【論文紹介】Language in Our Time An Empirical Analysis of Hashtags / introduce-language-in-our-time-an-empirical-analysis-of-hashtags

2019年5月24日 WEBエンジニア勉強会 #13 (https://web-engineer-meetup.connpass.com/event/128855/)での発表資料です。

WWW2019 の 『Language in Our Time: An Empirical Analysis of Hashtags』 (https://arxiv.org/abs/1905.04590) という論文についてさらっと紹介させていただきました。👀

Transcript

1. ʲ࿦จ঺հʳLanguage in Our Time: An Empirical Analysis of Hashtags - Instagram ʹ͓͚Δ #ϋογϡλά ͷ෼ੳ- WEBΤϯδχΞษڧձ #13 Yuya Matsumura (@yu-ya4) 24 May 2019
2. ✓ Yuya Matsumura ✓ Software Engineer ✓ Wantedly, Inc. Recommendation Team ✓ Interested in Information Retrieval, Machine Learning Self-Introduction @yu-ya4 @yu__ya4
3. ✓ Webٕज़શൠʹؔ͢Δ࠷΋ݖҖͷ͋Δࠃࡍձٞ ✓ Google, Baidu, Amazon, facebook…. ͳͲͷ໊ͩͨΔεϙϯαʔ ✓ ࠾୒཰͸ WWW 2018 Ͱ 14.8% (171/1155) ✓ 2019 ೥͸ 5/13 - 5/17 ʹαϯϑϥϯγείͰ։࠵͞Εͨ ✓ ݕࡧ΍ηΩϡϦςΟɼػցֶश͔Βࣾձ໰୊·Ͱ༷ʑͳτϐοΫ The Web Conference (WWW) 5IF
   8FC
   $POGFSFODF
   
   IUUQTXXXUIFXFCDPOGPSH
   
4. ✓ Who Watches the Watchmen: Exploring Complaints on the Web ✓ The Music Streaming Sessions Dataset ✓ Blockchain Mining Games with Pay-Forward ✓ GhostLink: Mining Latent Influence Networks for Influence-aware Item Recommendation ✓ Language in Our Time: An Empirical Analysis of Hashtags Examples of sessions IUUQTXXXUIFXFCDPOGPSHBDDFQUFEQBQFST
5. ✓ Who Watches the Watchmen: Exploring Complaints on the Web ✓ The Music Streaming Sessions Dataset ✓ Blockchain Mining Games with Pay-Forward ✓ GhostLink: Mining Latent Influence Networks for Influence-aware Item Recommendation ✓ Language in Our Time: An Empirical Analysis of Hashtags Examples of sessions IUUQTXXXUIFXFCDPOGPSHBDDFQUFEQBQFST ˠ͜ͷ࿦จͷ಺༰ʹ͍ͭͯ؆୯ʹ͝঺հ
6. ✓ ֓ཁΛ͞Βͬͱ঺հ͠·͢ (ৄࡉͳ࣮ݧઃఆ౳ʹ͸ݴٴ͠·ͤΜ) ✓ ͋͘·Ͱɼࢲ͕࿦จΛಡΜ্ͩͰͷࢲͷղऍͰ͢ ✓ ؒҧͬͨ͜ͱݴͬͯͨΒ͝ΊΜͳ͍͞ ✓ ൃද಺༰ͱॴଐ૊৫ͱ͸ؔ܎͍͟͝·ͤΜ (ΦωΦ) Excuse
7. Language in Our Time: An Empirical Analysis of Hashtags IUUQTBSYJWPSHBCT "VUIPST
   :BOH
   ;IBOH 1SPDFFEJOH
   888
   
