Elamur Research⌕
☰
home / papers / 2017 / Developing a concept-level knowledge base for sentiment analysis in Singlish

Developing a concept-level knowledge base for sentiment analysis in Singlish

Paper · 2017 · Rajiv Bajpai; Soujanya Poria; Danyun Ho; and Erik Cambria

Summary

The paper addresses the lack of sentiment-analysis resources for Singlish, the colloquial English of Singapore. Singlish borrows heavily from Hokkien, Cantonese, Mandarin, Malay and Tamil, so standard English tools handle it poorly. The authors build a Singlish sentiment lexicon in the SenticNet style. Each multiword concept (for example hum_sup, act_blur or shiok) gets emotion labels and a polarity value.

The concepts are not hand-labelled from WordNet or DBPedia. Instead, Singlish concepts are linked to emotion nodes as a graph, called Singlish AffectNet. This graph is blended with ConceptNet and English AffectNet, and the merged matrix is compressed with truncated SVD into a Singlish AffectiveSpace. The resulting knowledge exists in three forms: a semantic network, a matrix and a vector space.

Concepts are then placed on the Hourglass of Emotions. One ELM regressor per affective dimension predicts the emotional intensity, and that value is mapped to discrete sentic levels. The paper mainly describes how the resource is built rather than reporting a benchmark score. It matters because it is an early NLP resource for an under-served contact variety of English.

Key points

  • An earlier Singlish sentiment study used 425 political posts and reached only 35.5% accuracy, with 21% precision on positive posts and 94.4% on negative ones.
  • The knowledge comes from three sources: ConceptNet, AffectNet (ConceptNet blended with WNA emotion lists from SemEval 2007 task 14) and a new Singlish AffectNet.
  • In Singlish AffectNet, each Singlish concept points to one of the six basic emotions through a "HasProperty" edge.
  • After the matrices are blended, TSVD keeps 100 components. In the resulting space, the leading "eigenmood" axis roughly tracks positive versus negative valence.
  • Each Hourglass dimension (Pleasantness, Attention, Sensitivity, Aptitude) has its own ELM regressor. Each dimension has six sentic levels, giving 24 basic emotions, plus a neutral band around G(0).

Interesting ideas

  • I1. A dialect lexicon can be grafted onto an existing English knowledge graph by adding typed edges from the new words to shared nodes, so the new words inherit structure from the English graph.
  • I2. In "cumulative analogy," low-rank SVD of a concept×feature matrix suggests missing facts: a concept that shares most features with its neighbours probably also has their remaining feature (special_occasion inherits "Causes shiok" from birthday and hari_raya).
  • I3. Blending puts several matrices that share rows or columns into one matrix before decomposition, so the overlap between sources creates connections that none of them holds alone.
  • I4. Because the target categories (sentic levels) come from a continuous model, the authors treat prediction as regression and only then cut the result into bins, with a neutral band around zero.
  • I5. Affect is split into independent dimensions, each with its own predictor, instead of using a single sentiment score.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (world): Singlish or other dialect words could be attached to existing WordNet/Wikidata nodes through typed relations, so the ontology covers them without a separate resource.
  • I2 (world): this could propose candidate knowledge-graph edges, which are kept only if they pass an explicit check, while each suggestion records which neighbours justified it.
  • I3 (world): verb-class, FrameNet and WordNet feature tables could be merged this way to discover cross-resource links, with the merged links stored back as readable edges.
  • I4 (learn): graded labels such as intensity, certainty or politeness level could be trained as regression with explicit thresholds, which are easier to inspect than an opaque multi-class decision.
  • I5 (frames): irony and speech-act detection could compare per-dimension affect of the literal content with the context, giving an explainable signal when the two disagree.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Nanyang Technological UniversitySingapore · Singapore73 works · 24,390 citations

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 9,085 worksArtificial Intelligence · 8,888 worksArts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 worksPsychology · 798 worksExperimental and Cognitive Psychology · 463 works

