Python punkt.PunktSentenceTokenizer类代码示例

def preprocess(phys):
    '''
    :param fname: a text file
    :return: a json of sentences, processed for searchability
    '''

    phys = phys.decode('utf-8')
    phys = re.sub('(\n)+', '. ', phys)

    sentence_tokenizer = PunktSentenceTokenizer()
    sentences = sentence_tokenizer.tokenize(phys)

    for i in xrange(len(sentences)):
        sentence = unicode(sentences[i])
        sentence = sentence.replace('\n', ' ')
        sentence = re.sub(' +',' ',sentence)
        sentence = re.sub(r'\d+', '', sentence)
        sentence = sentence.replace("-"," ")
        exclude = string.punctuation
        sentence = ''.join(ch for ch in sentence if ch not in exclude)
        sentence = re.sub(' +',' ',sentence)
        sentences[i] = sentence
        # sentences[i] = sentence.encode('utf-8')
    count = 0
    for sentence in sentences:
        if sentence == ' ' or sentence == '':
            sentences.pop(count)
        count +=1

    # with open(fname.rstrip('txt')+'json', 'w') as outfile:
    #     json.dump(sentences, outfile)

    return sentences

 def _split_sentences(self, text):
     from nltk.tokenize.punkt import PunktSentenceTokenizer, PunktParameters
     punkt_param = PunktParameters()
     punkt_param.abbrev_types = set(['dr', 'vs', 'mr', 'mrs', 'prof', 'inc'])
     sentence_splitter = PunktSentenceTokenizer(punkt_param)
     sentences = sentence_splitter.tokenize(text)
     return sentences

 def summarize(self):
     punkt_param = PunktParameters()
     punkt_param.abbrev_types = set(['dr', 'vs', 'mr', 'mrs', 'prof', 'inc'])
     sentence_splitter = PunktSentenceTokenizer(punkt_param)
     sentences = sentence_splitter.tokenize(self.text)
     structure = {}
     sentence_objects = []
     for idx in range(len(sentences)):
         obj = {'text' : sentences[idx], 'index' : idx , 'data': {}}
         sentence_objects.append(obj)
     structure['sentences'] = sentence_objects
     self.sentencecount = len(structure['sentences'])
     structure['ordered'] = []
     structure['weights'] = {'words' : FreqDist(nltk.word_tokenize(preprocess(self.text))), 'total': 0, 'transformed': 0}
     structure['weights']['total'] = sum(structure['weights']['words'].values())
     self.sentenceIndex = 0
     for each_sent in structure['sentences']:
         each_sent['data']['tokens'] = nltk.word_tokenize(preprocess(each_sent['text']))
         each_sent['data']['sinTransform'] = (1-math.sin(self.sentenceIndex*(math.pi/self.sentencecount)))+1
         for each_word in structure['weights']['words']:
             if each_word in each_sent['data']['tokens']:
                 structure['weights']['words'][each_word] *= each_sent['data']['sinTransform']
         self.sentenceIndex += 1
     structure['weights']['transformed'] = sum(structure['weights']['words'].values())
     self.sentenceIndex = 0
     for each_sent in structure['sentences']:
         each_sent['data']['weights'] = {'words': self.calculate_relative_frequence(each_sent['data']['tokens'], structure['weights']['words']), 'total': 0}
         each_sent['data']['weights']['total'] = sum(each_sent['data']['weights']['words'].values())
         self.sentenceIndex += 1
     structure['ordered'] = sorted(structure['sentences'], key=lambda x:x['data']['weights']['total'], reverse=True)
     structure_keep = structure['ordered'][:self.quota]
     structure_keep.sort(key=lambda x:x['index'])
     for eac_sen in structure_keep:
         self.summary.append(eac_sen['text'])

def textrank(document):
    pst = PunktSentenceTokenizer()
    sentences = pst.tokenize(document)

    # Bag of Words
    from sklearn.feature_extraction.text import CountVectorizer
    cv = CountVectorizer()
    bow_matrix = cv.fit_transform(sentences)

    from sklearn.feature_extraction.text import TfidfTransformer
    normalized_matrix = TfidfTransformer().fit_transform(bow_matrix)