   5IF
   8PSME
   8JEF
   8FC
   $POGFSFODF
   Pages 2378-2389
8. എܠʢࡶʹʣ ͜͜5೥΄ͲͰΦϯϥΠϯιʔγϟϧωοτϫʔΫʢOSNsʣʹ͓͚Δ ϋογϡλάͷਓؾɾॏཁੑ͸ͱͯ΋େ͖͍΋ͷʹͳͬͨɻ ͜Ε·Ͱͷϋογϡλάʹ͍ͭͯͷݚڀ͸ओʹ Twitter ͷσʔλΛ༻͍ͨ΋ͷͰ͋ͬͨɻ Instagram ͷϋογϡλάΛ෼ੳ͠Α͏ʂ
9. Instagram ͷେن໛σʔληοτΛ࡞੒ͯ͠ 3 ͭͷҟͳΔ؍఺͔Βϋογϡλάʹ͍ͭͯ෼ੳͨ͠ ͜ͷ࿦จͰ΍͍ͬͯΔ͜ͱΛҰݴͰʂʂ
10. ✓ 2010೥ͷऴΘΓ͔Β2015೥ͷऴΘΓ·Ͱͷ໿5೥෼ͷσʔλΛऩू ✓ New York, Los Angeles, London ʹҐஔ৘ใ͖ͭͰ౤ߘͨ͜͠ͱͷ͋Δ 51,527 usersʢ౤ߘ਺ͳͲͰ଍੾Γ͋Γʣ ✓ more than 39 million posts, more than 7 million hashtags Instagram ͷେن໛σʔλɾηοτ
11. ౤ߘճ਺ TOP10 ͷϋογϡλά ✓ #nyc ΍ #london ͳͲͷ஍Ҭಛ༗ͷ΋ͷ΋ଟ͘ݟΒΕΔ ✓ ҰํͰɼ#love ͱ͔ #artɼ#travel Έ͍ͨͳ general ͳϋογϡλά΋ ͨ͘͞ΜݟΒΕΔͷͰσʔληοτͱͯ͠໰୊ͳͦ͞͏
12. ϋογϡλάͷγΣΞ͞Εͨճ਺ͱɼϋογϡλά෇͖ͷ౤ߘΛͨ͠Ϣʔβͷਓ਺ͷ෼෍ ΄ͱΜͲ͸਺ճఔ౓͔͠ ౤ߘ͞Ε͍ͯͳ͍ ͨ͘͞ΜγΣΞ͍ͯ͠Δ Ϣʔβ͸গͳ͍
13. ϋογϡλά෇͖ͷ౤ߘͱɼ౤ߘΛͨ͠Ϣʔβͷׂ߹ͷ࣌ؒมԽ 2% 70% 97% 18%
14. 1. What are the temporal and spatial patterns of hashtags?ʢ࣌ؒ΍৔ॴͱϋογϡλάͷؔ܎ʣ 2. Do hashtags exhibit semantic displacement? ʢϋογϡλάͷҙຯͷมԽʣ 3. Can hashtags be used to infer social relations?ʢϋογϡλάΛ༻͍ͨιʔγϟϧͳ༑ୡؔ܎ͷਪఆʣ 3 ͭͷ؍఺ʢResearch Questionsʣ