🧭 Topics 2Research subjects and languages the work is about.Sentiment & emotion · 609 worksKnowledge graphs & ontologies · 103 works

Cite

Cite as Rajiv Bajpai; Soujanya Poria; Danyun Ho; and Erik Cambria. 2017. Developing a concept-level knowledge base for sentiment analysis in Singlish. arXiv:1707.04408.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.1707.04408 · arXiv 1707.04408

@misc{ho2017developing,
  title = {{Developing a concept-level knowledge base for sentiment analysis in Singlish}},
  author = {Rajiv Bajpai; Soujanya Poria; Danyun Ho; and Erik Cambria},
  year = {2017},
  doi = {10.48550/arxiv.1707.04408},
  eprint = {1707.04408},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.1707.04408}
}

References

15 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. 1. Brown, A.: Singapore English in a nutshell: An alphabetical description of its features. Federal Publications (1999) 2. Cambria, E.: Affective computing and sentiment analysis. IEEE Intelligent Systems 31(2), 102–107 (2016) 3. Cambria, E., Fu, J., Bisio, F., Poria, S.: Affectivespace 2: Enabling affective intuition for concept-level sentiment analysis. In: AAAI. pp. 508–514 (2015) 4. Cambria, E., Hussain, A.: Sentic Computing: A Common-Sense-Based Framework for Concept-Level Sentiment Analysis. find
  2. Springer, Cham, Switzerland (2015) 5. Cambria, E., Hussain, A., Durrani, T., Havasi, C., Eckl, C., Munro, J.: Sentic computing for patient centered applications. In: ICSP). pp. 1279–1282. find
  3. IEEE (2010) 6. Cambria, E., Hussain, A., Havasi, C., Eckl, C.: Common sense computing: From the society of mind to digital intuition and beyond. In: Biometric ID Management and Multimodal Communication, Lecture Notes in Computer Science, vol. 5707, pp. 252–259. find
  4. Springer, Berlin Heidelberg (2009) 7. Cambria, E., Hussain, A., Havasi, C., Eckl, C.: Sentic computing: Exploitation of common sense for the development of emotion-sensitive systems. Lecture Notes in Computer Science, vol. 5967, pp. 148–156. find
  5. Springer, Berlin Heidelberg (2010) 8. Cambria, E., Olsher, D., Kwok, K.: Sentic activation: A two-level affective common sense reasoning framework. In: AAAI. pp. 186–192. find
  6. Toronto (2012) 9. Cambria, E., Poria, S., Bajpai, R., Schuller, B.W.: Senticnet 4: A semantic resource for sentiment analysis based on conceptual primitives. In: COLING. pp. 2666–2677 (2016) 10. Eckart, C., Young, G.: The approximation of one matrix by another of lower rank. Psychometrika 1(3), 211–218 (1936) 11. Fauconnier, G., Turner, M.: The way we think: Conceptual blending and the mind’s hidden complexities. find
  7. Basic Books (2008) 12. Gupta, A.F.: The step-tongue: Children’s English in Singapore, vol. 101. Multilingual matters (1994) 13. Harada, S.: The roles of singapore standard english and singlish. Information Research 40, 70–82 (2009) 14. Huang, G.B., Cambria, E., Toh, K.A., Widrow, B., Xu, Z.: New trends of learning in computational intelligence. IEEE Computational Intelligence Magazine 10(2), 16–17 (2015) 15. Lanczos, C.: An iteration method for the solution of the eigenvalue problem of linear differential and integral operators. United States Governm. find
  8. Press Office Los Angeles, CA (1950) 16. Lee, H., Grosse, R., Ranganath, R., Ng, A.Y.: Unsupervised learning of hierarchical representations with convolutional deep belief networks. Communications of the ACM 54(10), 95–103 (2011) 17. Low, E.L., Brown, A.: English in Singapore: an introduction. find