    ## mirrored matrix where the rows and columns correspond to 
    ## sentences, and the elements describe how similar the
    ## sentences are. score 1 means sentences are exactly the same.
    similarity_graph = normalized_matrix * normalized_matrix.T
    similarity_graph.toarray()

    # PageRank
    import networkx as nx
    nx_graph = nx.from_scipy_sparse_matrix(similarity_graph)

    ## mapping of sentence indices to scores. use them to associate
    ## back to the original sentences and sort them
    scores = nx.pagerank(nx_graph)
    ranked = sorted(((scores[i], s) for i,s in enumerate(sentences)), reverse=True)
    print ranked[0][1]

def featureize(F, observation_files):

    word_tokenizer = PunktSentenceTokenizer()
    sent_tokenizer = PunktSentenceTokenizer()

    m = len(observation_files)

    # X is Nx2
    X = np.zeros((m,2), dtype=np.float)

    for (i,filename) in enumerate(observation_files,start=0):

        file_text  = read_file(filename).decode('string_escape')

        try:
            num_sents = len(sent_tokenizer.sentences_from_text(file_text))
        except UnicodeDecodeError:
            num_sents = 2

        #num_tokens = len(word_tokenize(file_text))
        num_tokens = len(file_text.split())

        # Return two features: 
        # 1 (0) - Number of sentences per file
        # 2 (1) - Number of tokens per file
        X[i][0] = num_sents
        X[i][1] = num_tokens

    return X

    def get_key_sentences(self, n=5):
        '''
        Uses a simple implementation of TextRank to extract the top N sentences
        from a document.

        Sources:
        - Original paper: http://acl.ldc.upenn.edu/acl2004/emnlp/pdf/Mihalcea.pdf
        - Super useful blog post: http://joshbohde.com/blog/document-summarization
        - Wikipedia: http://en.wikipedia.org/wiki/Automatic_summarization#Unsupervised_keyphrase_extraction:_TextRank
        '''
        # Tokenize the document into sentences. More NLP preprocesing should also happen here. 
        sentence_tokenizer = PunktSentenceTokenizer()
        sentences = sentence_tokenizer.tokenize(self.doc)

        # Calculate word counts and TFIDF vectors
        word_counts = CountVectorizer(min_df=0).fit_transform(sentences)
        normalized = TfidfTransformer().fit_transform(word_counts) 

        # Normalized graph * its transpose yields a sentence-level similarity matrix
        similarity_graph = normalized * normalized.T
     
        nx_graph = nx.from_scipy_sparse_matrix(similarity_graph)
        scores = nx.pagerank(nx_graph)
        return sorted(((scores[i],s) for i,s in enumerate(sentences)),
                      reverse=True)[n]

    def tokenize_sentences(self, untokenized_string: str):
        """Tokenize sentences by reading trained tokenizer and invoking
        ``PunktSentenceTokenizer()``.
        :type untokenized_string: str
        :param untokenized_string: A string containing one of more sentences.
        :rtype : list of strings
        """
        # load tokenizer
        assert isinstance(untokenized_string, str), \
            'Incoming argument must be a string.'

        if self.language == 'latin':
            tokenizer = super()
        elif self.language == 'greek': # Workaround for regex tokenizer
            self.sent_end_chars=GreekLanguageVars.sent_end_chars
            self.sent_end_chars_regex = '|'.join(self.sent_end_chars)
            self.pattern = rf'(?<=[{self.sent_end_chars_regex}])\s'
        elif self.language in INDIAN_LANGUAGES:
            self.sent_end_chars=SanskritLanguageVars.sent_end_chars
            self.sent_end_chars_regex = '|'.join(self.sent_end_chars)
            self.pattern = rf'(?<=[{self.sent_end_chars_regex}])\s'
        else:
            # Warn that NLTK Punkt is being used by default???
            tokenizer = PunktSentenceTokenizer()

        # mk list of tokenized sentences
        if self.language == 'greek' or self.language in INDIAN_LANGUAGES:
            return re.split(self.pattern, untokenized_string)
        else:
            return tokenizer.tokenize(untokenized_string)

    def preprocessin(self, cell_value):