15. 1. What are the temporal and spatial patterns of hashtags?ʢ࣌ؒ΍৔ॴͱϋογϡλάͷؔ܎ʣ 2. Do hashtags exhibit semantic displacement? ʢϋογϡλάͷҙຯͷมԽʣ 3. Can hashtags be used to infer social relations?ʢϋογϡλάΛ༻͍ͨιʔγϟϧͳ༑ୡؔ܎ͷਪఆʣ 3 ͭͷ؍఺ʢResearch Questionsʣ
16. ✓ ʮ#newyearʯͳͲͷॕ೔ʹؔ܎͢ΔΑ͏ͳϋογϡλά͸पظతʹ ਓؾʹͳΔɻ ✓ λάͷछྨʹΑͬͯมΘΔਓؾ౓ͷ࣌ؒతมԽͷҧ͍Λ෼ੳ͍ͨ͠ɻ ✓ ϋογϡλάΛΫϥελϦϯάͯ͠ 4 ͭͷάϧʔϓʹ෼ྨͰ͖ͨʂ ✓ શγΣΞ਺ʹରͯ͠γΣΞ͞ΕΔճ਺ͷׂ߹ͷ࣌ؒతมԽΛݟͯΈΔ ϋογϡλάͷਓؾ౓ͷมԽͱ࣌ؒͷؔ܎
17. Temporal Patterns ʢ࣌ؒͱͷؔ܎ʣ ࠷ॳ૿͑ͨޙ҆ఆ نଇతʹมԽ ͣͬͱ૿͑ଓ͚Δ ҰॠͲʔΜͱ૿͑Δ
18. ✓ Ͳ͏͍͏৔ॴʹ͍ΔϢʔβ͸ϋογϡλά෇͖౤ߘΛ ΑΓଟ͘͢Δ܏޲ʹ͋Δͷ͔ʁͦͷٯ͸ʁ ✓ ౤ߘʹඥ෇͚ΒΕͨҐஔ৘ใͱFoursquare ͷ৔ॴͷ ΧςΰϦ৘ใΛར༻ͯ͠෼ੳͯ͠Έͨ Ϣʔβͷ͍Δ৔ॴʹΑΔϋογϡλά෇͖౤ߘͷ͠΍͢͞
19. ΧςΰϦ͝ͱͷ๚໰ऀ਺ׂ߹ʢVisitʣͱϋογϡλά౤ߘ਺ׂ߹ʢHashtagsʣ ✓ Bar ͳͲͰ͸๚໰ͯ͠΋͋·ΓϋογϡλάΛγΣΞ͠ͳ͍ ✓ ެԂͳͲͰ͸ϋογϡλάΛγΣΞ͕ͪ͠
20. 1. What are the temporal and spatial patterns of hashtags?ʢ࣌ؒ΍৔ॴͱϋογϡλάͷؔ܎ʣ 2. Do hashtags exhibit semantic displacement? ʢϋογϡλάͷҙຯͷมԽʣ 3. Can hashtags be used to infer social relations?ʢϋογϡλάΛ༻͍ͨιʔγϟϧͳ༑ୡؔ܎ͷਪఆʣ 3 ͭͷ؍఺ʢResearch Questionsʣ
21. ✓ ୯ޠͷҙຯͱ͍͏ͷ͸มԽ͍ͯ͘͠΋ͷɻ ✓ ϋογϡλάͷҙຯ΋࣌ؒͷܦաͱͱ΋ʹมԽ͍ͯ͘͠ͷͰ͸ͳ͍͔ʁ ✓ ΈΜͳେ޷͖ Word2Vec Ͱ୯ޠΛϕΫτϧʹม׵ͯ͠ྨࣅ౓Λଌఆ ✓ ظؒʢ1೥ʣ͝ͱʹϋογϡλάΛϕΫτϧʹม׵ʢ1 ͭͷϋογϡλάʹ͖ͭ࠷ େ 5 ͭͷϕΫτϧ͕Ͱ͖Δʣ͠ɼ࣌ؒͷมԽͱͱ΋ʹҙຯ͕มΘ͍ͬͯΔͷ͔ݕূ ϋογϡλάͷҙຯͷมԽ
22. ࣅ͍ͯΔ୯ޠͷྫ
23. ҙຯͷมԽ͕େ͖͔ͬͨϋογϡλάͷྫ
24. ҙຯͷมԽ͕খ͔ͬͨ͞ϋογϡλάͷྫ
