFieldAnalyzerWrapper): PerFieldAnalyzerWrapper.__init__(self, polish) - self.addAnalyzer("tags", simple_gap) + self.addAnalyzer("tags", simple) self.addAnalyzer("technical_editors", simple) self.addAnalyzer("editors", simple) self.addAnalyzer("url", keyword) self.addAnalyzer("source_url", keyword) self.addAnalyzer("source_name", simple) self.addAnalyzer("publisher", simple) - self.addAnalyzer("author", simple) + self.addAnalyzer("authors", simple) + self.addAnalyzer("title", simple) + self.addAnalyzer("is_book", keyword) + # shouldn't the title have two forms? _pl and simple? + + self.addAnalyzer("themes", simple) + self.addAnalyzer("themes_pl", polish) - self.addAnalyzer("themes", simple_gap) - self.addAnalyzer("themes_pl", polish_gap) + self.addAnalyzer("tag_name", simple) + self.addAnalyzer("tag_name_pl", polish) - self.addAnalyzer("tag_name", simple_gap) - self.addAnalyzer("tag_name_pl", polish_gap) + self.addAnalyzer("translators", simple) self.addAnalyzer("KEYWORD", keyword) self.addAnalyzer("SIMPLE", simple) @@ -70,6 +74,11 @@ class WLAnalyzer(PerFieldAnalyzerWrapper): class IndexStore(object): + """ + Provides access to search index. + + self.store - lucene index directory + """ def __init__(self): self.make_index_dir() self.store = SimpleFSDirectory(File(settings.SEARCH_INDEX)) @@ -94,6 +103,11 @@ class IndexChecker(IndexStore): class Snippets(object): + """ + This class manages snippet files for indexed object (book) + the snippets are concatenated together, and their positions and + lengths are kept in lucene index fields. + """ SNIPPET_DIR = "snippets" def __init__(self, book_id): @@ -107,6 +121,9 @@ class Snippets(object): self.file = None def open(self, mode='r'): + """ + Open the snippet file. Call .close() afterwards. + """ if not 'b' in mode: mode += 'b' self.file = open(os.path.join(settings.SEARCH_INDEX, self.SNIPPET_DIR, str(self.book_id)), mode) @@ -114,23 +131,38 @@ class Snippets(object): return self def add(self, snippet): - l = len(snippet) - self.file.write(snippet.encode('utf-8')) + """ + Append a snippet (unicode) to the snippet file. + Return a (position, length) tuple + """ + txt = snippet.encode('utf-8') + l = len(txt) + self.file.write(txt) pos = (self.position, l) self.position += l return pos def get(self, pos): + """ + Given a tuple of (position, length) return an unicode + of the snippet stored there. + """ self.file.seek(pos[0], 0) - return self.read(pos[1]).decode('utf-8') + txt = self.file.read(pos[1]).decode('utf-8') + return txt def close(self): + """Close snippet file""" self.file.close() -class Index(IndexStore): +class BaseIndex(IndexStore): + """ + Base index class. + Provides basic operations on index: opening, closing, optimizing. + """ def __init__(self, analyzer=None): - IndexStore.__init__(self) + super(BaseIndex, self).__init__() self.index = None if not analyzer: analyzer = WLAnalyzer() @@ -155,39 +187,76 @@ class Index(IndexStore): self.index.close() self.index = None + def __enter__(self): + self.open() + return self + + def __exit__(self, type, value, tb): + self.close() + + +class Index(BaseIndex): + """ + Class indexing books. + """ + def __init__(self, analyzer=None): + super(Index, self).__init__(analyzer) + def index_tags(self): + """ + Re-index global tag list. + Removes all tags from index, then index them again. + Indexed fields include: id, name (with and without polish stems), category + """ q = NumericRangeQuery.newIntRange("tag_id", 0, Integer.MAX_VALUE, True, True) self.index.deleteDocuments(q) - for tag in models.Tag.objects.all(): + for tag in catalogue.models.Tag.objects.all(): doc = Document() - doc.add(NumericField("tag_id", Field.Store.YES, True).setIntValue(tag.id)) + doc.add(NumericField("tag_id", Field.Store.YES, True).setIntValue(int(tag.id))) doc.add(Field("tag_name", tag.name, Field.Store.NO, Field.Index.ANALYZED)) doc.add(Field("tag_name_pl", tag.name, Field.Store.NO, Field.Index.ANALYZED)) doc.add(Field("tag_category", tag.category, Field.Store.NO, Field.Index.NOT_ANALYZED)) self.index.addDocument(doc) + def create_book_doc(self, book): + """ + Create a lucene document referring book id. + """ + doc = Document() + doc.add(NumericField("book_id", Field.Store.YES, True).setIntValue(int(book.id))) + if book.parent is not None: + doc.add(NumericField("parent_id", Field.Store.YES, True).setIntValue(int(book.parent.id))) + return doc + def remove_book(self, book): + """Removes a book from search index. + book - Book instance.""" q = NumericRangeQuery.newIntRange("book_id", book.id, book.id, True, True) self.index.deleteDocuments(q) - def index_book(self, book, overwrite=True): + def index_book(self, book, book_info=None, overwrite=True): + """ + Indexes the book. + Creates a lucene document for extracted metadata + and calls self.index_content() to index the contents of the book. + """ if overwrite: self.remove_book(book) book_doc = self.create_book_doc(book) - meta_fields = self.extract_metadata(book) + meta_fields = self.extract_metadata(book, book_info) for f in meta_fields.values(): - if isinstance(f, list): + if isinstance(f, list) or isinstance(f, tuple): for elem in f: book_doc.add(elem) else: book_doc.add(f) - + self.index.addDocument(book_doc) del book_doc - self.index_content(book, book_fields=[meta_fields['title'], meta_fields['author']]) + self.index_content(book, book_fields=[meta_fields['title'], meta_fields['authors'], meta_fields['published_date']]) master_tags = [ 'opowiadanie', @@ -195,29 +264,33 @@ class Index(IndexStore): 'dramat_wierszowany_l', 'dramat_wierszowany_lp', 'dramat_wspolczesny', 'liryka_l', 'liryka_lp', - 'wywiad' + 'wywiad', + ] + + ignore_content_tags = [ + 'uwaga', 'extra', + 'zastepnik_tekstu', 'sekcja_asterysk', 'separator_linia', 'zastepnik_wersu', + 'didaskalia', + 'naglowek_aktu', 'naglowek_sceny', 'naglowek_czesc', ] - skip_header_tags = ['autor_utworu', 'nazwa_utworu', 'dzielo_nadrzedne'] + footnote_tags = ['pa', 'pt', 'pr', 'pe'] - def create_book_doc(self, book): + skip_header_tags = ['autor_utworu', 'nazwa_utworu', 'dzielo_nadrzedne', '{http://www.w3.org/1999/02/22-rdf-syntax-ns#}RDF'] + + published_date_re = re.compile("([0-9]+)[\]. ]*$") + + def extract_metadata(self, book, book_info=None): """ - Create a lucene document connected to the book + Extract metadata from book and returns a map of fields keyed by fieldname """ - doc = Document() - doc.add(NumericField("book_id", Field.Store.YES, True).setIntValue(book.id)) - if book.parent is not None: - doc.add(NumericField("parent_id", Field.Store.YES, True).setIntValue(book.parent.id)) - return doc - - def extract_metadata(self, book): fields = {} - book_info = dcparser.parse(book.xml_file) - print("extract metadata for book %s id=%d, thread%d" % (book.slug, book.id, current_thread().ident)) + if book_info is None: + book_info = dcparser.parse(open(book.xml_file.path)) fields['slug'] = Field("slug", book.slug, Field.Store.NO, Field.Index.ANALYZED_NO_NORMS) - fields['tags'] = [Field("tags", t.name, Field.Store.NO, Field.Index.ANALYZED) for t in book.tags] + fields['tags'] = self.add_gaps([Field("tags", t.name, Field.Store.NO, Field.Index.ANALYZED) for t in book.tags], 'tags') fields['is_book'] = Field("is_book", 'true', Field.Store.NO, Field.Index.NOT_ANALYZED) # validator, name @@ -246,31 +319,67 @@ class Index(IndexStore): dt = getattr(book_info, field.name) fields[field.name] = Field(field.name, "%04d%02d%02d" %\ (dt.year, dt.month, dt.day), Field.Store.NO, Field.Index.NOT_ANALYZED) + + # get published date + source = book_info.source_name + match = self.published_date_re.search(source) + print("published date is %s %s" % (match, match is not None and match.groups())) + if match is not None: + fields["published_date"] = Field("published_date", str(match.groups()[0]), Field.Store.YES, Field.Index.NOT_ANALYZED) + return fields + def add_gaps(self, fields, fieldname): + """ + Interposes a list of fields with gap-fields, which are indexed spaces and returns it. + This allows for doing phrase queries which do not overlap the gaps (when slop is 0). + """ + def gap(): + while True: + yield Field(fieldname, ' ', Field.Store.NO, Field.Index.NOT_ANALYZED) + return reduce(lambda a, b: a + b, zip(fields, gap()))[0:-1] + def get_master(self, root): + """ + Returns the first master tag from an etree. + """ for master in root.iter(): if master.tag in self.master_tags: return master def index_content(self, book, book_fields=[]): - wld = WLDocument.from_file(book.xml_file.path) + """ + Walks the book XML and extract content from it. + Adds parts for each header tag and for each fragment. + """ + wld = WLDocument.from_file(book.xml_file.path, parse_dublincore=False) root = wld.edoc.getroot() master = self.get_master(root) if master is None: return [] - def walker(node): + def walker(node, ignore_tags=[]): yield node, None - for child in list(node): + for child in filter(lambda n: n.tag not in ignore_tags, list(node)): for b, e in walker(child): yield b, e yield None, node return def fix_format(text): - return re.sub("/$", "", text, flags=re.M) + # separator = [u" ", u"\t", u".", u";", u","] + if isinstance(text, list): + # need to join it first + text = filter(lambda s: s is not None, content) + text = u' '.join(text) + # for i in range(len(text)): + # if i > 0: + # if text[i][0] not in separator\ + # and text[i - 1][-1] not in separator: + # text.insert(i, u" ") + + return re.sub("(?m)/$", "", text) def add_part(snippets, **fields): doc = self.create_book_doc(book) @@ -285,21 +394,36 @@ class Index(IndexStore): doc.add(Field('content', fields["content"], Field.Store.NO, Field.Index.ANALYZED, \ Field.TermVector.WITH_POSITIONS_OFFSETS)) - snip_pos = snippets.add(content) + snip_pos = snippets.add(fields["content"]) doc.add(NumericField("snippets_position", Field.Store.YES, True).setIntValue(snip_pos[0])) doc.add(NumericField("snippets_length", Field.Store.YES, True).setIntValue(snip_pos[1])) if 'fragment_anchor' in fields: doc.add(Field("fragment_anchor", fields['fragment_anchor'], Field.Store.YES, Field.Index.NOT_ANALYZED)) - + if 'themes' in fields: - for theme in fields['themes']: - doc.add(Field("themes", theme, Field.Store.YES, Field.Index.ANALYZED, Field.TermVector.WITH_POSITIONS)) - doc.add(Field("themes_pl", theme, Field.Store.NO, Field.Index.ANALYZED, Field.TermVector.WITH_POSITIONS)) + themes, themes_pl = zip(*[ + (Field("themes", theme, Field.Store.YES, Field.Index.ANALYZED, Field.TermVector.WITH_POSITIONS), + Field("themes_pl", theme, Field.Store.NO, Field.Index.ANALYZED, Field.TermVector.WITH_POSITIONS)) + for theme in fields['themes']]) + + themes = self.add_gaps(themes, 'themes') + themes_pl = self.add_gaps(themes_pl, 'themes_pl') + + for t in themes: + doc.add(t) + for t in themes_pl: + doc.add(t) return doc + def give_me_utf8(s): + if isinstance(s, unicode): + return s.encode('utf-8') + else: + return s + fragments = {} snippets = Snippets(book.id).open('w') try: @@ -307,60 +431,73 @@ class Index(IndexStore): if header.tag in self.skip_header_tags: continue + if header.tag is etree.Comment: + continue - content = u' '.join([t for t in header.itertext()]) - content = fix_format(content) + # section content + content = [] + footnote = None - doc = add_part(snippets, header_index=position, header_type=header.tag, content=content) + for start, end in walker(header, ignore_tags=self.ignore_content_tags): + # handle footnotes + # if start is not None and start.tag in self.footnote_tags: + # footnote = ' '.join(start.itertext()) + # elif end is not None and footnote is not None and end.tag in self.footnote_tags: + # doc = add_part(snippets, header_index=position, header_type=header.tag, + # content=footnote) - self.index.addDocument(doc) + # self.index.addDocument(doc) - for start, end in walker(master): + # footnote = None + + # handle fragments and themes. if start is not None and start.tag == 'begin': fid = start.attrib['id'][1:] fragments[fid] = {'content': [], 'themes': [], 'start_section': position, 'start_header': header.tag} - fragments[fid]['content'].append(start.tail) + elif start is not None and start.tag == 'motyw': fid = start.attrib['id'][1:] - fragments[fid]['themes'].append(start.text) - fragments[fid]['content'].append(start.tail) + if start.text is not None: + fragments[fid]['themes'] += map(str.strip, map(give_me_utf8, start.text.split(','))) + elif start is not None and start.tag == 'end': fid = start.attrib['id'][1:] if fid not in fragments: continue # a broken node, skip it + # import pdb; pdb.set_trace() frag = fragments[fid] + if frag['themes'] == []: + continue # empty themes list. del fragments[fid] - def jstr(l): - return u' '.join(map( - lambda x: x == None and u'(none)' or unicode(x), - l)) - doc = add_part(snippets, header_type=frag['start_header'], header_index=frag['start_section'], header_span=position - frag['start_section'] + 1, fragment_anchor=fid, - content=u' '.join(filter(lambda s: s is not None, frag['content'])), + content=fix_format(frag['content']), themes=frag['themes']) self.index.addDocument(doc) + + # Collect content. elif start is not None: for frag in fragments.values(): frag['content'].append(start.text) + content.append(start.text) elif end is not None: for frag in fragments.values(): frag['content'].append(end.tail) - finally: - snippets.close() + content.append(end.tail) + # in the end, add a section text. + doc = add_part(snippets, header_index=position, header_type=header.tag, + content=fix_format(content)) - def __enter__(self): - self.open() - return self + self.index.addDocument(doc) - def __exit__(self, type, value, tb): - self.close() + finally: + snippets.close() def log_exception_wrapper(f): @@ -383,35 +520,23 @@ class ReusableIndex(Index): if you cannot rely on atexit, use ReusableIndex.close_reusable() yourself. """ index = None - pool = None - pool_jobs = None def open(self, analyzer=None, threads=4): if ReusableIndex.index is not None: self.index = ReusableIndex.index else: print("opening index") - ReusableIndex.pool = ThreadPool(threads, initializer=lambda: JVM.attachCurrentThread() ) - ReusableIndex.pool_jobs = [] Index.open(self, analyzer) ReusableIndex.index = self.index atexit.register(ReusableIndex.close_reusable) - def index_book(self, *args, **kw): - job = ReusableIndex.pool.apply_async(log_exception_wrapper(Index.index_book), (self,) + args, kw) - ReusableIndex.pool_jobs.append(job) + # def index_book(self, *args, **kw): + # job = ReusableIndex.pool.apply_async(log_exception_wrapper(Index.index_book), (self,) + args, kw) + # ReusableIndex.pool_jobs.append(job) @staticmethod def close_reusable(): if ReusableIndex.index is not None: - print("wait for indexing to finish") - for job in ReusableIndex.pool_jobs: - job.get() - sys.stdout.write('.') - sys.stdout.flush() - print("done.") - ReusableIndex.pool.close() - ReusableIndex.index.optimize() ReusableIndex.index.close() ReusableIndex.index = None @@ -420,20 +545,13 @@ class ReusableIndex(Index): pass -class Search(IndexStore): - def __init__(self, default_field="content"): - IndexStore.__init__(self) - self.analyzer = WLAnalyzer() #PolishAnalyzer(Version.LUCENE_34) - ## self.analyzer = WLAnalyzer() - self.searcher = IndexSearcher(self.store, True) - self.parser = QueryParser(Version.LUCENE_34, default_field, - self.analyzer) - - self.parent_filter = TermsFilter() - self.parent_filter.addTerm(Term("is_book", "true")) - - def query(self, query): - return self.parser.parse(query) +class JoinSearch(object): + """ + This mixin could be used to handle block join queries. + (currently unused) + """ + def __init__(self, *args, **kw): + super(JoinSearch, self).