Share
Embed
Transcript
50 slides
-
จϕΫτϧ: ࣗવݴޠจͷϕΫτϧදݱ 4 จϕΫτϧۭؒ ͜Ͳ͕Ոʹ͔͍ͬͯΔɻ ͜Ͳֶ͕ߍ͔ΒՈʹ͔͍ͬͯΔɻ ͜Ͳ͕ਤॻؗʹ͍Δɻ ͜Ͳ͕ޕޙʹา͍͍ͯΔɻ จಉ࢜ͷҙຯ͕͍ۙ
ϕΫτϧಉ͕͍࢜ۙ • ྨࣅจݕࡧ • ΫϥελϦϯά ɹͳͲ෯͍Ԡ༻ -
จϕΫτϧ: ࣗવݴޠจͷϕΫτϧදݱ 5 จϕΫτϧۭؒ ͜Ͳ͕Ոʹ͔͍ͬͯΔɻ ͜Ͳֶ͕ߍ͔ΒՈʹ͔͍ͬͯΔɻ ͜Ͳ͕ਤॻؗʹ͍Δɻ ͜Ͳ͕ޕޙʹา͍͍ͯΔɻ จಉ࢜ͷҙຯ͕͍ۙ
ϕΫτϧಉ͕͍࢜ۙ • ྨࣅจݕࡧ • ΫϥελϦϯά ɹͳͲ෯͍Ԡ༻ • ϕΫτϧͷ্࣭ • ϕΫτϧͷੑ࣭ཧղ ɹ͕༗༻ੑ্ʹ݁ จϕΫτϧʹ͍ͭͯͷཧղΛਂΊΔ ͜ͱ͕ࠓޙͷൃలͷͨΊʹॏཁ -
طଘݚڀͷ՝ จϕΫτϧݚڀͷݱঢ় •Ұ෦ͷϕϯνϚʔΫλεΫͰͷੑೳධՁ͕ओ •ͦΕͧΕͷख๏ͷཧղͦ͜·ͰਐΜͰ͍ͳ͍ ࣮Ԡ༻ͱͷΪϟοϓ •จϕΫτϧʹٻΊΒΕΔੑ࣭Ԡ༻ࣄྫ͝ͱʹมԽ͢Δ • ݕࡧͰจҙΑΓτϐοΫΧςΰϦ͕ॏཁ • ࣭ԠͰ࣭ͱճ͕ۙ͘ʹ͢Δ͜ͱ͕ॏཁ
•Ԡ༻ࣄྫ͝ͱʹదͳੑ࣭Λ࣋ͭจϕΫτϧΛ͍͍ͨ •ʮख๏͝ͱʹͲͷΑ͏ͳੑ࣭͕͋Δ͔ʁʯۃΊͯॏཁͳ͍ 7 -
ڭࢣ৴߸ʹணͨ͠จϕΫτϧͷੳ •ػցֶशϞσϧͷৼΔ͍܇࿅σʔλɾख๏ʹେ͖͘ґଘ •ҟͳΔڭࢣ৴߸ʹج͍ͮͯߏங͞ΕͨจϕΫτϧͷੑ࣭Λੳ •ͦΕͧΕͷख๏ʹΑΔจϕΫτϧͷੑ࣭Λੳ ੑ࣭ൺֱʹదͨ͠จϕΫτϧख๏ •ྨࣅͨ͠ΞʔΩςΫνϟΛ͕࣋ͭҟͳΔڭࢣ৴߸Λ༻͍Δ •ϕϯνϚʔΫλεΫʹ͓͍ͯಉͷੑೳΛࣔ͢ ຊݚڀͰରͱ͢Δख๏ •SBERT: ࣄલֶशࡁΈݴޠϞσϧ
+ ࣗવݴޠਪλεΫ •DefSent: ࣄલֶशࡁΈݴޠϞσϧ + ఆٛจˠ୯ޠ༧ଌλεΫ 10 -
SBERT: ࣗવݴޠਪλεΫʹجͮ͘ख๏ • ࣗવݴޠਪλεΫͰ จϕΫτϧϞσϧΛ܇࿅͢Δख๏ • ࣗવݴޠਪλεΫ: จϖΞͷҙຯؔΛ༧ଌ 12 จB
จA BERT BERT ໃ६ ؚҙ ͦͷଞ ϥϕϧ༧ଌ Pooling Pooling -
SBERT: ࣗવݴޠਪλεΫʹجͮ͘ख๏ • ࣗવݴޠਪλεΫͰ จϕΫτϧϞσϧΛ܇࿅͢Δख๏ • ࣗવݴޠਪλεΫ: จϖΞͷҙຯؔΛ༧ଌ SBERTʹΑΔfine-tuningͷखॱ 0.