25. ✓ ҰൠతʹɼΑ͘࢖ΘΕΔ୯ޠͷҙຯ͸มΘΓʹ͍͘ͱݴΘΕΔɻ ✓ ϋογϡλάʹ͓͍ͯ͸ͦ͏Ͱ͸ͳ͔ͬͨʢ૬ؔ܎਺ -0.2ʣɻ ✓ ϋογϡλάͷ࢖ΘΕํͷ͹Β͖ͭΛܭࢉ͢ΔϋογϡλάΤϯτϩϐʔͱ͍͏ ΋ͷΛఏҊ ✓ ͨ͘͞ΜͷϢʔβʹಉ͘͡Β͍ͷස౓Ͱ࢖ΘΕ͍ͯΔ΄Ͳখ͘͞ͳΔɻ ✓ ϋογϡλάΤϯτϩϐʔͱϋογϡλάͷҙຯͷมԽͷؔ܎Λ෼ੳɻ Ͳ͏͍͏ϋογϡλάͷҙຯ͕มԽ͠΍͍͔͢ʁ
26. ϋογϡλάΤϯτϩϐʔ͕খ͍͞΄Ͳҙຯ͕มΘΓ΍͍͢ →ͨ͘͞ΜͷϢʔβʹಉ͘͡Β͍ͷස౓Ͱ࢖ΘΕ͍ͯΔϋογϡλά΄Ͳ ɹҙຯ͕มΘΓ΍͍͢
27. 1. What are the temporal and spatial patterns of hashtags?ʢ࣌ؒ΍৔ॴͱϋογϡλάͷؔ܎ʣ 2. Do hashtags exhibit semantic displacement? ʢϋογϡλάͷҙຯͷมԽʣ 3. Can hashtags be used to infer social relations?ʢϋογϡλάΛ༻͍ͨιʔγϟϧͳ༑ୡؔ܎ͷਪఆʣ 3 ͭͷ؍఺ʢResearch Questionsʣ
28. ✓ ༑ୡؔ܎ʹ͋Ε͹ಉ͡Α͏ͳϋογϡλάΛ౤ߘ͢ΔͷͰ͸ͳ͍͔ʁ ✓ ڞ௨ͷϋογϡλάͷ਺͕ଟ͚Ε͹༑ୡʁ → ͚ͬ͜͏গͳ͍͔Βݫ͍͠ ✓ User ͱ Hashtag ͷؔ܎ΛάϥϑͰද͢ ✓ Random Walk తͳΞϓϩʔνΛ༻͍ͯɼUser ΛϕΫτϧͰදݱʢྲྀߦΓ ͷ Graph Embeddingʣͯ͠ྨࣅ౓Ͱਪఆʂ ϋογϡλάΛ༻͍ͨιʔγϟϧͳ༑ୡؔ܎ͷਪఆ
29. Πϝʔδ Users Hashtags
30. Users Hashtags ༑ୡ͔΋ʂʁ Πϝʔδ
31. ༑ୡؔ܎ͷਪఆͷ࣮ݧ݁Ռ ✓ 3 ͭͷϕʔεϥΠϯͱൺֱͯ͠ߴ͍ਫ਼౓ ✓ ڞ௨ͷϋογϡλά౤ߘ͕ͳ͍༑ͩͪؔ܎΋ਪఆͰ͖ͨʂ
32. ✓ Instagram ͷσʔλΛ༻͍ͯϋογϡλάͷ෼ੳΛߦͬͨ ✓ 3 ͭͷ؍఺͔Β෼ੳ 1. ࣌ؒ΍ۭؒͱϋογϡλάͷؔ܎ੑ 2. ϋογϡλάͷҙຯͷมԽ 3. ϋογϡλάΛ༻͍ͨιʔγϟϧͳ༑ୡؔ܎ͷਪఆ ·ͱΊ
33. ✓ The Web Conference ͍ͬͯ͏ Web ܥͷΧϯϑΝϨϯε͕͋ΔΑʂ ✓ ঺հͨ͠Α͏ͳׂͱಡΈ΍͍͢࿦จ΋౤ߘ͞ΕͯΔΑʂ ✓ ීஈͷۀ຿ʹ׆͔ͤΔ͔΋ʂʁ ✓ Έͳ͞Μ΋ੋඇڵຯΛ͍͚࣋ͬͯͨͩΕ͹ʂ ఻͔͑ͨͬͨ͜ͱ