國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
51
第五章 結論
總結本研究三個部分的實驗結果，可以觀察到對於品味的感知的確具有⾼
度主觀的特性，即便是透過專業的品評⽂字與風味圖譜進⾏訓練，電腦品味模 型仍無法在每個不同風味屬性都達到良好的預測結果。然⽽，普遍來說，電腦 模型預測結果與專家品評結果的相似度仍較⼀般消費者評測結果與專家品評結 果之間明顯為⾼。對於⼀些較特殊的風味感知屬性，像是許多蘇格蘭威⼠忌獨 有的 Peaty（泥煤味），電腦模型能透過學習專家品評筆記之品味⽂本與風味 感知屬性評分⽽達到相對優秀的預測效能，甚⾄可能具有近似於⼈類⼀般消費 者的感知模式，此⼀發現相當值得未來後續研究做進⼀步的探索。
⾄於⼀般消費者就品味測試的感官分析評測結果，普遍來說⼤致呼應 Ares 等⼈在 2017 年所做的相關研究結論[50]，⼀般消費者評測的準確度隨著個體 的感知能⼒差異有著很⼤的變異性。
總和以上實驗之結果，可以認定當前電腦模型對於⽂字風味的品味能⼒，
在特定領域訓練資料的輔助下，已有機會能凌駕於⼀般消費者之感知⽔平，此
⼀成果亦可提供感官分析的研究領域作為產業應用未來⽅向之參考，如能透過 電腦模型的輔助，替代⼀部分的消費者感官品評測試執⾏，除可在⼀定程度上 加強研究結果的客觀性，更能節省⼤量的時間與⾦錢成本。
在資訊科學的情感設計領域，未來若能通過自然語⾔處理之情緒分析模型 進⼀步理解⼈對於風味的⽂字解讀與表達能⼒，例如：消費者對於特殊、不討
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
喜風味的發現、偵測能⼒較強，但對於日常頻繁接觸的風味之強度評估能⼒較 弱，以及對於味覺的辨別⼒較嗅覺為佳等等命題，皆有機會透過更多的資料與 實驗來驗證。透過這些頗具潛⼒的延伸研究，或有機會在未來開發出更接近真
⼈感知⽔平的品味相關情感設計應用。
最後，在現有深度學習任務延伸應用的部分，目前透過學習Distiller 酒飲 網站資料之品味相關詞彙意義⽽產⽣的預訓練模型，目前對於現有Yelp 之餐廳 評論情緒分析任務並未帶來預期中的顯著成效提升，然⽽，考量到目前模型效 能尚不突出，未來若能持續改進模型效能、擴充對其他品味相關資料集的學習 訓練，甚⾄是進⼀步建立品味相關之專門術語字典作為輔助，預期將有機會對 於加強與嗅覺、味覺相關之分類任務成效帶來⼀定程度的幫助。
最後，本研究衷⼼盼望，此次對於品味分析研究的初步嘗試，能夠發揮拋 磚引⽟的功效，能促成未來更多與嗅覺、味覺感知屬性相關，整合資訊科學與 感官分析領域的品味分析研究，藉此對於跨領域知識科技的整合以及網際網路 數位內容資料的運用形式等應用研究發展，貢獻微薄之⼼⼒。
‧
1. Vaswani, A., et al., Attention is all you need, in Proceedings of the 31st International Conference on Neural Information Processing Systems. 2017, Curran Associates Inc.:
Long Beach, California, USA. p. 6000–6010.
2. Soleymani, M., et al., A survey of multimodal sentiment analysis. Image and Vision Computing, 2017. 65: p. 3-14.
3. Lehrer, K. and A. Lehrer, The language of taste. Inquiry, 2016. 59(6): p. 752-765.
4. Trivedi, B.P., Gustatory system: The finer points of taste. 2012. 486(7403): p. S2-S3.
5. Chiras, D.D., Human Biology. 2013: Jones & Bartlett Learning.
6. Piggott, J., Alcoholic beverages: Sensory evaluation and consumer research. 2011. 1-491.
7. Ross, C.F., Sensory science at the human–machine interface. Trends in Food Science
& Technology, 2009. 20(2): p. 63-72.