ࣄલֶशࡁΈݴޠϞσϧΛ༻ҙ 1. จϖΞΛͦΕͧΕจϕΫτϧʹ 2. ಘΒΕͨจϕΫτϧͷϖΞ͔Β จϖΞͷҙຯؔΛ༧ଌ 3. ਖ਼͍͠ҙຯؔΛ༧ଌͰ͖Δ Α͏ʹϞσϧΛ܇࿅ 13 จB จA BERT BERT ໃ६ ؚҙ ͦͷଞ ϥϕϧ༧ଌ Pooling Pooling -
DefSent: ఆٛจˠ୯ޠ༧ଌλεΫʹجͮ͘ख๏ • ఆٛจˠ୯ޠ༧ଌλεΫʹΑͬͯ จϕΫτϧϞσϧΛ܇࿅͢Δख๏ 14 ఆٛจ จB จA w
|V| w1 w2 w3 ... BERT BERT ໃ६ ؚҙ ͦͷଞ ϥϕϧ༧ଌ BERT ୯ޠ༧ଌ Pooling Pooling Pooling -
DefSent: ఆٛจˠ୯ޠ༧ଌλεΫʹجͮ͘ख๏ • ఆٛจˠ୯ޠ༧ଌλεΫʹΑͬͯ จϕΫτϧϞσϧΛ܇࿅͢Δख๏ DefSentʹΑΔfine-tuningͷखॱ 0. ࣄલֶशࡁΈݴޠϞσϧΛ༻ҙ 1. ఆٛจΛBERTʹೖྗͯ͠
จϕΫτϧΛ֫ಘ 2. ಘΒΕͨϕΫτϧ͔Βఆٛจ ʹରԠ͢Δ୯ޠΛ༧ଌ 3. ఆٛจ͕ද͢୯ޠͷ֬Λ ࠷େԽ͢ΔΑ͏ʹ܇࿅ 15 ఆٛจ จB จA w |V| w1 w2 w3 ... BERT BERT ໃ६ ؚҙ ͦͷଞ ϥϕϧ༧ଌ BERT ୯ޠ༧ଌ Pooling Pooling Pooling -
ຊݚڀͷ֓ཁ 16 ڭࢣ৴߸ͷҧ͍ʹ ணͨ͠จϕΫτϧͷ ൺֱɾ౷߹ SBERT DefSent BERT ؚҙؔೝࣝͰ fine-tuning
ఆٛจ→୯ޠ ༧ଌͰfine-tuning จϕΫτϧ Ϟσϧ -
ຊݚڀͷ֓ཁ 17 Semantic Textual Similarity (STS) ᶃ จͷιʔε ᶄ จϖΞͷදతྨࣅ
SentEval ᶅ ײɾ੍࣌ྨͳͲͷԼྲྀλεΫ ᶆ ݴޠֶతใͷྨλεΫ ൺֱ •SBERT→DefSent •DefSent→SBERT •ϚϧνλεΫֶश •Average •Concat ౷߹ ڭࢣ৴߸ͷҧ͍ʹ ணͨ͠จϕΫτϧͷ ൺֱɾ౷߹ SBERT DefSent BERT ؚҙؔೝࣝͰ fine-tuning ఆٛจ→୯ޠ ༧ଌͰfine-tuning จϕΫτϧ Ϟσϧ -
จϕΫτϧͷੑ࣭ൺֱ: STS 19 Semantic Textual Similarity (STS) ᶃ จͷιʔε ᶄ
จϖΞͷදతྨࣅ จϖΞ (จϕΫτϧͷϖΞ) ೖྗ ൺֱ؍ ਓؒධՁͱจϕΫτϧಉ࢜ͷ ྨࣅͱͷॱҐ૬ؔ ධՁࢦඪ STSͷධՁखॱ ᶃ จϕΫτϧϞσϧΛ༻ҙ ᶄ จϖΞͦΕͧΕΛจϕΫτϧʹม ᶅ จϕΫτϧͷϖΞͷྨࣅΛܭࢉ ᶆ ਓؒධՁͱͷ૬ؔΛܭࢉ จA จB จϕΫτϧϞσϧ ਓखධՁͱͷ ૬ؔͰධՁ จྨࣅ ᶄ ᶃ ᶅ ᶆ -
