Twitterのリプライを機械学習用に整形する

随時更新していきます。

pip install emoji

cleanup.py

import re, emoji
from xml.sax.saxutils import unescape

def sentence(sentence):
    if sentence is None:
        return ''

    # 特殊文字デコード
    sentence = unescape(sentence)

    # ユーザー名削除
    sentence = re.sub(r'@[0-9a-zA-Z_:]*', "", sentence)

    # ハッシュタグ削除
    sentence = re.sub(r'#.*', "", sentence)

    # URL削除
    sentence = re.sub(r'(https?)(:\/\/[-_.!~*\'()a-zA-Z0-9;\/?:\@&=+\$,%#]+)', "", sentence)

    # RT削除
    sentence = re.sub(r'(RT ).*$', "", sentence)

    # 「w」はすべて半角1文字に
    sentence = re.sub(r'ｗ+', "w", sentence)
    sentence = re.sub(r'w+', "w", sentence)

    # 「・」を3点リーダにまとめる
    sentence = re.sub(r'・・+', "…", sentence)
    sentence = re.sub(r'･･+', "…", sentence)
    sentence = re.sub(r'\.\.+', "…", sentence)

    # 「！」はすべて全角単！に
    sentence = re.sub(r'！+', "！", sentence)
    sentence = re.sub(r'!+', "！", sentence)

    # 「？」はすべて全角単？に
    sentence = re.sub(r'？+', "？", sentence)
    sentence = re.sub(r'\?+', "？", sentence)

    # 連続する「っ」「ッ」をつめる
    sentence = re.sub(r'っ+', "っ", sentence)
    sentence = re.sub(r'ッ+', "ッ", sentence)

    # 連続する「あ行」をつめる（3文字以上）
    sentence = re.sub(r'あああ+', "あああ", sentence)
    sentence = re.sub(r'いいい+', "いいい", sentence)
    sentence = re.sub(r'ううう+', "ううう", sentence)
    sentence = re.sub(r'えええ+', "えええ", sentence)
    sentence = re.sub(r'おおお+', "おおお", sentence)
    sentence = re.sub(r'ぁぁぁ+', "ぁぁぁ", sentence)
    sentence = re.sub(r'ぃぃぃ+', "ぃぃぃ", sentence)
    sentence = re.sub(r'ぅぅぅ+', "ぅぅぅ", sentence)
    sentence = re.sub(r'ぇぇぇ+', "ぇぇぇ", sentence)
    sentence = re.sub(r'ぉぉぉ+', "ぉぉぉ", sentence)
    sentence = re.sub(r'アアア+', "アアア", sentence)
    sentence = re.sub(r'イイイ+', "イイイ", sentence)
    sentence = re.sub(r'ウウウ+', "ウウウ", sentence)
    sentence = re.sub(r'エエエ+', "エエエ", sentence)
    sentence = re.sub(r'オオオ+', "オオオ", sentence)
    sentence = re.sub(r'ァァァ+', "ァァァ", sentence)
    sentence = re.sub(r'ィィィ+', "ィィィ", sentence)
    sentence = re.sub(r'ゥゥゥ+', "ゥゥゥ", sentence)
    sentence = re.sub(r'ェェェ+', "ェェェ", sentence)
    sentence = re.sub(r'ォォォ+', "ォォォ", sentence)

    # 連続する「笑」をつめる
    sentence = re.sub(r'笑+', "笑", sentence)

    # 連続する「b」をつめる
    sentence = re.sub(r'ｂ+', "b", sentence)
    sentence = re.sub(r'b+', "b", sentence)

    # 連続する「＞」「＜」をつめる
    sentence = re.sub(r'＞+', "＞", sentence)
    sentence = re.sub(r'＜+', "＜", sentence)
    sentence = re.sub(r'>+', ">", sentence)
    sentence = re.sub(r'<+', "<", sentence)

    # 連続する「ー」「～」をつめる
    sentence = re.sub(r'ー+', "ー", sentence)
    sentence = re.sub(r'～+', "～", sentence)
    sentence = re.sub(r'~+', "～", sentence)
    sentence = re.sub(r'〜+', "～", sentence)

    # 連続する3点リーダをつめる
    sentence = re.sub(r'…+', "…", sentence)

    # 連続する読点をつめる
    sentence = re.sub(r'、+', "、", sentence)

    # 連続する「！？」「？！」をつめる
    sentence = re.sub(r'(！？)+', "！？", sentence)
    sentence = re.sub(r'(？！)+', "？！", sentence)

    # 連続する絵文字をつめる
    sentence = prepare_emoji(sentence)

    # 「（」「）」はすべて半角に
    sentence = sentence.replace("（", "(")
    sentence = sentence.replace("）", ")")

    # スペースすべて削除
    sentence = sentence.replace(" ", "")
    sentence = sentence.replace("　", "")

    # 改行を半角スペースに変換
    sentence = sentence.replace("\r\n", " ")
    sentence = sentence.replace("\r", " ")
    sentence = sentence.replace("\n", " ")