        # to tokenize the tweet into sentences
        tweet = PunktSentenceTokenizer().tokenize(cell_value)
        # to remove 'u'
        tweet = '\n'.join(tweet)
        # to remove html tags
        tweet = self.remTags(tweet)
        # to lower aplphabets
        tweet = tweet.lower()

        ##Removing all junk
        tweet = re.sub(u'(RT |\\\\|\u201c)"[email protected]*?[: ]', ' ', tweet)
        tweet = re.sub('@', ' ', tweet)
        tweet = re.sub(r'[^\x00-\x7F]', ' ', tweet)
        tweet = re.sub('[\s]+', ' ', tweet)
        tweet = re.sub('_', ' ', tweet)
        tweet = re.sub('((www\.[\s]+)|(https?://[^\s]+))', '', tweet)
        tweet = re.sub(r'\\([^\s]+)', ' ', tweet)
        tweet = re.sub(u'[\u2018\u2019]', '\'', tweet)
        tweet = re.sub('(^|)?http?s?:?/?/?.*?( |$)', ' ', tweet)
        tweet = re.sub(u'\u2026', ' ', tweet)
        tweet = re.sub('---', ' ', tweet)
        tweet = re.sub(u'[\u201c\u201d]', '"', tweet)
        tweet = re.sub('\[email protected]*?( |:|$)', ' ', tweet)
        tweet = re.sub(r"\.\.+", ' ', tweet)
        tweet = re.sub('&amp', ' ', tweet)
        tweet = re.sub('\.\.\.', ' ', tweet)
        tweet = tweet.strip('\'"')
        tweet = re.sub('(, |\.( |$))', ' ', tweet)
        tweet = re.sub('[][!"$*,/;<=>[email protected]\\\\^_`{|}~]', ' ', tweet)
        tweet = re.sub('( - )', ' ', tweet)

        return tweet

def get_todo_items(text):
    all_items = list()
    tokenizer = PunktSentenceTokenizer()
    sen_tokens = tokenizer.tokenize(text)

    for sen_token in sen_tokens:
        todo_items = list()
        tokens = nltk.word_tokenize(sen_token)

        tags = tagger.tag(tokens)
        stop_words = [word for (word, tag) in tags if tag in (tagVB, tagVBP)]

        ind = -1
        for word in stop_words:
            curr_ind = tokens.index(word)
            if curr_ind != 0 and tags[curr_ind - 1][1] in (tagCC, tagRB):
                to_ind = curr_ind - 1
            else: to_ind = curr_ind
            if ind != -1 and abs(to_ind - ind) > 1:
                todo_items.append(' '.join(tokens[ind:get_punctuation_index(tokens, ind, to_ind)]))
            elif ind != -1 and len(todo_items) > 0:
                last_ind = len(todo_items)
                todo_items[last_ind - 1] = ' '.join([todo_items[last_ind - 1], tokens[to_ind - 1]])
            ind = curr_ind

        if ind != -1 and abs(len(tokens) - ind) > 1:
            todo_items.append(' '.join(tokens[ind:get_punctuation_index(tokens, ind, len(tokens))]))
        elif ind != -1 and len(todo_items) > 0:
            last_ind = len(todo_items)
            todo_items[last_ind - 1] = ' '.join([todo_items[last_ind - 1], tokens[len(tokens) - 1]])

        all_items.extend(todo_items)

