再帰型ニューラルネットを用いたCSの返答自動化の検討

カスタマーサポートで質問に自動返答できるシステムが欲しかったのでプロトタイプを作ってみました

Transcript

1. ࠶ؼܕχϡʔϥϧωοτΛ༻ ͍ͨCSͷฦ౴ࣗಈԽͷݕ౼ ాத Ժࣝ (Tanaka Yasunori)
2. About Me • ాத Ժࣝ (Tanaka Yasunori) • @yanak174 • https://blog.codingecho.com • ॴଐ: גࣜձࣾenish • ΍ͬͯΔ͜ͱ: ήʔϜΞϓϦͷαʔόʔαΠυΤϯδχΞ
3. ಺༰ 1. CS (ΧελϚʔαϙʔτ) ͷݱঢ় 2. ΍Γ͍ͨ͜ͱ 3. σʔληοτ 4. ࣭໰ςΩετͷલॲཧ 5. LSTMͰֶश 6. ·ͱΊ
4. CSͷݱঢ় • 1. ଟ͘ͷϢʔβʔ͕͢Δ࣭໰ͱɺ2. ൺֱతগͳ͍Ϣʔβʔ͕ ૺ۰͢Δ໰୊ʹର͢Δ࣭໰͕͋Δ • ଟ͘ͷϢʔβʔ͕͢Δ࣭໰ (ҎԼɺFAQ)͸ɺ͋Δఔ౓ܾΊΒΕ ͨϑΥʔϚοτʹैͬͯճ౴͞ΕΔ͜ͱ͕ଟ͍ • εϚʔτϑΥϯ޲͚ήʔϜͷதʹ͸ɺFAQʹ֘౰͢Δ࣭໰ͱ͠ ͯ՝ۚʹؔ͢Δෆ۩߹ͱΞΧ΢ϯτফࣦʹؔ͢Δ಺༰͕͋Δ • ͜ΕΒͷFAQ͸Ϣʔβʔʹରͯ͠ఆܕԽͨ͠ඞཁ߲໨ͷώΞ Ϧϯά͕ඞཁͱͳΔ
5. ΍Γ͍ͨ͜ͱ • ඞཁ߲໨ͷώΞϦϯά෦෼ʹରͯࣗ͠ಈฦ౴͕ద༻ՄೳͰ ͋Δ͔ݕ౼͢Δ
6. Data set • ՝ۚʹؔ͢Δෆ۩߹ɺΞΧ΢ϯτফࣦɺͦͷଞͷ3ͭͷΧ ςΰϦʔͷ࣭໰Λར༻͢Δ • ֤ΧςΰϦʔ͸700αϯϓϧɺ߹ܭ໿2100αϯϓϧ • ϑΝΠϧ͸TSVͰid, question, answer, categoryʹผΕͯ ͍Δ • question, categoryΛ࢖༻͢Δ
7. ࣭໰ςΩετͷલॲཧ ه߸΍ϝʔϧΞυϨεɺ਺ࣈͳͲͷจࣈΛ࡟আ͢Δ filtered_text = [] text = ["͓࣌ؒΛ௖ଷ͓ͯ͠Γ·͢ɻversion 1.2.3 ----------------------------------------"] for t in issues: result = re.compile('-+').sub('', t) result = re.compile('[0-9]+').sub('0', result) result = re.compile('\s+').sub('', result) # ... ͜ͷΑ͏ͳஔ׵ॲཧ͕ෳ਺ܨ͕͍ͬͯ·͢ # ࣭໰ςΩετ͕ۭจࣈʹͳΔ͜ͱ͕͋ΔͷͰͦͷߦ͸ؚΊͳ͍Α͏ʹ͠·͢ if len(result) > 0: sub_texts.append(result) filtered_text.append(result) print("text:%s" % result) # text:͓࣌ؒΛ௖ଷ͓ͯ͠Γ·͢ɻ
8. αϯϓϧͱϥϕϧΛ࡞੒ labels = [] samples = [] threshold = 700 cnt1 = 0 cnt2 = 0 cnt3 = 0 for i, row in enumerate(filtered_samples): if 'Account' in row[2]: if cnt2 < threashold: cnt1 += 1 labels.append(2) samples.append(row[0]) elif 'Payment' in row[2]: if cnt3 < threashold: cnt3 += 1 labels.append(3) samples.append(row[0]) else: if cnt1 < threashold: cnt1 += 1 labels.append(1) samples.append(row[0])