    # 句点をすべて半角スペースに変換
    sentence = sentence.replace("。", " ")

    # 「！」「？」に半角スペースをつける
    sentence = sentence.replace("！", "！ ")
    sentence = sentence.replace("？", "？ ")

    # 「！w」から半角スペースを消す
    sentence = sentence.replace("！ w", "！w ")

    # 「！？」「？！」から半角スペースを消す
    sentence = sentence.replace("！ ？ ", "！？ ")
    sentence = sentence.replace("？ ！ ", "？！ ")

    # 「)」前から半角スペースを消す
    sentence = sentence.replace(" )", ")")

    # 「」」前から半角スペースを消す
    sentence = sentence.replace(" 」", "」")

    # 「w」の前から半角スペースを消す
    sentence = sentence.replace(" w", "w")

    # 連続するスペースをつめる
    sentence = re.sub(r' +', " ", sentence)
    sentence = re.sub(r'　+', " ", sentence)

    # 全角空白削除
    sentence = sentence.replace("ㅤ", "")

    # 余白削除
    sentence = sentence.strip()

    # ペアになってないカッコを修正
    sentence = prepare_pair(sentence, '(', ')')
    sentence = prepare_pair(sentence, '[', ']')
    sentence = prepare_pair(sentence, '「', '」')
    sentence = prepare_pair(sentence, '『', '』')
    sentence = prepare_pair(sentence, '【', '】')
    sentence = prepare_pair(sentence, '［', '］')

    # AAっぽいツイートは除外
    if sentence.find('＿＿') >= 0 or sentence.find('人人人') >= 0 \
        or sentence.find('|') >= 0 or sentence.find('｜') >= 0 or sentence.find('／') >= 0 or sentence.find('＼') >= 0 \
        or sentence.find('┏') >= 0 or sentence.find('┓') >= 0 or sentence.find('┗') >= 0 or sentence.find('┛') >= 0 \
        or sentence.find('┳') >= 0 or sentence.find('┻') >= 0 or sentence.find('━') >= 0 or sentence.find('┃') >= 0:
        return ''

    return sentence

def chunk(chunk):
    # 余白削除
    chunk = chunk.strip()

    # ペアになってないカッコを修正
    chunk = prepare_pair(chunk, '(', ')')
    chunk = prepare_pair(chunk, '[', ']')
    chunk = prepare_pair(chunk, '「', '」')
    chunk = prepare_pair(chunk, '『', '』')
    chunk = prepare_pair(chunk, '【', '】')
    chunk = prepare_pair(chunk, '［', '］')

    # 先頭と末尾にあるノイズを除去
    chunk = chunk \
        .strip('←').strip('↑').strip('→').strip('↓').strip('⇐').strip('⇑').strip('⇒').strip('⇓') \
        .strip('>>').strip('.').strip('、').strip('゛').strip('*').strip('“')

    # 絵文字以外の１文字は除外
    if len(chunk) == 1 and chunk[0] not in emoji.UNICODE_EMOJI:
        return ''

    return chunk

def prepare_emoji(text):
    prepared_text = ""
    prev_emoji = ''
    for char in list(text):
        if char in emoji.UNICODE_EMOJI:
            if prev_emoji == char:
                continue
            prev_emoji = char
        prepared_text += char
    return prepared_text

def prepare_pair(text, start_char, end_char):
    start_pos = 0
    end_pos = 0
    hit_flag = False
    prepared_text = ""
    for char in list(text):
        if char == start_char:
            if hit_flag:
                start_pos += 1
                prepared_text += text[start_pos:end_pos]
                start_pos = end_pos
            else:
                prepared_text += text[start_pos:end_pos]
                start_pos = end_pos
            hit_flag = True
        elif char == end_char:
            if hit_flag:
                prepared_text += text[start_pos:end_pos]
                start_pos = end_pos
            else:
                prepared_text += text[start_pos:end_pos]
                start_pos = end_pos + 1
            hit_flag = False
        end_pos += 1
    prepared_text += text[start_pos:end_pos]
    if hit_flag:
        prepared_text += end_char
    return prepared_text

if __name__ == '__main__':
    tweet = "君がッ！！！泣くまで！！！！殴るのをやめないッ！！！！！」"
    reply = "「このきたならしい阿呆がァーーーーーーッ！！"

    tweet = sentence(tweet)
    reply = sentence(reply)

    tweet_chunks = tweet.split()
    reply_chunks = reply.split()

    for tweet_chunk in tweet_chunks:
        for reply_chunk in reply_chunks:
            tweet_chunk = chunk(tweet_chunk)
            reply_chunk = chunk(reply_chunk)

            if tweet_chunk == '' or reply_chunk == '':
                continue

            print("%(tweet_chunk)s -> %(reply_chunk)s" % {'tweet_chunk': tweet_chunk, 'reply_chunk': reply_chunk})

君がッ！ -> 「このきたならしい阿呆がァーッ！」
泣くまで！ -> 「このきたならしい阿呆がァーッ！」
殴るのをやめないッ！ -> 「このきたならしい阿呆がァーッ！」

今後の予定

• 顔文字を崩さないようにする
• 文節で切ってさらに水増しさせる

参考サイト

pythonで絵文字を駆逐する