จϕΫτϧͷੑ࣭ൺֱ: STS 20 Semantic Textual Similarity (STS) ᶃ จͷιʔε ᶄ
จϖΞͷදతྨࣅ จϖΞ (จϕΫτϧͷϖΞ) ೖྗ ൺֱ؍ ਓؒධՁͱจϕΫτϧಉ࢜ͷ ྨࣅͱͷॱҐ૬ؔ ධՁࢦඪ -
จϕΫτϧͷੑ࣭ൺֱ: STS 21 Semantic Textual Similarity (STS) ᶃ จͷιʔε ᶄ
จϖΞͷදతྨࣅ จϖΞ (จϕΫτϧͷϖΞ) ೖྗ ൺֱ؍ ਓؒධՁͱจϕΫτϧಉ࢜ͷ ྨࣅͱͷॱҐ૬ؔ ධՁࢦඪ • 2ͭͷ؍ͰσʔληοτΛׂ • ੑೳͷมԽΛ؍ -
จϕΫτϧͷੑ࣭ൺֱ: STS 22 Semantic Textual Similarity (STS) ᶃ จͷιʔε ᶄ
จϖΞͷදతྨࣅ จϖΞ (จϕΫτϧͷϖΞ) ೖྗ ൺֱ؍ ਓؒධՁͱจϕΫτϧಉ࢜ͷ ྨࣅͱͷॱҐ૬ؔ ධՁࢦඪ • จͷιʔεʹΑͬͯੑೳʹࠩ • ֤ख๏ͷ܇࿅σʔληοτʹ͍ۙ จͷํ͕͏·͘ྨࣅΛଌΕΔ • 2ͭͷ؍ͰσʔληοτΛׂ • ੑೳͷมԽΛ؍ -
จϕΫτϧͷੑ࣭ൺֱ: STS 23 දతྨࣅͱੑೳͷؔ SBERT DefSent Semantic Textual Similarity (STS)
ᶃ จͷιʔε ᶄ จϖΞͷදతྨࣅ จϖΞ (จϕΫτϧͷϖΞ) ೖྗ ൺֱ؍ ਓؒධՁͱจϕΫτϧಉ࢜ͷ ྨࣅͱͷॱҐ૬ؔ ධՁࢦඪ • දతྨࣅʹΑͬͯੑೳࠩ • SBERT (ؚҙؔ)දతྨ ࣅͷӨڹΛड͚ͮΒ͍ • DefSent (ఆٛจ)දతʹྨ ࣅ͍ͯ͠ͳ͍จͷྨࣅΛ ൺֱతਖ਼͘͠ਪఆͰ͖Δ • จͷιʔεʹΑͬͯੑೳʹࠩ • ֤ख๏ͷ܇࿅σʔληοτʹ͍ۙ จͷํ͕͏·͘ྨࣅΛଌΕΔ • 2ͭͷ؍ͰσʔληοτΛׂ • ੑೳͷมԽΛ؍ -
จϕΫτϧͷੑ࣭ൺֱ: SentEval 24 SentEval ᶅ ԼྲྀλεΫ͝ͱͷੑೳ ᶆ ݴޠֶతใͷྨੑೳ จϕΫτϧ ೖྗ
ൺֱ؍ จϕΫτϧΛೖྗͱ͢Δ ઢܗྨثͷྨੑೳ ධՁࢦඪ จ ྨੑೳ͔Β จຒΊࠐΈͷ࣭ΛධՁ จϕΫτϧϞσϧ ྨث ᶄ ᶃ ᶅ ᶆ SentEvalͷධՁखॱ ᶃ จຒΊࠐΈϞσϧΛ༻ҙ ᶄ ֤จΛจϕΫτϧʹม ᶅ จϕΫτϧΛೖྗͱ͢ΔྨثΛ܇࿅ ᶆ ྨثͷੑೳ͔ΒจϕΫτϧͷ࣭ΛධՁ -
จϕΫτϧͷੑ࣭ൺֱ: SentEval 25 SentEval ᶅ ԼྲྀλεΫ͝ͱͷੑೳ ᶆ ݴޠֶతใͷྨੑೳ จϕΫτϧ ೖྗ
ൺֱ؍ จϕΫτϧΛೖྗͱ͢Δ ઢܗྨثͷྨੑೳ ධՁࢦඪ -
จϕΫτϧͷੑ࣭ൺֱ: SentEval 26 SentEval ᶅ ԼྲྀλεΫ͝ͱͷੑೳ ᶆ ݴޠֶతใͷྨੑೳ จϕΫτϧ ೖྗ
