store contents in index for highlighting fragments
[wolnelektury.git] / apps / search / index.py
index 8f7722a..36f4247 100644 (file)
-
+# -*- coding: utf-8 -*-
 from django.conf import settings
-from lucene import SimpleFSDirectory, IndexWriter
+from lucene import SimpleFSDirectory, IndexWriter, File, Field, \
+    NumericField, Version, Document, JavaError, IndexSearcher, \
+    QueryParser, Term, PerFieldAnalyzerWrapper, \
+    SimpleAnalyzer, PolishAnalyzer, ArrayList, \
+    KeywordAnalyzer, NumericRangeQuery, BooleanQuery, \
+    BlockJoinQuery, BlockJoinCollector, TermsFilter, \
+    HashSet, BooleanClause, Term, CharTermAttribute, \
+    PhraseQuery, StringReader, TermQuery, BlockJoinQuery, \
+    Sort, Integer
+    # KeywordAnalyzer
+import sys
 import os
+import errno
+from librarian import dcparser
+from librarian.parser import WLDocument
+import catalogue.models
+from multiprocessing.pool import ThreadPool
+from threading import current_thread
+import atexit
 
 
-class BookSearch(object):
+class WLAnalyzer(PerFieldAnalyzerWrapper):
     def __init__(self):
-        if not os.exists(settings.SEARCH_INDEX):
-            os.mkdir(settings.SEARCH_INDEX)
-        self.store = IndexWriter(store, )
-                         
+        polish = PolishAnalyzer(Version.LUCENE_34)
+        simple = SimpleAnalyzer(Version.LUCENE_34)
+        keyword = KeywordAnalyzer(Version.LUCENE_34)
+        # not sure if needed: there's NOT_ANALYZED meaning basically the same
+
+        PerFieldAnalyzerWrapper.__init__(self, polish)
+
+        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("is_book", keyword)
+
+        #self.addanalyzer("fragment_anchor", keyword)
+
+
+class IndexStore(object):
+    def __init__(self):
+        self.make_index_dir()
+        self.store = SimpleFSDirectory(File(settings.SEARCH_INDEX))
+
+    def make_index_dir(self):
+        try:
+            os.makedirs(settings.SEARCH_INDEX)
+        except OSError as exc:
+            if exc.errno == errno.EEXIST:
+                pass
+            else: raise
+
+
+class Index(IndexStore):
+    def __init__(self, analyzer=None):
+        IndexStore.__init__(self)
+        self.index = None
+        if not analyzer:
+            analyzer = WLAnalyzer()
+        self.analyzer = analyzer
+
+    def open(self, analyzer=None):
+        if self.index:
+            raise Exception("Index is already opened")
+        self.index = IndexWriter(self.store, self.analyzer,\
+                                 IndexWriter.MaxFieldLength.LIMITED)
+        return self.index
+
+    def close(self):
+        self.index.optimize()
+        self.index.close()
+        self.index = None
+
+    def remove_book(self, book):
+        q = NumericRangeQuery.newIntRange("book_id", book.id, book.id, True,True)
+        self.index.deleteDocuments(q)
+
+    def index_book(self, book, overwrite=True):
+        if overwrite:
+            self.remove_book(book)
+
+
+        doc = self.extract_metadata(book)
+        parts = self.extract_content(book)
+        block = ArrayList().of_(Document)
+
+        for p in parts:
+            block.add(p)
+        block.add(doc)
+        self.index.addDocuments(block)
+
+    master_tags = [
+        'opowiadanie',
+        'powiesc',
+        'dramat_wierszowany_l',
+        'dramat_wierszowany_lp',
+        'dramat_wspolczesny', 'liryka_l', 'liryka_lp',
+        'wywiad'
+        ]
+
+    skip_header_tags = ['autor_utworu', 'nazwa_utworu', 'dzielo_nadrzedne']
+
+    def create_book_doc(self, book):
+        """
+        Create a lucene document connected to the book
+        """
+        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):
+        book_info = dcparser.parse(book.xml_file)
+
+        print("extract metadata for book %s id=%d, thread%d" % (book.slug, book.id, current_thread().ident))
+        
+        doc = self.create_book_doc(book)
+        doc.add(Field("slug", book.slug, Field.Store.NO, Field.Index.ANALYZED_NO_NORMS))
+        doc.add(Field("tags", ','.join([t.name for t in book.tags]), Field.Store.NO, Field.Index.ANALYZED))
+        doc.add(Field("is_book", 'true', Field.Store.NO, Field.Index.NOT_ANALYZED))
+
+        # validator, name
+        for field in dcparser.BookInfo.FIELDS:
+            if hasattr(book_info, field.name):
+                if not getattr(book_info, field.name):
+                    continue
+                # since no type information is available, we use validator
+                type_indicator = field.validator
+                if type_indicator == dcparser.as_unicode:
+                    s = getattr(book_info, field.name)
+                    if field.multiple:
+                        s = ', '.join(s)
+                    try:
+                        doc.add(Field(field.name, s, Field.Store.NO, Field.Index.ANALYZED))
+                    except JavaError as je:
+                        raise Exception("failed to add field: %s = '%s', %s(%s)" % (field.name, s, je.message, je.args))
+                elif type_indicator == dcparser.as_person:
+                    p = getattr(book_info, field.name)
+                    if isinstance(p, dcparser.Person):
+                        persons = unicode(p)
+                    else:
+                        persons = ', '.join(map(unicode, p))
+                    doc.add(Field(field.name, persons, Field.Store.NO, Field.Index.ANALYZED))
+                elif type_indicator == dcparser.as_date:
+                    dt = getattr(book_info, field.name)
+                    doc.add(Field(field.name, "%04d%02d%02d" % (dt.year, dt.month, dt.day), Field.Store.NO, Field.Index.NOT_ANALYZED))
+        return doc
+
+    def get_master(self, root):
+        for master in root.iter():
+            if master.tag in self.master_tags:
+                return master
+
+    
+    def extract_content(self, book):
+        wld = WLDocument.from_file(book.xml_file.path)
+        root = wld.edoc.getroot()
+
+        # first we build a sequence of top-level items.
+        # book_id
+        # header_index - the 0-indexed position of header element.
+        # content
+        master = self.get_master(root)
+        if master is None:
+            return []
+        
+        header_docs = []
+        for header, position in zip(list(master), range(len(master))):
+            if header.tag in self.skip_header_tags:
+                continue
+            doc = self.create_book_doc(book)
+            doc.add(NumericField("header_index", Field.Store.YES, True).setIntValue(position))
+            doc.add(Field("header_type", header.tag, Field.Store.YES, Field.Index.NOT_ANALYZED))
+            content = u' '.join([t for t in header.itertext()])
+            doc.add(Field("content", content, Field.Store.YES, Field.Index.ANALYZED))
+            header_docs.append(doc)
+
+        def walker(node):
+            yield node, None
+            for child in list(node):
+                for b, e in walker(child):
+                    yield b, e
+            yield None, node
+            return
+
+        # Then we create a document for each fragments
+        # fragment_anchor - the anchor
+        # themes - list of themes [not indexed]
+        fragment_docs = []
+        # will contain (framgent id -> { content: [], themes: [] }
+        fragments = {}
+        for start, end in walker(master):
+            if start is not None and start.tag == 'begin':
+                fid = start.attrib['id'][1:]
+                fragments[fid] = {'content': [], 'themes': []}
+                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)
+            elif start is not None and start.tag == 'end':
+                fid = start.attrib['id'][1:]
+                if fid not in fragments:
+                    continue  # a broken <end> node, skip it
+                frag = fragments[fid]
+                del fragments[fid]
+
+                def jstr(l):
+                    return u' '.join(map(
+                        lambda x: x == None and u'(none)' or unicode(x),
+                        l))
+
+                doc = self.create_book_doc(book)
+                doc.add(Field("fragment_anchor", fid,
+                              Field.Store.YES, Field.Index.NOT_ANALYZED))
+                doc.add(Field("content",
+                              u' '.join(filter(lambda s: s is not None, frag['content'])),
+                              Field.Store.YES, Field.Index.ANALYZED))
+                doc.add(Field("themes",
+                              u' '.join(filter(lambda s: s is not None, frag['themes'])),
+                              Field.Store.NO, Field.Index.ANALYZED))
+
+                fragment_docs.append(doc)
+            elif start is not None:
+                for frag in fragments.values():
+                    frag['content'].append(start.text)
+            elif end is not None:
+                for frag in fragments.values():
+                    frag['content'].append(end.tail)
+
+        return header_docs + fragment_docs
+
+    def __enter__(self):
+        self.open()
+        return self
+
+    def __exit__(self, type, value, tb):
+        self.close()
+
+
+class ReusableIndex(Index):
+    """
+    Works like index, but does not close/optimize Lucene index
+    until program exit (uses atexit hook).