__init__(*args, **kw) def wrapjoins(self, query, fields=[]): """ @@ -458,28 +576,6 @@ class Search(IndexStore): return BlockJoinQuery(query, self.parent_filter, BlockJoinQuery.ScoreMode.Total) - def simple_search(self, query, max_results=50): - """Returns (books, total_hits) - """ - - tops = self.searcher.search(self.query(query), max_results) - bks = [] - for found in tops.scoreDocs: - doc = self.searcher.doc(found.doc) - bks.append(models.Book.objects.get(id=doc.get("book_id"))) - return (bks, tops.totalHits) - - def search(self, query, max_results=50): - query = self.query(query) - query = self.wrapjoins(query, ["content", "themes"]) - - tops = self.searcher.search(query, max_results) - bks = [] - for found in tops.scoreDocs: - doc = self.searcher.doc(found.doc) - bks.append(models.Book.objects.get(id=doc.get("book_id"))) - return (bks, tops.totalHits) - def bsearch(self, query, max_results=50): q = self.query(query) bjq = BlockJoinQuery(q, self.parent_filter, BlockJoinQuery.ScoreMode.Avg) @@ -488,84 +584,154 @@ class Search(IndexStore): bks = [] for found in tops.scoreDocs: doc = self.searcher.doc(found.doc) - bks.append(models.Book.objects.get(id=doc.get("book_id"))) + bks.append(catalogue.models.Book.objects.get(id=doc.get("book_id"))) return (bks, tops.totalHits) -# TokenStream tokenStream = analyzer.tokenStream(fieldName, reader); -# OffsetAttribute offsetAttribute = tokenStream.getAttribute(OffsetAttribute.class); -# CharTermAttribute charTermAttribute = tokenStream.getAttribute(CharTermAttribute.class); - -# while (tokenStream.incrementToken()) { -# int startOffset = offsetAttribute.startOffset(); -# int endOffset = offsetAttribute.endOffset(); -# String term = charTermAttribute.toString(); -# } - class SearchResult(object): - def __init__(self, searcher, scoreDocs, score=None, how_found=None, snippets_cb=None): + def __init__(self, search, scoreDocs, score=None, how_found=None, snippets=None, searched=None, tokens_cache=None): + if tokens_cache is None: tokens_cache = {} + if score: - self.score = score + self._score = score else: - self.score = scoreDocs.score + self._score = scoreDocs.score + + self.boost = 1.0 - self.hits = [] + self._hits = [] + self._processed_hits = None # processed hits - stored = searcher.doc(scoreDocs.doc) + stored = search.searcher.doc(scoreDocs.doc) self.book_id = int(stored.get("book_id")) header_type = stored.get("header_type") + if not header_type: + return + sec = (header_type, int(stored.get("header_index"))) header_span = stored.get('header_span') header_span = header_span is not None and int(header_span) or 1 - stored = searcher.doc(scoreDocs.doc) - self.book_id = int(stored.get("book_id")) fragment = stored.get("fragment_anchor") - hit = (sec + (header_span,), fragment, scoreDocs.score, {'how_found': how_found, 'snippets_cb': snippets_cb}) + pd = stored.get("published_date") + if pd is None: + print "published_date is none for book %d" % self.book_id + pd = 0 + self.published_date = int(pd) + + if snippets: + snippets = snippets.replace("/\n", "\n") + hit = (sec + (header_span,), fragment, scoreDocs.score, {'how_found': how_found, 'snippets': snippets and [snippets] or []}) - self.hits.append(hit) + self._hits.append(hit) + + self.search = search + self.searched = searched + self.tokens_cache = tokens_cache + + @property + def score(self): + return self._score * self.boost def merge(self, other): if self.book_id != other.book_id: raise ValueError("this search result is or book %d; tried to merge with %d" % (self.book_id, other.book_id)) - self.hits += other.hits + self._hits += other._hits if other.score > self.score: - self.score = other.score - return self - - def add_snippets(self, snippets): - self.snippets += snippets + self._score = other._score return self def get_book(self): - return models.Book.objects.get(id=self.book_id) + return catalogue.models.Book.objects.get(id=self.book_id) book = property(get_book) - def get_parts(self): - book = self.book - - def sections_covered(results): - frags = filter(lambda r: r[1] is not None, results) - sect = filter(lambda r: r[1] is None, results) - sect = filter(lambda s: 0 == len(filter( - lambda f: s[0][1] >= f[0][1] and s[0][1] < f[0][1] + f[0][2], - frags)), sect) - print "filtered, non overlapped sections: %s" % sect - return frags + sect - - - parts = [{"header": s[0], "position": s[1], '_score_key': s} for s in self.sections] \ - + [{"fragment": book.fragments.get(anchor=f), '_score_key':f} for f in self.fragments] + @property + def hits(self): + if self._processed_hits is not None: + return self._processed_hits + + POSITION = 0 + FRAGMENT = 1 + POSITION_INDEX = 1 + POSITION_SPAN = 2 + SCORE = 2 + OTHER = 3 + + # to sections and fragments + frags = filter(lambda r: r[FRAGMENT] is not None, self._hits) + sect = filter(lambda r: r[FRAGMENT] is None, self._hits) + sect = filter(lambda s: 0 == len(filter( + lambda f: s[POSITION][POSITION_INDEX] >= f[POSITION][POSITION_INDEX] + and s[POSITION][POSITION_INDEX] < f[POSITION][POSITION_INDEX] + f[POSITION][POSITION_SPAN], + frags)), sect) + + hits = [] + + # remove duplicate fragments + fragments = {} + for f in frags: + fid = f[FRAGMENT] + if fid in fragments: + if fragments[fid][SCORE] >= f[SCORE]: + continue + fragments[fid] = f + frags = fragments.values() - parts.sort(lambda a, b: cmp(self.scores[a['_score_key']], self.scores[b['_score_key']])) - print("bookid: %d parts: %s" % (self.book_id, parts)) - return parts + # remove duplicate sections + sections = {} - parts = property(get_parts) + for s in sect: + si = s[POSITION][POSITION_INDEX] + # skip existing + if si in sections: + if sections[si]['score'] >= s[SCORE]: + continue + m = {'score': s[SCORE], + 'section_number': s[POSITION][POSITION_INDEX] + 1, + } + m.update(s[OTHER]) + sections[si] = m + + hits = sections.values() + + for f in frags: + try: + frag = catalogue.models.Fragment.objects.get(anchor=f[FRAGMENT]) + except catalogue.models.Fragment.DoesNotExist: + # stale index + continue + + # Figure out if we were searching for a token matching some word in theme name. + themes = frag.tags.filter(category='theme') + themes_hit = [] + if self.searched is not None: + tokens = self.search.get_tokens(self.searched, 'POLISH', cached=self.tokens_cache) + for theme in themes: + name_tokens = self.search.get_tokens(theme.name, 'POLISH') + for t in tokens: + if t in name_tokens: + if not theme in themes_hit: + themes_hit.append(theme) + break + + m = {'score': f[SCORE], + 'fragment': frag, + 'section_number': f[POSITION][POSITION_INDEX] + 1, + 'themes': themes, + 'themes_hit': themes_hit + } + m.update(f[OTHER]) + hits.append(m) + + hits.sort(lambda a, b: cmp(a['score'], b['score']), reverse=True) + + self._processed_hits = hits + + return hits def __unicode__(self): return u'SearchResult(book_id=%d, score=%d)' % (self.book_id, self.score) @@ -583,16 +749,149 @@ class SearchResult(object): return books.values() def __cmp__(self, other): - return cmp(self.score, other.score) + c = cmp(self.score, other.score) + if c == 0: + # this is inverted, because earlier date is better + return cmp(other.published_date, self.published_date) + else: + return c + +class Hint(object): + """ + Given some hint information (information we already know about) + our search target - like author, title (specific book), epoch, genre, kind + we can narrow down search using filters. + """ + def __init__(self, search): + """ + Accepts a Searcher instance. + """ + self.search = search + self.book_tags = {} + self.part_tags = [] + self._books = [] -class MultiSearch(Search): - """Class capable of IMDb-like searching""" - def get_tokens(self, searched, field='content'): + def books(self, *books): + """ + Give a hint that we search these books. + """ + self._books = books + + def tags(self, tags): + """ + Give a hint that these Tag objects (a list of) + is necessary. + """ + for t in tags: + if t.category in ['author', 'title', 'epoch', 'genre', 'kind']: + lst = self.book_tags.get(t.category, []) + lst.append(t) + self.book_tags[t.category] = lst + if t.category in ['theme', 'theme_pl']: + self.part_tags.append(t) + + def tag_filter(self, tags, field='tags'): + """ + Given a lsit of tags and an optional field (but they are normally in tags field) + returns a filter accepting only books with specific tags. + """ + q = BooleanQuery() + + for tag in tags: + toks = self.search.get_tokens(tag.name, field=field) + tag_phrase = PhraseQuery() + for tok in toks: + tag_phrase.add(Term(field, tok)) + q.add(BooleanClause(tag_phrase, BooleanClause.Occur.MUST)) + + return QueryWrapperFilter(q) + + def book_filter(self): + """ + Filters using book tags (all tag kinds except a theme) + """ + tags = reduce(lambda a, b: a + b, self.book_tags.values(), []) + if tags: + return self.tag_filter(tags) + else: + return None + + def part_filter(self): + """ + This filter can be used to look for book parts. + It filters on book id and/or themes. + """ + fs = [] + if self.part_tags: + fs.append(self.tag_filter(self.part_tags, field='themes')) + + if self._books != []: + bf = BooleanFilter() + for b in self._books: + id_filter = NumericRangeFilter.newIntRange('book_id', b.id, b.id, True, True) + bf.add(FilterClause(id_filter, BooleanClause.Occur.SHOULD)) + fs.append(bf) + + return Search.chain_filters(fs) + + def should_search_for_book(self): + return self._books == [] + + def just_search_in(self, all): + """Holds logic to figure out which indexes should be search, when we have some hinst already""" + some = [] + for field in all: + if field == 'authors' and 'author' in self.book_tags: + continue + if field == 'title' and self._books != []: + continue + if (field == 'themes' or field == 'themes_pl') and self.part_tags: + continue + some.append(field) + return some + + +class Search(IndexStore): + """ + Search facilities. + """ + def __init__(self, default_field="content"): + IndexStore.