ൺֱ؍ จϕΫτϧΛೖྗͱ͢Δ ઢܗྨثͷྨੑೳ ධՁࢦඪ • SBERTҙຯతͳใΛ๛ ʹຒΊࠐΜͰ͍Δ • DefSentදతใ͕๛ • ϑϨʔζͷߏಘҙ -
Length WordContent Tense SubjNumber จϕΫτϧͷੑ࣭ൺֱ: SentEval 27 SentEval ᶅ ԼྲྀλεΫ͝ͱͷੑೳ
ᶆ ݴޠֶతใͷྨੑೳ จϕΫτϧ ೖྗ ൺֱ؍ จϕΫτϧΛೖྗͱ͢Δ ઢܗྨثͷྨੑೳ ධՁࢦඪ • SBERTҙຯతͳใΛ๛ ʹຒΊࠐΜͰ͍Δ • DefSentදతใ͕๛ • ϑϨʔζͷߏಘҙ • DefSent੍࣌จத୯ޠͷใ ͳͲදతͳใ͕ൺֱత๛ 50 60 70 80 90 Length WordContent Tense จ༧ଌ 50 60 70 80 90 Length WordContent Tense จத୯ޠ༧ଌ ੍࣌༧ଌ • SBERTจͷදใ͕ॏཁͳλεΫ ͷੑೳ͕͍ • จத୯ޠͳͲͷใগͳΊ -
จϕΫτϧͷੑ࣭ൺֱ: ·ͱΊ 28 Semantic Textual Similarity (STS) ᶃ จͷιʔε ᶄ
จϖΞͷදతྨࣅ จϖΞ (จϕΫτϧͷϖΞ) ೖྗ ൺֱ؍ ਓؒධՁͱจϕΫτϧಉ࢜ͷ ྨࣅͱͷॱҐ૬ؔ ධՁࢦඪ SentEval ᶅ ԼྲྀλεΫ͝ͱͷੑೳ ᶆ ݴޠֶతใͷྨੑೳ จϕΫτϧ ೖྗ ൺֱ؍ จϕΫτϧΛೖྗͱ͢Δ ઢܗྨثͷྨੑೳ ධՁࢦඪ -
จϕΫτϧͷੑ࣭ൺֱ: ·ͱΊ 29 දతྨࣅͱੑೳͷؔ SBERT DefSent Semantic Textual Similarity (STS)
ᶃ จͷιʔε ᶄ จϖΞͷදతྨࣅ จϖΞ (จϕΫτϧͷϖΞ) ೖྗ ൺֱ؍ ਓؒධՁͱจϕΫτϧಉ࢜ͷ ྨࣅͱͷॱҐ૬ؔ ධՁࢦඪ SentEval ᶅ ԼྲྀλεΫ͝ͱͷੑೳ ᶆ ݴޠֶతใͷྨੑೳ จϕΫτϧ ೖྗ ൺֱ؍ จϕΫτϧΛೖྗͱ͢Δ ઢܗྨثͷྨੑೳ ධՁࢦඪ SBERT • λεΫ: ࣗવݴޠਪ • ײۃੑͳͲҙຯతใ͕๛ • දతใগͳΊ DefSent • λεΫ: ఆٛจˠ୯ޠ༧ଌ • จ੍࣌ͳͲදతใ͕๛ • ϑϨʔζͷߏಘҙ -
จϕΫτϧͷ౷߹ ੑ࣭ൺֱͷ݁Ռ •ಉ͡ϞσϧΛϕʔεͱ͠ɺҟͳΔڭࢣ৴߸Λ༻͍Δख๏Λൺֱ • SBERTͱDefSentҟͳΔੑ࣭Λ࣋ͭ͜ͱ͕Θ͔ͬͨ •ҟͳΔৼΔ͍Λ͢ΔϞσϧͷΞϯαϯϒϧ༗ͳखཱͯ • จϕΫτϧʹ͓͍ͯҟͳΔϞσϧͷ౷߹༗༻͔ʁ ҟͳΔੑ࣭Λ࣋ͭจϕΫτϧͷ౷߹ •5ͭͷ౷߹ख๏ʹ͍࣮ͭͯݧ
• S+D, D+S • Multi • Average • Concat 32 • ౷߹ʹΑͬͯੑೳ্͢Δ͔ʁ • ࠓޙͷจϕΫτϧݚڀʹ͓͍ͯෳ ͷڭࢣ৴߸ͷΈ߹Θͤ༗͔ʁ -