    return all_items

 def fractal_representation(self):
     punkt_param = PunktParameters()
     for each_paragraph in self.paragraphs:
         buffer_p = paragraph()
         buffer_p.paragraph = each_paragraph
         buffer_p.tokens = nltk.word_tokenize(preprocess(each_paragraph))
         buffer_p.weights['words'] = FreqDist(buffer_p.tokens)
         buffer_p.weights['total'] = {'words':0, 'sentences':0}    
         punkt_param.abbrev_types = set(['dr', 'vs', 'mr', 'mrs', 'prof', 'inc'])
         sentence_splitter = PunktSentenceTokenizer(punkt_param)
         sentences = sentence_splitter.tokenize(each_paragraph)
         for each_sentence in sentences:
             self.stotal += 1
             buffer_s = sentence()
             buffer_s.sentence = each_sentence
             buffer_s.tokens = nltk.word_tokenize(preprocess(each_sentence))
             if len(buffer_s.tokens) > 0:
                 buffer_s.weights['sentence'] = FreqDist(buffer_s.tokens)
                 buffer_s.weights['paragraph'] = self.calculate_relative_frequence(buffer_s.tokens, buffer_p.weights['words'])
                 buffer_s.weights['document'] = self.calculate_relative_frequence(buffer_s.tokens, self.fractal.weights)
                 buffer_s.weights['total'] = {}
                 buffer_s.weights['total']['sentence'] = 1
                 buffer_s.weights['total']['paragraph'] = sum(buffer_s.weights['paragraph'].values())
                 buffer_s.weights['total']['document'] = sum(buffer_s.weights['document'].values())
                 self.s_weight += buffer_s.weights['total']['document']
                 buffer_p.weights['total']['sentences'] += buffer_s.weights['total']['document']
                 buffer_p.sentences.append(buffer_s)
         buffer_p.weights['total']['words'] = sum(buffer_p.weights['words'].values())
         self.fractal.paragraphs.append(buffer_p)
         self.pindex += 1

    def sentence_tokenizer(self, untokenized_string, language):
        """Reads language .pickle for right language"""
        if language == 'greek':
            pickle_path = os.path.expanduser('~/cltk_data/greek/cltk_linguistic_data/tokenizers/sentence/greek.pickle')
            language_punkt_vars = PunktLanguageVars
            language_punkt_vars.sent_end_chars = ('.', ';')
            language_punkt_vars.internal_punctuation = (',', '·')
        elif language == 'latin':
            pickle_path = os.path.expanduser('~/cltk_data/latin/cltk_linguistic_data/tokenizers/sentence/latin.pickle')
            language_punkt_vars = PunktLanguageVars
            language_punkt_vars.sent_end_chars = ('.', '?', ':')
            language_punkt_vars.internal_punctuation = (',', ';')
        else:
            print("No sentence tokenizer for this language available.")

        with open(pickle_path, 'rb') as open_pickle:
            tokenizer = pickle.load(open_pickle)
        tokenizer.INCLUDE_ALL_COLLOCS = True
        tokenizer.INCLUDE_ABBREV_COLLOCS = True
        params = tokenizer.get_params()
        sbd = PunktSentenceTokenizer(params)
        tokenized_sentences = []
        for sentence in sbd.sentences_from_text(untokenized_string,
                                                realign_boundaries=True):
            tokenized_sentences.append(sentence)
        return tokenized_sentences

def _punkt_sent_tokenize(text):
    '''
     Sentence segmentation using nltk PunktSentenceTokenizer.
    '''
    punkt_param = PunktParameters()
    punkt_param.abbrev_types = set(config.tokenize_abbrev)
    sentence_splitter = PunktSentenceTokenizer(punkt_param)
    return sentence_splitter.tokenize(text)

 def tokenize(self):
     """
     Returns a list of tokenized sentences
     """
     sentence_tokenizer = PunktSentenceTokenizer()
     sentences = sentence_tokenizer.sentences_from_text(self.text)
     sentences = [sentence.split() for sentence in sentences]
     sentences = [[word.strip(",.?!") for word in sentence] for sentence in sentences]
     return sentences

 def transform(self,documents):
     sentence_splitter = PunktSentenceTokenizer()
     for doc in documents:
         if not 'sentences' in doc.ext:
             doc.ext['sentences'] = [s.strip() for s in sentence_splitter.tokenize(doc.text)]
     # for doc in documents:
     #     if not 'sentences' in doc.ext:
     #         doc.ext['sentences'] = [s.strip() for s in doc.text.split('.') if s]
     return documents

def parse (text):
    """Use nltk's PunktSentenceTokenizer to convert the text string into
    a list of English-language sentences."""

    punkt_param = PunktParameters()
    punkt_param.abbrev_types = set(ABBREVIATIONS)
    sentence_splitter = PunktSentenceTokenizer(punkt_param)

    return sentence_splitter.tokenize(preprocess(text))