Emotion Analysis on Social Big Data

doi:10.3969/j.issn.1673-5188.2017.S2.005

摘要/Abstract

Abstract:

In this paper, we describe a method of emotion analysis on social big data. Social big data means text data that is emerging on Internet social networking services.We collect multilingual web corpora and annotated emotion tags to these corpora for the purpose of emotion analysis. Because these data are constructed by manual annotation, their quality is high but their quantity is low. If we create an emotion analysis model based on this corpus with high quality and use the model for the analysis of social big data, we might be able to statistically analyze emotional sensesand behavior of the people in Internet communications, which we could not know before. In this paper, we create an emotion analysis model that integrate the high-quality emotion corpus and the automatic-constructed corpus that we created in our past studies, and then analyze a large-scale corpus consisting of Twitter tweets based on the model. As the result of time-series analysis on the large-scale corpus and the result of model evaluation, we show the effectiveness of our proposed method.

Key words: emotion analysis, social big data analysis, affective computing

. [J]. ZTE Communications, 2017, 15(S2): 30-37.

REN Fuji, Kazuyuki Matsumoto. Emotion Analysis on Social Big Data[J]. ZTE Communications, 2017, 15(S2): 30-37.

图/表 13

Figure 1. Making flow of hybrid emotion estimation model.

Table 1 Examples of frequently appearing tags

Favorite tag		Unfavorite tag		Personal tag
Anime	746	Insect	688	Female	1258
Game	575	Horror	220	Male	891
Music	575	Cockroach	194	Blood type A	584
Cat	425	Male	157	Blood type O	473
VOCALOID	340	Female	155	Blood type B	409
Comic	333	Cat	153	High-school student	381
Voice Actor	309	Anime	151	Student	296

Figure 2. Number of users in each cluster.

Figure 3. Results of time-series analysis of emotion when the number of clusters was 5.

Figure 4. Results of time-series analysis of emotion analysis when the number of clusters was 10.

Figure 5. Results of time-series analysis of emotion when the number of clusters was 15.

Table 2 Evaluation test set and training data

Test data (labeled)		Training data (labeled)		Pretraining data (unlabeled)
4483 tweets (Male: 15 users)	4479 tweets (Female: 15 users)	MAC: 46,979 sentences	AAC: 198,968 sentences	1,100,000 tweets

Figure 6. Experimental results of the emotion estimation model by automatically annotated corpus (male).

Figure 7. Experimental results of the emotion estimation model by automatically annotated corpus (female).

Table 3 Experimental results under Condition 2)

Label	Male			Female
Label	Precision	Recall	F1-score	Precision	Recall	F1-score
Anxiety	0.292	0.480	0.363	0.261	0.657	0.373
Love	0.220	0.791	0.344	0.251	0.775	0.379
Joy	0.931	0.905	0.918	0.940	0.950	0.945
Sorrow	0.226	0.599	0.328	0.194	0.570	0.289
Neutral	0.397	0.386	0.391	0.372	0.297	0.331
Anger	0.141	0.755	0.238	0.168	0.663	0.268
Surprise	0.251	0.110	0.153	0.314	0.116	0.169
Average	0.351	0.575	0.391	0.357	0.575	0.394

Table 4 Experimental results with Condition 3) (MSM)

Label	Male			Female
Label	Precision	Recall	F1-score	Precision	Recall	F1-score
Anxiety	0.288	0.478	0.360	0.244	0.629	0.351
Love	0.219	0.797	0.343	0.251	0.777	0.380
Joy	0.931	0.905	0.918	0.940	0.950	0.945
Sorrow	0.228	0.598	0.331	0.196	0.576	0.292
Neutral	0.395	0.386	0.390	0.369	0.280	0.318
Anger	0.141	0.752	0.238	0.164	0.620	0.260
Surprise	0.249	0.103	0.146	0.247	0.120	0.161
Average	0.350	0.574	0.389	0.344	0.564	0.387

Table 5 Experimental results with Condition 4) (MMM)

Label	Male			Female
Label	Precision	Recall	F1-score	Precision	Recall	F1-score
Anxiety	0.271	0.442	0.336	0.224	0.554	0.319
Love	0.213	0.769	0.333	0.247	0.774	0.375
Joy	0.898	0.968	0.932	0.918	0.958	0.938
Sorrow	0.262	0.584	0.362	0.226	0.692	0.341
Neutral	0.446	0.296	0.356	0.431	0.321	0.368
Anger	0.134	0.706	0.225	0.217	0.337	0.264
Surprise	0.178	0.251	0.208	0.158	0.225	0.186
Average	0.343	0.574	0.393	0.346	0.552	0.399

Table 6 Accuracy using AAC model for each gender and CR unk, CR emoji, CR face, and CR em

	Male				Female
Accuracy	0.573 (43/75)				0.688 (353/513)
	$C R unk$	$C R emoji$	$C R face$	$C R em$	$C R unk$	$C R emoji$	$C R face$	$C R em$
Failure	0.616	0.038	0.001	0.005	0.676	0.065	0.005	0.007
Success	0.653	0.055	0.002	0.006	0.675	0.059	0.227	0.009

Table 6 Accuracy using AAC model for each gender and CR unk, CR emoji, CR face, and CR em

	Male				Female
Accuracy	0.573 (43/75)				0.688 (353/513)
	$C R unk$	$C R emoji$	$C R face$	$C R em$	$C R unk$	$C R emoji$	$C R face$	$C R em$
Failure	0.616	0.038	0.001	0.005	0.676	0.065	0.005	0.007
Success	0.653	0.055	0.002	0.006	0.675	0.059	0.227	0.009