9. MeCabͰ෼͔ͪॻ͖ import MeCab import re def tokenize(text): wakati = MeCab.Tagger("-O wakati") wakati.parse("") words = wakati.parse(text) # Make word list if words[-1] == u"\n": words = words[:-1] return words texts = [tokenize(a) for a in samples] ௕Β͓࣌ؒ͘Λ௖ଷ͓ͯ͠Γ·͢ ௕Β͘ ͓ ࣌ؒ Λ ௖ଷ ͠ ͯ ͓Γ ·͢
10. Word embedding Deep Learning with Python: https://www.manning.com/books/deep-learning-with-python
11. LSTMͰֶश Deep Learning with Python: https://www.manning.com/books/deep-learning-with-python
12. LSTMͰֶश from keras.models import Sequential from keras.layers import Flatten, Dense, Embedding from keras.layers import LSTM model = Sequential() model.add(Embedding(15000, 100, input_length=maxlen)) model.add(LSTM(32)) model.add(Dense(4, activation='sigmoid')) model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['acc']) model.summary() Word embedding΋ಉ࣌ʹֶश͢Δ
13. Results Validation accuracy 90%
14. LSTMͷจ຺ཧղ A. ͜ͷલ͸ΞΧ΢ϯτͷҾ͖ܧ͗ͷ໰୊ղܾ͋Γ͕ͱ͏͍͟͝·ͨ͠ɻࠓճ ͷෆ۩߹͸ΞΠςϜΛߪೖͨ͠ͷʹ൓ө͞Ε·ͤΜɻ͜ͷΑ͏ͳ͜ͱ͕ଓ ͘ͷ͸൵͍͠Ͱ͢ɻ B. ͜ͷલ͸ΞΠςϜߪೖͷ໰୊ղܾ͋Γ͕ͱ͏͍͟͝·ͨ͠ɻࠓճͷෆ۩߹ ͸ΞΧ΢ϯτͷҾ͖ܧ͕͗Ͱ͖ͳ͍໰୊Ͱ͢ɻ͜ͷΑ͏ͳ͜ͱ͕ଓ͘ͷ͸ ൵͍͠Ͱ͢ɻ
15. LSTMͷจ຺ཧղ
16. LSTMͷจ຺ཧղ • Aͷ࣭໰͸՝ۚෆ۩߹ʹؔ͢Δ࣭໰Ͱ͢ɻBͷ࣭໰͸ΞΧ΢ϯτফࣦʹؔ͢ Δ࣭໰Ͱ͢ɻ͜ΕΒͷ࣭໰ʹ͸෼ྨର৅ͷΧςΰϦʔ͕ࠞࡏ͍ͯ͠·͢ɻ2 ͭͷ࣭໰ͷจ຺Ͱ͸લޙͷΧςΰϦʔΛ൓సͤ͞Δ͜ͱͰɺޓ͍ʹٯͷҙຯ ʹͳΔΑ͏ʹ͍ͯ͠·͢ɻ • Ұ൪Լͷpredictions͸0࢝·ΓͰɺ1ྻ໨͕"ͦͷଞ"ɺ2ྻ໨͕"ΞΧ΢ϯτফ ࣦ”ɺ3ྻ໨͕"՝ۚ"ʹؔ͢Δ֬৴౓Λද͍ͯ͠·͢ɻ஋͕େ͖͍΄ͲͦͷΧ ςΰϦ͕ਖ਼ղͩͱ֬৴͍ͯ͠Δ͜ͱΛҙຯ͍ͯ͠·͢ɻ·ͨɺ2ͭͷ࣭໰͸ ΧςΰϦʔͷ෼ྨʹਖ਼ղ͍ͯ͠·͢ɻ • ͜ͷ਺஋ΛݟΔͱ2ͭͷ࣭໰ͱ΋ਖ਼ղͷΧςΰϦʔʹରͯ͠ɺଞͷෆਖ਼ղͷ ΧςΰϦͱେ͖ͳࠩΛ͚ͭͯਖ਼ղͷΧςΰϦΛਖ਼ղͰ͋Δͱ֬৴͍ͯ͠Δ͜ ͱ͕Θ͔Γ·͢ɻ
17. ·ͱΊ • ະ஌ͷ࣭໰ʹରͯ͠90%ͷਫ਼౓Ͱ෼ྨͰ͖ͨ • LSTMͷจ຺ΛཧղͰ͖Δೳྗ͕෼ྨਫ਼౓ͷ޲্ʹظ଴Ͱ ͖Δ͜ͱ͕Θ͔ͬͨ • ςΩετͷલॲཧͱLSTM͕ࣗಈฦ౴ͷΞϓϩʔνͱͯ͠ ޮՌతͰ͋Δ͔΋͠Εͳ͍
18. ࠓޙͷ՝୊ A. ֶशࡁΈWord embeddingΛར༻ͨ͠৔߹ B. LSTMΛ1D-convolutional networkʹஔ͖׵͑Δ C. αϯϓϧ਺Λ૿΍͢ ਫ਼౓޲্ͷͨΊҎԼͷํ๏ͷݕূ͕ඞཁ
19. ࢀߟจݙ • Deep Learning with Python: https://www.manning.com/ books/deep-learning-with-python