8. Ares, G., Methodological challenges in sensory characterization. Current Opinion in Food Science, 2015. 3: p. 1-5.
9. Krantz, J., Experiencing Sensation and Perception. 2012: Pearson Education, Limited.
10. Stets, J.E., Emotions and Sentiments, in Handbook of Social Psychology, J. Delamater, Editor. 2006, Springer US: Boston, MA. p. 309-335.
11. Hu, X., K. Choi, and J.S. Downie, A framework for evaluating multimodal music mood classification. Journal of the Association for Information Science and Technology, 2017. 68(2): p. 273-285.
12. Hu, X. and J. Downie, When Lyrics Outperform Audio for Music Mood Classification: A Feature Analysis. 2010. 619-624.
13. Baltrusaitis, T., C. Ahuja, and L.-P. Morency, Multimodal Machine Learning: A Survey and Taxonomy. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019. 41(2): p. 423-443.
14. Flanagin, A.J. and M.J. Metzger, Trusting expert- versus user-generated ratings online: The role of information volume, valence, and consumer characteristics.
Computers in Human Behavior, 2013. 29(4): p. 1626-1634.
15. Parikh, A.A., et al., Comparative content analysis of professional, semi-professional, and user-generated restaurant reviews. 2016: p. 1-15.
16. Minim, V.P.R., et al., Optimized Descriptive Profile: A rapid methodology for sensory description. Food Quality and Preference, 2012. 24(1): p. 190-200.
17. Murray, J.M., C.M. Delahunty, and I.A. Baxter, Descriptive sensory analysis: past, present and future. Food Research International, 2001. 34(6): p. 461-471.
18. Valentin, D., et al., Quick and dirty but still pretty good: a review of new descriptive methods in food science. International Journal of Food Science and Technology, 2012. 47(8): p. 1563-1578.
‧
19. Granitto, P.M., et al., Modern data mining tools in descriptive sensory analysis: A case study with a Random forest approach. Food Quality and Preference, 2007.
18(4): p. 681-689.
20. Tao, J. and T. Tan, Affective Computing: A Review. 2005, Springer Berlin Heidelberg.
p. 981-995.
21. Cai, G. and B. Xia, Convolutional Neural Networks for Multimedia Sentiment Analysis.
2015, Springer International Publishing. p. 159-167.
22. Purwins, H., et al., Deep Learning for Audio Signal Processing. Vol. 13. 2019.
23. 楊子萲, 應用深度學習架構於社群網路資料分析：以Twitter圖文資料為例, in
資訊科學系. 2018, 國立政治大學. p. 73.
24. 沈昱成, 基於社群媒體情感分析歸納產品屬性優缺點, in 資訊工程學系. 2016,
國立成功大學: 台南市. p. 44.
25. Zhang, L., S. Wang, and B. Liu, Deep Learning for Sentiment Analysis : A Survey. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, 2018.
26. 陳禔多, 基於歌詞文本分析技術探討音樂情緒辨識之方法研究 Exploring Music
Emotion Recognition via Textual Analysis on Song Lyrics. 2017.
27. Ortigosa-Hernández, J., et al., Approaching Sentiment Analysis by using semi-supervised learning of multi-dimensional classifiers. Neurocomputing, 2012. 92: p.
98-115.
28. Levy, O. and Y. Goldberg, Neural word embedding as implicit matrix factorization, in Proceedings of the 27th International Conference on Neural Information Processing Systems - Volume 2. 2014, MIT Press: Montreal, Canada. p. 2177–2185.
29. Firat, O., K. Cho, and Y. Bengio. Multi-Way, Multilingual Neural Machine Translation with a Shared Attention Mechanism. in Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2016. San Diego, California: Association for Computational Linguistics.
30. 李孟. 淺談神經機器翻譯 & 用 Transformer 與 TensorFlow 2 英翻中. 2019 [cited 2020 June, 9]; Available from:
https://leemeng.tw/neural-machine-translation-with-transformer-and-tensorflow2.html.
31. Devlin, J., et al. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. in Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). 2019. Minneapolis, Minnesota:
Association for Computational Linguistics.
32. 李孟. 進擊的 BERT：NLP 界的巨人之力與遷移學習. 2019 [cited 2020 June, 9];
Available from: https://leemeng.tw/attack_on_bert_transfer_learning_in_nlp.html.