จϕΫτϧͷ౷߹: ୯ҰϞσϧͷ౷߹ 33 • S+D: ୯ҰϞσϧʹSBERT, DefSentʹΑΔfine-tuningΛॱʹ࣮ࢪ • D+S: ୯ҰϞσϧʹDefSent,
SBERTʹΑΔfine-tuningΛॱʹ࣮ࢪ • Multi: SBERTͱDefSentʹΑΔfine-tuningΛަޓʹ࣮ࢪ -
จϕΫτϧͷ౷߹: ୯ҰϞσϧͷ౷߹ 34 • S+D: ୯ҰϞσϧʹSBERT, DefSentʹΑΔfine-tuningΛॱʹ࣮ࢪ • D+S: ୯ҰϞσϧʹDefSent,
SBERTʹΑΔfine-tuningΛॱʹ࣮ࢪ • Multi: SBERTͱDefSentʹΑΔfine-tuningΛަޓʹ࣮ࢪ -
จϕΫτϧͷ౷߹: ୯ҰϞσϧͷ౷߹ 35 • S+D: ୯ҰϞσϧʹSBERT, DefSentʹΑΔfine-tuningΛॱʹ࣮ࢪ • D+S: ୯ҰϞσϧʹDefSent,
SBERTʹΑΔfine-tuningΛॱʹ࣮ࢪ • Multi: SBERTͱDefSentʹΑΔfine-tuningΛަޓʹ࣮ࢪ -
จϕΫτϧͷ౷߹: ෳϞσϧͷ౷߹ 36 • Average: SBERT, DefSentʹΑΔจϕΫτϧΛฏۉ • Concat: SBERT,
DefSentʹΑΔจϕΫτϧΛ࿈݁ -
จϕΫτϧͷ౷߹: ෳϞσϧͷ౷߹ 37 • Average: SBERT, DefSentʹΑΔจϕΫτϧΛฏۉ • Concat: SBERT,
DefSentʹΑΔจϕΫτϧΛ࿈݁ -
จϕΫτϧͷ౷߹: ධՁ࣮ݧ 38 •౷߹ख๏͝ͱʹϞσϧΛ܇࿅ɾධՁ •STSͰ10ճ, SentEvalͰ3ճϞσϧΛ ܇࿅ͯ͠ฏۉੑೳΛใࠂ ࣮ݧઃఆ ධՁର •SBERT
•DefSent •S+D (SBERT→DefSent) •D+S (DefSent→SBERT) •Multi •Average •Concat ୯Ұख๏ͱੑೳΛൺֱ ධՁλεΫ •STS •SentEval -
จϕΫτϧͷ౷߹: ධՁ࣮ݧ 39 BERT-base STS SentEval SBERT 73.19 86.49 DefSent
75.20 86.61 S+D 78.45 86.80 D+S 72.89 86.09 Multi 72.89 86.23 Average 77.82 87.47 Concat 76.03 87.93 ֤౷߹ख๏͝ͱͷSTSͱSentEvalͷฏۉੑೳ (%) SBERT→DefSentͱ Average͕ߴੑೳ • SentEvalͰConcatͷੑೳ͕ ྑ͍͕ɺจϕΫτϧͷ࣍ݩ͕ େ͖͘༗རͳͷͰҙ -