  9. McGraw-Hill (2005) 18. Osgood, C.E., May, W.H., Miron, M.S.: Cross-cultural universals of affective meaning. University of Illinois Press (1975) Singlish Sentiment Lexicon 19. Pang, B., Lee, L., et al.: Opinion mining and sentiment analysis. Foundations and Trends® in Information Retrieval 2(1–2), 1–135 (2008) 20. Phua, Y.L.: Social media sentiment analysis and topic detection for singapore english. Tech. rep., DTIC Document (2013) 21. Platt, J.T., Weber, H.: English in Singapore and Malaysia: Status, features, functions. find
  10. Oxford University Press (1980) 22. Plutchik, R.: The nature of emotions. American Scientist 89(4), 344–350 (2001) 23. Poria, S., Cambria, E., Gelbukh, A.F.: Deep convolutional neural network textual features and multiple kernel learning for utterance-level multimodal sentiment analysis. (2015) 24. Poria, S., Cambria, E., Hazarika, D., Mazumder, N., Zadeh, A., Morency, L.P.: Context-dependent sentiment analysis in user-generated videos. In: Association for Computational Linguistics (2017) 25. Poria, S., Chaturvedi, I., Cambria, E., Hussain, A.: Convolutional mkl based multimodal emotion recognition and sentiment analysis. In: Data Mining (ICDM), 2016 IEEE 16th International Conference on. pp. 439–448.
  11. IEEE (2016) 26. Poria, S., Gelbukh, A., Cambria, E., Das, D., Bandyopadhyay, S.: Enriching Sentic-Net polarity scores through semi-supervised fuzzy clustering. In: IEEE ICDM. pp. 709–716. find
  12. Brussels (2012) 27. Poria, S., Gelbukh, A., Cambria, E., Hussain, A., Huang, G.B.: EmoSenticSpace: A novel framework for affective common-sense reasoning. Knowledge-Based Systems 69, 108–123 (2014) 28. Poria, S., Gelbukh, A., Cambria, E., Yang, P., Hussain, A., Durrani, T.: Merging senticnet and wordnet-affect emotion lists for sentiment analysis. In: ICSP. pp. 1251–1255. find
  13. IEEE (2012) 29. Sato, Y., Kim, C.: Radical pro drop and the role of syntactic agreement in colloquial singapore english. Lingua 122(8), 858–873 (2012) 30. Strapparava, C., Valitutti, A., et al.: Wordnet affect: an affective extension of wordnet. In: LREC. vol. 4, pp. 1083–1086 (2004) 31. Tay, M.W.: The uses, users and features of english in singapore. New Englishes pp. 51–70 (1982) 32. Tong, E., Zadeh, A., Morency, L.P.: Combating human trafficking with deep multimodal models. In: Association for Computational Linguistics (2017) 33. Warschauer, M.: The internet and linguistic pluralism. Silicon literacies: Communication, innovation and education in the electronic age pp. 62–74 (2002) 34. Zadeh, A.: Micro-opinion sentiment intensity analysis and summarization in online videos. In: Proceedings of the 2015 ACM on International Conference on Multimodal Interaction. pp. 587–591.
  14. ACM (2015) 35. Zadeh, A., Baltrušaitis, T., Morency, L.P.: Convolutional experts constrained local model for facial landmark detection. In: Computer Vision and Pattern Recognition Workshop (CVPRW). find
  15. IEEE (2017) 36. Zadeh, A., Chen, M., Poria, S., Cambria, E., Morency, L.P.: Tensor fusion network for multimodal sentiment analysis. In: Empirical Methods in NLP (2017) 37. Zadeh, A., Zellers, R., Pincus, E., Morency, L.P.: Mosi: multimodal corpus of sentiment intensity and subjectivity analysis in online opinion videos. arXiv preprint arXiv:1606.06259 (2016) 38. Zadeh, A., Zellers, R., Pincus, E., Morency, L.P.: Multimodal sentiment intensity analysis in videos: Facial gestures and verbal messages. IEEE Intelligent Systems 31(6), 82–88 (2016) ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/bajpai-2017-developing-concept-level-knowledge-base/