33. Yang, Z., et al., XLNet: Generalized Autoregressive Pretraining for Language Understanding. 2019.
‧
35. Mehra, N.K. and S. Gupta. Survey on Multiclass Classification Methods. 2013.
36. Salman, R. and V. Kecman. Regression as classification. 2012. IEEE.
37. The Yelp Restaurant Review. [cited 2020 June, 9]; Available from:
https://www.yelp.com/dataset/.
38. Distiller. [cited 2020 March, 30]; Available from: https://distiller.com/.
39. Yu, N. and S. Kubler. Semi-supervised Learning for Opinion Detection. 2010. IEEE.
40. Banned Word List. 2009 [cited 2020 July, 7]; Available from:
http://www.bannedwordlist.com/.
41. Wine, W.F.a. Describing Food. [cited 2020 June, 9]; Available from: https://world-food-and-wine.com/describing-food.
42. Wishart, D., The flavour of whisky. 2009. 6(1): p. 20-26.
43. How to Write a Menu Describing Your Food. 2020 Feb, 11 2020 [cited 2020 July, 7];
Available from: https://www.webstaurantstore.com/article/53/how-to-write-a-menu.html.
44. Rajapakse, T. Simple Transformers — Multi-Class Text Classification with BERT, RoBERTa, XLNet, XLM, and DistilBERT. 2019 [cited 2020 June, 9]; Available from:
https://medium.com/swlh/simple-transformers-multi-class-text-classification-with-bert-roberta-xlnet-xlm-and-8b585000ce3a.
45. Transformers. 2020 [cited 2020 June, 9]; Available from:
https://huggingface.co/transformers/.
46. 3.3. Metrics and scoring: quantifying the quality of predictions. 2020 [cited 2020 2020, Aug 16]; Available from:
https://scikit-learn.org/stable/modules/model_evaluation.html#label-ranking-average-precision.
47. Afonja, T. Accuracy Paradox. 2017 Dec, 8 [cited 2020 July, 9]; Available from:
https://towardsdatascience.com/accuracy-paradox-897a69e2dd9b.
48. Gorodkin, J., Comparing two K-category assignments by a K-category correlation coefficient. Computational Biology and Chemistry, 2004. 28(5): p. 367-374.
49. Ares, G., et al., Evaluation of a rating-based variant of check-all-that-apply questions:
Rate-all-that-apply (RATA). Food Quality and Preference, 2014. 36: p. 87-95.
50. Ares, G. and P. Varela, Trained vs. consumer panels for analytical testing: Fueling a long lasting debate in the field. Food Quality and Preference, 2017. 61: p. 79-86.
51. Meyners, M., S. Jaeger, and G. Ares, On the analysis of Rate-All-That-Apply (RATA) data. Food Quality and Preference, 2015. 49.
52. Cosine similarity. 2020 2020, Aug 10 [cited 2020 2020, Aug 14]; Available from:
https://en.wikipedia.org/wiki/Cosine_similarity.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
53. 6.8. Pairwise metrics, Affinities and Kernels. 2020 [cited 2020 2020, Aug 16];
Available from: https://scikit-learn.org/stable/modules/metrics.html#cosine-similarity.
54. pingouin.mwu. 2020 [cited 2020 July, 12]; Available from: https://pingouin-stats.org/generated/pingouin.mwu.html.
55. Mann–Whitney U test. 2020 [cited 2020 July, 8]; Available from:
https://en.wikipedia.org/wiki/Mann%E2%80%93Whitney_U_test.
56. scipy.stats.spearmanr. 2020 2020, July 23 [cited 2020 2020, Aug 15]; Available from:
https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.spearmanr.html.
57. Spearman's rank correlation coefficient. 2020 2020, July 10 [cited 2020 2020, Aug 15]; Available from:
https://en.wikipedia.org/wiki/Spearman%27s_rank_correlation_coefficient.
58. Rajapakse, T. Simple Transformers — Introducing The Easiest Way To Use BERT, RoBERTa, XLNet, and XLM. 2019 [cited 2020 June, 9]; Available from:
https://towardsdatascience.com/simple-transformers-introducing-the-easiest-bert-roberta-xlnet-and-xlm-library-58bf8c59b2a3.
59. McNemar's test. 2020 2020, June 12 [cited 2020 2020, Aug 14]; Available from:
https://en.wikipedia.org/wiki/McNemar%27s_test.
60. Model: xlnet-base-cased. 2019 [cited 2020 2020, Aug 17]; Available from:
https://huggingface.co/xlnet-base-cased.