X-Git-Url: https://git.mdrn.pl/wolnelektury.git/blobdiff_plain/124258c72bb9e69e6336de195f3a3b3e74509b95..faef1a564460905fd4c78953bf2f7023b5a94fc7:/apps/search/index.py diff --git a/apps/search/index.py b/apps/search/index.py index 77ce877af..6d97047a0 100644 --- a/apps/search/index.py +++ b/apps/search/index.py @@ -25,6 +25,7 @@ import re import errno from librarian import dcparser from librarian.parser import WLDocument +from lxml import etree import catalogue.models from multiprocessing.pool import ThreadPool from threading import current_thread @@ -54,6 +55,8 @@ class WLAnalyzer(PerFieldAnalyzerWrapper): self.addAnalyzer("source_name", simple) self.addAnalyzer("publisher", simple) self.addAnalyzer("authors", simple) + self.addAnalyzer("title", simple) + self.addAnalyzer("is_book", keyword) # shouldn't the title have two forms? _pl and simple? @@ -210,7 +213,7 @@ class Index(BaseIndex): 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)) @@ -221,9 +224,9 @@ class Index(BaseIndex): Create a lucene document referring book id. """ doc = Document() - doc.add(NumericField("book_id", Field.Store.YES, True).setIntValue(book.id)) + 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(book.parent.id)) + doc.add(NumericField("parent_id", Field.Store.YES, True).setIntValue(int(book.parent.id))) return doc def remove_book(self, book): @@ -253,7 +256,7 @@ class Index(BaseIndex): self.index.addDocument(book_doc) del book_doc - self.index_content(book, book_fields=[meta_fields['title'], meta_fields['authors']]) + self.index_content(book, book_fields=[meta_fields['title'], meta_fields['authors'], meta_fields['published_date']]) master_tags = [ 'opowiadanie', @@ -261,10 +264,21 @@ class Index(BaseIndex): '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'] + + 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]+)[\]. ]*$") def extract_metadata(self, book, book_info=None): """ @@ -306,6 +320,12 @@ class Index(BaseIndex): fields[field.name] = Field(field.name, "%04d%02d%02d" %\ (dt.year, dt.month, dt.day), Field.Store.NO, Field.Index.NOT_ANALYZED) + # get published date + source = book_info.source_name + match = self.published_date_re.search(source) + if match is not None: + fields["published_date"] = Field("published_date", str(match.groups()[0]), Field.Store.YES, Field.Index.NOT_ANALYZED) + return fields def add_gaps(self, fields, fieldname): @@ -338,15 +358,26 @@ class Index(BaseIndex): if master is None: return [] - def walker(node): + def walker(node, ignore_tags=[]): yield node, None - for child in list(node): + for child in filter(lambda n: n.tag not in ignore_tags, list(node)): for b, e in walker(child): yield b, e yield None, node return def fix_format(text): + # separator = [u" ", u"\t", u".", u";", u","] + if isinstance(text, list): + # need to join it first + text = filter(lambda s: s is not None, content) + text = u' '.join(text) + # for i in range(len(text)): + # if i > 0: + # if text[i][0] not in separator\ + # and text[i - 1][-1] not in separator: + # text.insert(i, u" ") + return re.sub("(?m)/$", "", text) def add_part(snippets, **fields): @@ -399,53 +430,71 @@ class Index(BaseIndex): 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(header): + # 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:] if start.text is not None: fragments[fid]['themes'] += map(str.strip, map(give_me_utf8, start.text.split(','))) - fragments[fid]['content'].append(start.tail) + 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) + 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)) + + self.index.addDocument(doc) + finally: snippets.close() @@ -539,17 +588,20 @@ class JoinSearch(object): class SearchResult(object): - def __init__(self, searcher, scoreDocs, score=None, how_found=None, snippets=None): - self.snippets = [] + 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.hits = [] + self.boost = 1.0 - stored = searcher.doc(scoreDocs.doc) + self._hits = [] + self._processed_hits = None # processed hits + + stored = search.searcher.doc(scoreDocs.doc) self.book_id = int(stored.get("book_id")) header_type = stored.get("header_type") @@ -562,16 +614,31 @@ class SearchResult(object): fragment = stored.get("fragment_anchor") - hit = (sec + (header_span,), fragment, scoreDocs.score, {'how_found': how_found, 'snippets': [snippets]}) + pd = stored.get("published_date") + if pd is None: + 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.search = search + self.searched = searched + self.tokens_cache = tokens_cache - self.hits.append(hit) + @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 + self._score = other._score return self def get_book(self): @@ -579,34 +646,88 @@ class SearchResult(object): book = property(get_book) - def process_hits(self): - frags = filter(lambda r: r[1] is not None, self.hits) - sect = filter(lambda r: r[1] is None, self.hits) + @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[0][1] >= f[0][1] and