จϕΫτϧͷ౷߹: ධՁ࣮ݧ 40 BERT-base STS SentEval SBERT 73.19 86.49 DefSent
75.20 86.61 S+D 78.45 86.80 D+S 72.89 86.09 Multi 72.89 86.23 Average 77.82 87.47 Concat 76.03 87.93 ֤౷߹ख๏͝ͱͷSTSͱSentEvalͷฏۉੑೳ (%) • ౷߹ख๏͕୯Ұख๏ ΛԼճΔ߹ • ഁ໓త٫ͷӨڹ͔ SBERT→DefSentͱ Average͕ߴੑೳ DefSent→SBERT ੑೳ͕ѱԽ • SentEvalͰConcatͷੑೳ͕ ྑ͍͕ɺจϕΫτϧͷ࣍ݩ͕ େ͖͘༗རͳͷͰҙ -
จϕΫτϧͷ౷߹: ධՁ࣮ݧ 41 BERT-base STS SentEval SBERT 73.19 86.49 DefSent
75.20 86.61 S+D 78.45 86.80 D+S 72.89 86.09 Multi 72.89 86.23 Average 77.82 87.47 Concat 76.03 87.93 ֤౷߹ख๏͝ͱͷSTSͱSentEvalͷฏۉੑೳ (%) • SentEvalͰConcatͷੑೳ͕ ྑ͍͕ɺจϕΫτϧͷ࣍ݩ͕ େ͖͘༗རͳͷͰҙ • ౷߹ख๏͕୯Ұख๏ ΛԼճΔ߹ • ഁ໓త٫ͷӨڹ͔ • จϕΫτϧͷ୯७ฏۉ͕Α͍ੑೳ • ౷߹ख๏ʹΑΔϕΫτϧͷੑ࣭ੳ ࠓޙͷ՝ SBERT→DefSentͱ Average͕ߴੑೳ DefSent→SBERT ੑೳ͕ѱԽ -
·ͱΊɾࠓޙͷ՝ 42 ౷߹ ڭࢣ৴߸ͷҧ͍ʹண͠จϕΫτϧͷ ੑ࣭Λൺֱੳɾ౷߹ ൺֱ • จͷιʔεʹΑΔख๏͝ͱͷੑೳ͕ࠩݦஶ SBERT •
ײۃੑͳͲҙຯతใ͕๛ • දతྨࣅͷӨڹΛड͚ͮΒ͍ DefSent • จ੍࣌ͳͲදతใ͕๛ • ϑϨʔζͷߏಘҙ • දతྨࣅ͕͍จϖΞʹڧ͍ • ౷߹ʹΑͬͯੑೳ্ • SBERT→DefSent Average͕ߴੑೳ • ഁ໓త٫ͷӨڹͰੑ ೳ͕Լ͢Δ߹ ੳର ɹSBERT: ࣗવݴޠਪϕʔε ɹDefSent: ఆٛจ→୯ޠ༧ଌϕʔε ࠓޙͷ՝ 1. ΑΓ൚ͳϞσϧɾจϕΫτϧख๏ͷௐࠪ 2. ౷߹ख๏Ͱߏ͞ΕͨϕΫτϧͷੑ࣭ੳ 3. ΑΓΑ͍౷߹ख๏ͷ։ൃ -
·ͱΊɾࠓޙͷ՝ 43 ౷߹ ڭࢣ৴߸ͷҧ͍ʹண͠จϕΫτϧͷ ੑ࣭Λൺֱੳɾ౷߹ ൺֱ • จͷιʔεʹΑΔख๏͝ͱͷੑೳ͕ࠩݦஶ SBERT •