+    This is usefull for importbooks command.
+
+    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)
+            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(Index.index_book, (self,)+ args, kw)
+        ReusableIndex.pool_jobs.append(job)
+
+    @staticmethod
+    def close_reusable():
+        if ReusableIndex.index is not None:
+            print("closing index")
+            for job in ReusableIndex.pool_jobs:
+                job.wait()
+            ReusableIndex.pool.close()
+
+            ReusableIndex.index.optimize()
+            ReusableIndex.index.close()
+            ReusableIndex.index = None
+
+    def close(self):
+        pass
+
+
+class Search(IndexStore):
+    def __init__(self, default_field="content"):
+        IndexStore.__init__(self)
+        self.analyzer = 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)
+
+    def wrapjoins(self, query, fields=[]):
+        """
+        This functions modifies the query in a recursive way,
+        so Term and Phrase Queries contained, which match
+        provided fields are wrapped in a BlockJoinQuery,
+        and so delegated to children documents.
+        """
+        if BooleanQuery.instance_(query):
+            qs = BooleanQuery.cast_(query)
+            for clause in qs:
+                clause = BooleanClause.cast_(clause)
+                clause.setQuery(self.wrapjoins(clause.getQuery(), fields))
+            return qs
+        else:
+            termset = HashSet()
+            query.extractTerms(termset)
+            for t in termset:
+                t = Term.cast_(t)
+                if t.field() not in fields:
+                    return query
+            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(catalogue.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(catalogue.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)
+
+        tops = self.searcher.search(bjq, 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)
+
+# 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 MultiSearch(Search):
+    """Class capable of IMDb-like searching"""
+    def get_tokens(self, queryreader):
+        if isinstance(queryreader, str) or isinstance(queryreader, unicode):
+            queryreader = StringReader(queryreader)
+        queryreader.reset()
+        tokens = self.analyzer.reusableTokenStream('content', queryreader)
+        toks = []
+        while tokens.incrementToken():
+            cta = tokens.getAttribute(CharTermAttribute.class_)
+            toks.append(cta.toString())
+        return toks
+
+    def make_phrase(self, tokens, field='content', slop=2):
+        phrase = PhraseQuery()
+        phrase.setSlop(slop)
+        for t in tokens:
+            term = Term(field, t)
+            phrase.add(term)
+        return phrase
+
+    def make_term_query(self, tokens, field='content', modal=BooleanClause.Occur.SHOULD):
+        q = BooleanQuery()
+        for t in tokens:
+            term = Term(field, t)
+            q.add(BooleanClause(TermQuery(term), modal))
+        return q
+
+    def content_query(self, query):
+        return BlockJoinQuery(query, self.parent_filter,
+                              BlockJoinQuery.ScoreMode.Total)
+
+    def search_perfect(self, tokens, max_results=20):
+        qrys = [self.make_phrase(tokens, field=fld) for fld in ['author', 'title', 'content']]
+
+        books = []
+        for q in qrys:
+            top = self.searcher.search(q, max_results)
+            for found in top.scoreDocs:
+                book_info = self.searcher.doc(found.doc)
+                books.append((found.score, catalogue.models.Book.objects.get(id=book_info.get("book_id")), []))
+        return books
+