__init__(self) + self.analyzer = WLAnalyzer() # PolishAnalyzer(Version.LUCENE_34) + # self.analyzer = WLAnalyzer() + self.searcher = IndexSearcher(self.store, True) + self.parser = QueryParser(Version.LUCENE_34, default_field, + self.analyzer) + + self.parent_filter = TermsFilter() + self.parent_filter.addTerm(Term("is_book", "true")) + + def query(self, query): + """Parse query in default Lucene Syntax. (for humans) + """ + return self.parser.parse(query) + + def simple_search(self, query, max_results=50): + """Runs a query for books using lucene syntax. (for humans) + Returns (books, total_hits) + """ + + tops = self.searcher.search(self.query(query), max_results) + bks = [] + for found in tops.scoreDocs: + doc = self.searcher.doc(found.doc) + bks.append(catalogue.models.Book.objects.get(id=doc.get("book_id"))) + return (bks, tops.totalHits) + + def get_tokens(self, searched, field='content', cached=None): """returns tokens analyzed by a proper (for a field) analyzer argument can be: StringReader, string/unicode, or tokens. In the last case they will just be returned (so we can reuse tokens, if we don't change the analyzer) """ + if cached is not None and field in cached: + return cached[field] + if isinstance(searched, str) or isinstance(searched, unicode): searched = StringReader(searched) elif isinstance(searched, list): @@ -604,9 +903,14 @@ class MultiSearch(Search): while tokens.incrementToken(): cta = tokens.getAttribute(CharTermAttribute.class_) toks.append(cta.toString()) + + if cached is not None: + cached[field] = toks + return toks def fuzziness(self, fuzzy): + """Helper method to sanitize fuzziness""" if not fuzzy: return None if isinstance(fuzzy, float) and fuzzy > 0.0 and fuzzy <= 1.0: @@ -615,6 +919,9 @@ class MultiSearch(Search): return 0.5 def make_phrase(self, tokens, field='content', slop=2, fuzzy=False): + """ + Return a PhraseQuery with a series of tokens. + """ if fuzzy: phrase = MultiPhraseQuery() for t in tokens: @@ -641,6 +948,11 @@ class MultiSearch(Search): return phrase def make_term_query(self, tokens, field='content', modal=BooleanClause.Occur.SHOULD, fuzzy=False): + """ + Returns term queries joined by boolean query. + modal - applies to boolean query + fuzzy - should the query by fuzzy. + """ q = BooleanQuery() for t in tokens: term = Term(field, t) @@ -651,85 +963,178 @@ class MultiSearch(Search): q.add(BooleanClause(term, modal)) return q - def content_query(self, query): - return BlockJoinQuery(query, self.parent_filter, - BlockJoinQuery.ScoreMode.Total) + def search_phrase(self, searched, field, book=True, max_results=20, fuzzy=False, + filters=None, tokens_cache=None, boost=None, snippets=False, slop=2): + if filters is None: filters = [] + if tokens_cache is None: tokens_cache = {} + + tokens = self.get_tokens(searched, field, cached=tokens_cache) + + query = self.make_phrase(tokens, field=field, fuzzy=fuzzy, slop=slop) + if book: + filters.append(self.term_filter(Term('is_book', 'true'))) + top = self.searcher.search(query, self.chain_filters(filters), max_results) + + return [SearchResult(self, found, snippets=(snippets and self.get_snippets(found, query) or None), searched=searched) for found in top.scoreDocs] + + def search_some(self, searched, fields, book=True, max_results=20, fuzzy=False, + filters=None, tokens_cache=None, boost=None, snippets=True): + if filters is None: filters = [] + if tokens_cache is None: tokens_cache = {} - def search_perfect_book(self, searched, max_results=20, fuzzy=False): - qrys = [self.make_phrase(self.get_tokens(searched, field=fld), field=fld, fuzzy=fuzzy) for fld in ['author', 'title']] + if book: + filters.append(self.term_filter(Term('is_book', 'true'))) + + query = BooleanQuery() + + for fld in fields: + tokens = self.get_tokens(searched, fld, cached=tokens_cache) + + query.add(BooleanClause(self.make_term_query(tokens, field=fld, + fuzzy=fuzzy), BooleanClause.Occur.SHOULD)) + + top = self.searcher.search(query, self.chain_filters(filters), max_results) + + return [SearchResult(self, found, searched=searched, tokens_cache=tokens_cache, + snippets=(snippets and self.get_snippets(found, query) or None)) for found in top.scoreDocs] + + def search_perfect_book(self, searched, max_results=20, fuzzy=False, hint=None): + """ + Search for perfect book matches. Just see if the query matches with some author or title, + taking hints into account. + """ + fields_to_search = ['authors', 'title'] + only_in = None + if hint: + if not hint.should_search_for_book(): + return [] + fields_to_search = hint.just_search_in(fields_to_search) + only_in = hint.book_filter() + + qrys = [self.make_phrase(self.get_tokens(searched, field=fld), field=fld, fuzzy=fuzzy) for fld in fields_to_search] books = [] for q in qrys: - top = self.searcher.search(q, max_results) + top = self.searcher.search(q, + self.chain_filters([only_in, self.term_filter(Term('is_book', 'true'))]), + max_results) for found in top.scoreDocs: - books.append(SearchResult(self.searcher, found)) + books.append(SearchResult(self, found, how_found="search_perfect_book")) return books - def search_perfect_parts(self, searched, max_results=20, fuzzy=False): + def search_book(self, searched, max_results=20, fuzzy=False, hint=None): + fields_to_search = ['tags', 'authors', 'title'] + + only_in = None + if hint: + if not hint.should_search_for_book(): + return [] + fields_to_search = hint.just_search_in(fields_to_search) + only_in = hint.book_filter() + + tokens = self.get_tokens(searched, field='SIMPLE') + + q = BooleanQuery() + + for fld in fields_to_search: + q.add(BooleanClause(self.make_term_query(tokens, field=fld, + fuzzy=fuzzy), BooleanClause.Occur.SHOULD)) + + books = [] + top = self.searcher.search(q, + self.chain_filters([only_in, self.term_filter(Term('is_book', 'true'))]), + max_results) + for found in top.scoreDocs: + books.append(SearchResult(self, found, how_found="search_book")) + + return books + + def search_perfect_parts(self, searched, max_results=20, fuzzy=False, hint=None): + """ + Search for book parts which contains a phrase perfectly matching (with a slop of 2, default for make_phrase()) + some part/fragment of the book. + """ qrys = [self.make_phrase(self.get_tokens(searched), field=fld, fuzzy=fuzzy) for fld in ['content']] + flt = None + if hint: + flt = hint.part_filter() + books = [] for q in qrys: - top = self.searcher.search(q, max_results) + top = self.searcher.search(q, + self.chain_filters([self.term_filter(Term('is_book', 'true'), inverse=True), + flt]), + max_results) for found in top.scoreDocs: - books.append(SearchResult(self.searcher, found).add_snippets(self.get_snippets(found, q))) + books.append(SearchResult(self, found, snippets=self.get_snippets(found, q), how_found='search_perfect_parts')) return books - def search_everywhere(self, searched, max_results=20, fuzzy=False): + def search_everywhere(self, searched, max_results=20, fuzzy=False, hint=None, tokens_cache=None): + """ + Tries to use search terms to match different fields of book (or its parts). + E.g. one word can be an author survey, another be a part of the title, and the rest + are some words from third chapter. + """ + if tokens_cache is None: tokens_cache = {} books = [] + only_in = None + + if hint: + only_in = hint.part_filter() # content only query : themes x content q = BooleanQuery() - tokens = self.get_tokens(searched) - q.add(BooleanClause(self.make_term_query(tokens, field='themes', fuzzy=fuzzy), BooleanClause.Occur.MUST)) - q.add(BooleanClause(self.make_term_query(tokens, field='content', fuzzy=fuzzy), BooleanClause.Occur.SHOULD)) + tokens_pl = self.get_tokens(searched, field='content', cached=tokens_cache) + tokens = self.get_tokens(searched, field='SIMPLE', cached=tokens_cache) - topDocs = self.searcher.search(q, max_results) - for found in topDocs.scoreDocs: - books.append(SearchResult(self.searcher, found)) + # only search in themes when we do not already filter by themes + if hint is None or hint.just_search_in(['themes']) != []: + q.add(BooleanClause(self.make_term_query(tokens_pl, field='themes_pl', + fuzzy=fuzzy), BooleanClause.Occur.MUST)) + + q.add(BooleanClause(self.make_term_query(tokens_pl, field='content', + fuzzy=fuzzy), BooleanClause.Occur.SHOULD)) - # joined query themes/content x author/title/epochs/genres/kinds - # q = BooleanQuery() - # in_meta = BooleanQuery() - # in_content = BooleanQuery() + topDocs = self.searcher.search(q, only_in, max_results) + for found in topDocs.scoreDocs: + books.append(SearchResult(self, found, how_found='search_everywhere_themesXcontent', searched=searched)) + print "* %s theme x content: %s" % (searched, books[-1]._hits) - # for fld in ['themes', 'content']: - # in_content.add(BooleanClause(self.make_term_query(tokens, field=fld, fuzzy=False), BooleanClause.Occur.SHOULD)) + # query themes/content x author/title/tags + q = BooleanQuery() + in_content = BooleanQuery() + in_meta = BooleanQuery() - # in_meta.add(BooleanClause(self.make_term_query( - # self.get_tokens(searched, field='author'), field='author', fuzzy=fuzzy), BooleanClause.Occur.SHOULD)) + for fld in ['themes_pl', 'content']: + in_content.add(BooleanClause(self.make_term_query(tokens_pl, field=fld, fuzzy=False), BooleanClause.Occur.SHOULD)) - # for fld in ['title', 'epochs', 'genres', 'kinds']: - # in_meta.add(BooleanClause(self.make_term_query(tokens, field=fld, fuzzy=fuzzy), BooleanClause.Occur.SHOULD)) + for fld in ['tags', 'authors', 'title']: + in_meta.add(BooleanClause(self.make_term_query(tokens, field=fld, fuzzy=False), BooleanClause.Occur.SHOULD)) - # q.add(BooleanClause(in_meta, BooleanClause.Occur.MUST)) - # in_content_join = self.content_query(in_content) - # q.add(BooleanClause(in_content_join, BooleanClause.Occur.MUST)) - # # import pdb; pdb.set_trace() - # collector = BlockJoinCollector(Sort.RELEVANCE, 100, True, True) + q.add(BooleanClause(in_content, BooleanClause.Occur.MUST)) + q.add(BooleanClause(in_meta, BooleanClause.Occur.SHOULD)) - # self.searcher.search(q, collector) + topDocs = self.searcher.search(q, only_in, max_results) + for found in topDocs.scoreDocs: + books.append(SearchResult(self, found, how_found='search_everywhere', searched=searched)) + print "* %s scatter search: %s" % (searched, books[-1]._hits) - # top_groups = collector.getTopGroups(in_content_join, Sort.RELEVANCE, 0, max_results, 0, True) - # if top_groups: - # for grp in top_groups.groups: - # for part in grp.scoreDocs: - # books.append(SearchResult(self.searcher, part, score=grp.maxScore)) return books - def multisearch(self, query, max_results=50): - """ - Search strategy: - - (phrase) OR -> content - -> title - -> author - - (keywords) -> author - -> motyw - -> tags - -> content - """ + # def multisearch(self, query, max_results=50): + # """ + # Search strategy: + # - (phrase) OR -> content + # -> title + # -> authors + # - (keywords) -> authors + # -> motyw + # -> tags + # -> content + # """ # queryreader = StringReader(query) # tokens = self.get_tokens(queryreader) @@ -759,39 +1164,34 @@ class MultiSearch(Search): # top_level.add(BooleanClause(phrase_level, Should)) # top_level.add(BooleanClause(kw_level, Should)) - return None - - def book_search(self, query, filter=None, max_results=50, collector=None): - tops = self.searcher.search(query, filter, max_results) - #tops = self.searcher.search(p_content, max_results) - - bks = [] - for found in tops.scoreDocs: - doc = self.searcher.doc(found.doc) - b = models.Book.objects.get(id=doc.get("book_id")) - bks.append(b) - print "%s (%d) -> %f" % (b, b.id, found.score) - return bks + # return None def get_snippets(self, scoreDoc, query, field='content'): + """ + Returns a snippet for found scoreDoc. + """ htmlFormatter = SimpleHTMLFormatter() highlighter = Highlighter(htmlFormatter, QueryScorer(query)) stored = self.searcher.doc(scoreDoc.doc) + position = stored.get('snippets_position') + length = stored.get('snippets_length') + if position is None or length is None: + return None # locate content. snippets = Snippets(stored.get('book_id')).open() try: - text = snippets.get(stored.get('snippets_position'), stored.get('snippets_length')) + text = snippets.get((int(position), + int(length))) finally: snippets.close() tokenStream = TokenSources.getAnyTokenStream(self.searcher.getIndexReader(), scoreDoc.doc, field, self.analyzer) # highlighter.getBestTextFragments(tokenStream, text, False, 10) snip = highlighter.getBestFragments(tokenStream, text, 3, "...") - print('snips: %s' % snip) - return [snip] + return snip @staticmethod def enum_to_array(enum): @@ -811,23 +1211,37 @@ class MultiSearch(Search): return JArray('object')(terms, Term) def search_tags(self, query, filter=None, max_results=40): + """ + Search for Tag objects using query. + """ tops = self.searcher.search(query, filter, max_results) tags = [] for found in tops.scoreDocs: doc = self.searcher.doc(found.doc) - tag = models.Tag.objects.get(id=doc.get("tag_id")) + tag = catalogue.models.Tag.objects.get(id=doc.get("tag_id")) tags.append(tag) print "%s (%d) -> %f" % (tag, tag.id, found.score) return tags + def search_books(self, query, filter=None, max_results=10): + """ + Searches for Book objects using query + """ + bks = [] + tops = self.searcher.search(query, filter, max_results) + for found in tops.scoreDocs: + doc = self.searcher.doc(found.doc) + bks.append(catalogue.models.Book.objects.get(id=doc.get("book_id"))) + return bks + def create_prefix_phrase(self, toks, field): q = MultiPhraseQuery() for i in range(len(toks)): t = Term(field, toks[i]) if i == len(toks) - 1: - pterms = MultiSearch.enum_to_array(PrefixTermEnum(self.searcher.getIndexReader(), t)) + pterms = Search.enum_to_array(PrefixTermEnum(self.searcher.getIndexReader(), t)) if pterms: q.add(pterms) else: @@ -836,7 +1250,23 @@ class MultiSearch(Search): q.add(t) return q + @staticmethod + def term_filter(term, inverse=False): + only_term = TermsFilter() + only_term.addTerm(term) + + if inverse: + neg = BooleanFilter() + neg.add(FilterClause(only_term, BooleanClause.Occur.MUST_NOT)) + only_term = neg + + return only_term + def hint_tags(self, string, max_results=50): + """ + Return auto-complete hints for tags + using prefix search. + """ toks = self.get_tokens(string, field='SIMPLE') top = BooleanQuery() @@ -844,37 +1274,41 @@ class MultiSearch(Search): q = self.create_prefix_phrase(toks, field) top.add(BooleanClause(q, BooleanClause.Occur.SHOULD)) - book_cat = TermsFilter() - book_cat.addTerm(Term("tag_category", "book")) - - no_book_cat = BooleanFilter() - no_book_cat.add(FilterClause(book_cat, BooleanClause.Occur.MUST_NOT)) + no_book_cat = self.term_filter(Term("tag_category", "book"), inverse=True) return self.search_tags(top, no_book_cat, max_results=max_results) def hint_books(self, string, max_results=50): + """ + Returns auto-complete hints for book titles + Because we do not index 'pseudo' title-tags. + Prefix search. + """ toks = self.get_tokens(string, field='SIMPLE') q = self.create_prefix_phrase(toks, 'title') - only_books = TermsFilter() - only_books.addTerm(Term("is_book", "true")) - - return self.book_search(q, only_books, max_results=max_results) - def filter_by_tags(self, tags): - q = BooleanQuery() - - for tag in tags: - toks = self.get_tokens(tag.name, field='tags') - tag_phrase = PhraseQuery() - for tok in toks: - tag_phrase.add(tok) - q.add(BooleanClause(tok, BooleanClause.Occur.MUST)) + return self.search_books(q, self.term_filter(Term("is_book", "true")), max_results=max_results) - return QueryWrapperFilter(q) + @staticmethod + def chain_filters(filters, op=ChainedFilter.AND): + """ + Chains a filter list together + """ + filters = filter(lambda x: x is not None, filters) + if not filters or filters is []: + return None + chf = ChainedFilter(JArray('object')(filters, Filter), op) + return chf def filtered_categories(self, tags): + """ + Return a list of tag categories, present in tags list. + """ cats = {} for t in tags: cats[t.category] = True return cats.keys() + + def hint(self): + return Hint(self)