ײۃੑͳͲҙຯతใ͕๛ • දతྨࣅͷӨڹΛड͚ͮΒ͍ DefSent • จ੍࣌ͳͲදతใ͕๛ • ϑϨʔζͷߏಘҙ • දతྨࣅ͕͍จϖΞʹڧ͍ • ౷߹ʹΑͬͯੑೳ্ • SBERT→DefSent Average͕ߴੑೳ • ഁ໓త٫ͷӨڹͰੑ ೳ͕Լ͢Δ߹ ੳର ɹSBERT: ࣗવݴޠਪϕʔε ɹDefSent: ఆٛจ→୯ޠ༧ଌϕʔε ࠓޙͷ՝ 1. ΑΓ൚ͳϞσϧɾจϕΫτϧख๏ͷௐࠪ 2. ౷߹ख๏Ͱߏ͞ΕͨϕΫτϧͷੑ࣭ੳ 3. ΑΓΑ͍౷߹ख๏ͷ։ൃ -
ത࢜ޙظ՝ఔͷల •จϕΫτϧͷকདྷతͳൃలͷͨΊͷॏཁͳ՝ • طଘख๏จͷݶΒΕͨଆ໘ʹ͔͠ண͍ͯ͠ͳ͍ • طଘͷࣄલֶशࡁΈݴޠϞσϧจϕΫτϧͷදݱྗ͕ෆ ɺܭࢉྔେ͖͍ •͋ΒΏΔจϕΫτϧख๏ͷج൫ͱͯ͠༻͍Δ͜ͱ͕Ͱ͖Δɺ൚ ༻ੑʹ༏ΕͨϞσϧ͕ඞཁ จϕΫτϧͷͨΊͷج൫Ϟσϧͷ։ൃ
•ݚڀ1: จϕΫτϧͷੑ࣭ੳ •ݚڀ2: จϕΫτϧʹ͓͚Δج൫ϞσϧͷఏҊ •ݚڀ3: จϕΫτϧʹ͓͚Δج൫ϞσϧͷԠ༻ 45 -
ݚڀۀ ࠃจࢽ (ࠪಡ͋Γ) • ௩ӽॣ, ྒྷฏ, ాߒҰ. ఆٛจΛ༻͍ͨจຒΊࠐΈߏ๏, ࣗવݴޠॲཧ Vol.
30 No. 1 (ൃߦ༧ఆ). ࠃࡍձٞ (ࠪಡ͋Γ) • Hayato Tsukagoshi, Ryohei Sasano, Koichi Takeda. Comparison and Combination of Sentence Embeddings Derived from Different Supervision Signals, in Proceedings of the 11th Joint Conference on Lexical and Computational Semantics (*SEM 2022). • Hayato Tsukagoshi, Ryohei Sasano, Koichi Takeda. DefSent: Sentence Embeddings using Definition Sentences, in Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (ACL-IJCNLP 2021). ࠃձٞ (ࠪಡͳ͠) • ཅాᠳฏ, ௩ӽॣ, ྒྷฏ, ాߒҰ. ΨεຒΊࠐΈʹجͮ͘จදݱੜ, ݴޠॲཧֶձ ୈ29ճ࣍େձ (NLP2023) ൃද༧ఆ. • ௩ӽॣ, ฏඌ, Լກ, ࠤࠀݾ, ྒྷฏ, ాߒҰ. ࣗવݴޠਪͱ࠶ݱثΛ༻͍ͨSplit and Rephraseʹ ͓͚Δੜจͷ্࣭, ݴޠॲཧֶձ ୈ28ճ࣍େձ (NLP2022). • ௩ӽॣ, ྒྷฏ, ాߒҰ. ఆٛจΛ༻͍ͨจຒΊࠐΈߏ๏, ݴޠॲཧֶձ ୈ27ճ࣍େձ (NLP2021). ͦͷଞ • 2023 ຊֶज़ৼڵձ ಛผݚڀһ-DC1 ࠾༻ఆ • 2023 ໊ݹେֶ༥߹ϑϩϯςΟΞϑΣϩʔೝఆ 50