+    def search_everywhere(self, tokens, max_results=20):
+        q = BooleanQuery()
+        in_meta = BooleanQuery()
+        in_content = BooleanQuery()
+
+        for fld in ['themes', 'content']:
+            in_content.add(BooleanClause(self.make_term_query(tokens, field=fld), BooleanClause.Occur.SHOULD))
+
+        for fld in ['author', 'title', 'epochs', 'genres', 'kinds']:
+            in_meta.add(BooleanClause(self.make_term_query(tokens, field=fld), 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))
+
+        collector = BlockJoinCollector(Sort.RELEVANCE, 100, True, True)
+
+        self.searcher.search(q, collector)
+
+        books = []
+
+        top_groups = collector.getTopGroups(in_content_join, Sort.RELEVANCE, 0, max_results, 0, True)
+        if top_groups:
+            for grp in top_groups.groups:
+                doc_id = Integer.cast_(grp.groupValue).intValue()
+                book_data = self.searcher.doc(doc_id)
+                book = catalogue.models.Book.objects.get(id=book_data.get("book_id"))
+                parts = []
+                for part in grp.scoreDocs:
+                    part_data = self.searcher.doc(part.doc)
+                    header_type = part_data.get("header_type")
+                    if header_type:
+                        parts.append((part.score, {"header": header_type, "position": int(part_data.get("header_index"))}))
+                    fragment = part_data.get("fragment_anchor")
+                    if fragment:
+                        fragment = book.fragments.get(anchor=fragment)
+                        parts.append((part.score, {"fragment": fragment}))
+                books.append((grp.maxScore, book, parts))
+                
+        return books
+
+
+    def multisearch(self, query, max_results=50):
+        """
+        Search strategy:
+        - (phrase) OR -> content
+                      -> title
+                      -> author
+        - (keywords)  -> author
+                      -> motyw
+                      -> tags
+                      -> content
+        """
+        # queryreader = StringReader(query)
+        # tokens = self.get_tokens(queryreader)
+
+        # top_level = BooleanQuery()
+        # Should = BooleanClause.Occur.SHOULD
+
+        # phrase_level = BooleanQuery()
+        # phrase_level.setBoost(1.3)
+
+        # p_content = self.make_phrase(tokens, joined=True)
+        # p_title = self.make_phrase(tokens, 'title')
+        # p_author = self.make_phrase(tokens, 'author')
+
+        # phrase_level.add(BooleanClause(p_content, Should))
+        # phrase_level.add(BooleanClause(p_title, Should))
+        # phrase_level.add(BooleanClause(p_author, Should))
+
+        # kw_level = BooleanQuery()
+
+        # kw_level.add(self.make_term_query(tokens, 'author'), Should)
+        # j_themes = self.make_term_query(tokens, 'themes', joined=True)
+        # kw_level.add(j_themes, Should)
+        # kw_level.add(self.make_term_query(tokens, 'tags'), Should)
+        # j_con = self.make_term_query(tokens, joined=True)
+        # kw_level.add(j_con, Should)
+
+        # top_level.add(BooleanClause(phrase_level, Should))
+        # top_level.add(BooleanClause(kw_level, Should))
+
+        return None
+
+    
+    def do_search(self, query, max_results=50, collector=None):
+        tops = self.searcher.search(query, max_results)
+        #tops = self.searcher.search(p_content, max_results)
+
+        bks = []
+        for found in tops.scoreDocs:
+            doc = self.searcher.doc(found.doc)
+            b = catalogue.models.Book.objects.get(id=doc.get("book_id"))
+            bks.append(b)
+            print "%s (%d) -> %f" % (b, b.id, found.score)
+        return (bks, tops.totalHits)