参考文献 24

[1]	F. Ren and Z. Huang , “Automatic facial expression learning method based on humanoid robot XIN-REN,” IEEE Transactions on Human-Machine Systems, vol. 46, no. 6, pp. 810-821, Dec. 2016. doi: 10.1109/THMS.2016.2599495.
[2]	F. Ren and X. Sun , “Current Situation and Development of Intelligence Robots,” ZTE Communications, vol. 14, no. S1, pp. 25-34, 2016. doi: 10.3969/j.issn.1673-5188.2016.S1.005.
[3]	F. Ren, Y. Wang, C. Quan , “A novel factored POMDP model for affective dialogue management,” Journal of Intelligent and Fuzzy Systems, vol.31, pp. 127-136, 2016.
[4]	C. Quan and F. Ren , “Automatic annotation of word emotion in sentences based on Ren-CECps,” in Proc.International Conference on Language Resources and Evaluation, Valletta, Malta, 2010, pp. 1146-1151.
[5]	C. Quan and F. Ren , “Recognizing sentence emotions based on polynomial kernel method using Ren-CECps,” in Proc. 5th IEEE International Conference on Natural Language Processing and Knowledge Engineering, Dalian, China, 2009. doi: 10.1109/NLPKE.2009.5313834.
[6]	C. Quan and F. Ren , “Recognition of word emotion state in sentences,” IEEJ Transactions on Electrical and Electronic Engineering, vol. 6, no. S1, pp. S34-S51, 2011. doi: 10.1002/tee.20618.
[7]	C. Quan, F. Ren , “Sentence emotion analysis and recognition based on emotion words using Ren-CECps,” International Journal of Advanced Intelligence, vol. 2, no. 1, pp. 105-117, 2010.
[8]	L. Wang, C. Quan, Y. Bao, F. Ren , “Construction of a Chinese emotion lexicon from Ren-CECps,” in Lecture Notes in Computer Science, vol. 8589. Berlin/Heidelberg, Germany: Springer, 2014.
[9]	J. Wang, L.-C. Yu, K. R. Lai, and X. Zhang , “Locally weighted linear regression for cross-lingual valence-arousal prediction of affective words,” Neurocomputing, vol. 194, pp. 271-278, Jun. 2016. doi: 10.1016/j.neucom.2016.02.057.
[10]	A. Pak and P. Paroubek , “Twitter for sentiment analysis: when language resources are not available,” in 22nd International Conference on Database and Expert Systems Applications, Toulouse, France, 2011. doi: 10.1109/DEXA.2011.86.
[11]	K. Matsumoto and F. Ren , “Estimation of word emotions based on part of speech and positional information,” Computers in Human Behavior, vol. 27, no. 5, pp. 1553-1564, 2011.
[12]	D. Bahdanau, K. Cho, Y. Bengio , “Neural machine translation by jointlylearning to align and translate,” presented at ICLR, San Diego, USA, 2015.
[13]	K. Matsumoto, F. Ren, M. Yoshida, K. Kita , “Refinement by filtering translation candidates and similarity based approach to expand emotion tagged corpus,” Communications in Computer and Information Science (CCIS), vol.631, pp. 260-280, 2017.
[14]	F. Ren and K. Matsumoto , “Semi-automatic creation of youth slang corpus and its application to affective computing,” IEEE Transactions on Affective Computing, vol. 7, no. 2, 2016. doi: 10.1109/TAFFC.2015.2457915.
[15]	K. Matsumoto, N. Yoshioka, K. Kita, F. Ren , “Judgment of depressive tendency from emotional fluctuation in weblog,” in Proc. 16th Annual KES Conference on Advances in Knowledge-Based and Intelligent Information and Engineering Systems, San Sebastian, Spain, 2012, pp. 545-554. doi: 10.3233/978-1-61499-105-2-545.
[16]	F. Ren, C. Quan, K. Matsumoto , “Enriching mental engineering,” International Journal of Innovative Computing, Information and Control, vol. 9, no. 8, 2013.
[17]	K. Matsumoto, S. Tanaka, M. Yoshida, K. Kita, F. Ren , “Ego-state Estimation from short texts based on sentence distributed representation,” International Journal of Advanced Intelligence (IJAI), vol. 9, no. 2, pp. 145-161, 2017.
[18]	IBM. (2018, Jan. 4). IBM Watson—personality insights [Online]. Available: https://www.ibm.com/watson/services/personality-insights
[19]	B. Eisner, T. Rocktäschel, I. Augenstein, M. Bošnjak, S. Riedel , “emoji2vec: learning emoji representations from their description,” in Proc. 4th International Workshop on Natural Language Processing for Social Media at EMNLP, Austin, USA, 2016.
[20]	Twifile. (2018, Jan. 4). Twifile [Online]. Available: http://twpf.jp
[21]	K. Matsumoto, M. Yoshida, K. Kita, Y. Wu, I. Fumihiro , “Effect of users attribute on emotion estimation from twitter, ” in Proc. 2nd IEEE International Conference on Computer and Communications, Chengdu, China, 2016, vol. 3, pp. 1186-1190.
[22]	P. Bojanowski, E. Grave, A. Joulin, T. Mikolov , “Enriching word vectors with subword information, ” arXiv preprint arXiv: 1607.04606, 2016.
[23]	A. Joulin, E. Grave, P. Bojanowski, T. Mikolov , “Bag of tricks for efficient text classification, ” arXiv preprint arXiv:1607.01759, 2016.
[24]	A. Joulin, E. Grave, P. Bojanowski , et al., “FastText.zip: compressing text classification models, ” arXiv preprint arXiv:1612.03651, 2016.