s[0][1] < f[0][1] + f[0][2], + 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() + + # remove duplicate sections + sections = {} + for s in sect: - m = {'score': s[2], - 'header_index': s[0][1] + 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[3]) - hits.append(m) + m.update(s[OTHER]) + sections[si] = m + + hits = sections.values() for f in frags: - frag = catalogue.models.Fragment.objects.get(anchor=f[1]) - m = {'score': f[2], + 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, - 'themes': frag.tags.filter(category='theme') + 'section_number': f[POSITION][POSITION_INDEX] + 1, + 'themes': themes, + 'themes_hit': themes_hit } - m.update(f[3]) + m.update(f[OTHER]) hits.append(m) hits.sort(lambda a, b: cmp(a['score'], b['score']), reverse=True) - print("--- %s" % hits) + self._processed_hits = hits return hits @@ -626,7 +747,12 @@ 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): @@ -660,7 +786,7 @@ class Hint(object): lst = self.book_tags.get(t.category, []) lst.append(t) self.book_tags[t.category] = lst - if t.category in ['theme']: + if t.category in ['theme', 'theme_pl']: self.part_tags.append(t) def tag_filter(self, tags, field='tags'): @@ -756,11 +882,14 @@ class Search(IndexStore): bks.append(catalogue.models.Book.objects.get(id=doc.get("book_id"))) return (bks, tops.totalHits) - def get_tokens(self, searched, field='content'): + 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): @@ -772,6 +901,10 @@ class Search(IndexStore): 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): @@ -828,9 +961,40 @@ class Search(IndexStore): 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 = {} + + 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): """ @@ -853,12 +1017,39 @@ class Search(IndexStore): 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_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 containt a phrase perfectly matching (with a slop of 2, default for make_phrase()) + 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']] @@ -874,16 +1065,17 @@ class Search(IndexStore): flt]), max_results) for found in top.scoreDocs: - books.append(SearchResult(self.searcher, found, 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, hint=None): + 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 @@ -893,29 +1085,40 @@ class Search(IndexStore): # content only query : themes x content q = BooleanQuery() - tokens = self.get_tokens(searched) - if hint is None or hint.just_search_in(['themes_pl']) != []: - q.add(BooleanClause(self.make_term_query(tokens, field='themes_pl', + tokens_pl = self.get_tokens(searched, field='content', cached=tokens_cache) + tokens = self.get_tokens(searched, field='SIMPLE', cached=tokens_cache) + + # 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, field='content', + q.add(BooleanClause(self.make_term_query(tokens_pl, field='content', fuzzy=fuzzy), BooleanClause.Occur.SHOULD)) topDocs = self.searcher.search(q, only_in, max_results) for found in topDocs.scoreDocs: - books.append(SearchResult(self.searcher, found)) + books.append(SearchResult(self, found, how_found='search_everywhere_themesXcontent', searched=searched)) + print "* %s theme x content: %s" % (searched, books[-1]._hits) # query themes/content x author/title/tags q = BooleanQuery() - # in_meta = BooleanQuery() in_content = BooleanQuery() + in_meta = BooleanQuery() - for fld in ['themes', 'content', 'tags', 'authors', 'title']: - in_content.add(BooleanClause(self.make_term_query(tokens, field=fld, fuzzy=False), 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 ['tags', 'authors', 'title']: + in_meta.add(BooleanClause(self.make_term_query(tokens, field=fld, fuzzy=False), BooleanClause.Occur.SHOULD)) + + q.add(BooleanClause(in_content, BooleanClause.Occur.MUST)) + q.add(BooleanClause(in_meta, BooleanClause.Occur.SHOULD)) topDocs = self.searcher.search(q, only_in, max_results) for found in topDocs.scoreDocs: - books.append(SearchResult(self.searcher, found)) + books.append(SearchResult(self, found, how_found='search_everywhere', searched=searched)) + print "* %s scatter search: %s" % (searched, books[-1]._hits) return books @@ -961,7 +1164,6 @@ class Search(IndexStore): # return None - def get_snippets(self, scoreDoc, query, field='content'): """ Returns a snippet for found scoreDoc. @@ -971,17 +1173,20 @@ class Search(IndexStore): 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((int(stored.get('snippets_position')), - int(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) - # import pdb; pdb.set_trace() snip = highlighter.getBestFragments(tokenStream, text, 3, "...") return snip @@ -1089,7 +1294,7 @@ class Search(IndexStore): Chains a filter list together """ filters = filter(lambda x: x is not None, filters) - if not filters: + if not filters or filters is []: return None chf = ChainedFilter(JArray('object')(filters, Filter), op) return chf