---
title: "INGEOTEC at SemEval 2017 Task 4: A B4MSA Ensemble based"
description: "The paper describes the INGEOTEC entry for SemEval-2017 Task 4, Subtask A. The task is to label… 12 references, cited by 2 papers here."
authors: ["Sabino Miranda-Jiménez", "Mario Graff", "Eric Sadit Tellez", "Daniela Moctezuma"]
year: 2017
venue: "Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017)"
source: "https://aclanthology.org/S17-2130/"
doi: "10.18653/v1/s17-2130"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/jimenez-2017-ingeotec-semeval-2017-task-4/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/jimenez-2017-ingeotec-semeval-2017-task-4/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2017](https://md.research.elamur.ai/free/en/papers/2017/page/5/) / INGEOTEC at SemEval 2017 Task 4: A B4MSA Ensemble based on Genetic Programming for Twitter Sentiment Analysis

# INGEOTEC at SemEval 2017 Task 4: A B4MSA Ensemble based on Genetic Programming for Twitter Sentiment Analysis

**Authors:** [Sabino Miranda-Jiménez](https://md.research.elamur.ai/free/en/authors/sabino-miranda-jimenez/), Mario Graff, Eric Sadit Tellez and Daniela Moctezuma · **Year:** 2017 · **Venue:** [Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017)](https://md.research.elamur.ai/free/en/venues/semeval/) ([SemEval 2017](https://md.research.elamur.ai/free/en/venues/semeval/2017/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/S17-2130/) · [PDF](https://aclanthology.org/S17-2130.pdf) · [DOI 10.18653/v1/s17-2130](https://doi.org/10.18653/v1/s17-2130) · [OpenAlex](https://openalex.org/W2750618026)

**Citations:** 16 citations (OpenAlex, 2026-10-14) · 2 in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper describes the INGEOTEC entry for SemEval-2017 Task 4, Subtask A. The task is to label tweets as positive, negative or neutral, in English and Arabic. The system has two layers. The first layer is a set of B4MSA classifiers, which are SVMs over TF-IDF features. Each one's text-normalisation steps and tokenizers (word n-grams, character q-grams, skip-grams) were chosen automatically by hyper-heuristic search over a huge configuration space. Some of these classifiers learn from human-labelled data: SemEval 2013–2016 plus the HL dataset of Mozetič et al. Others learn from a large distant-supervision corpus, labelled with an affective lexicon built from SentiSense, WordNet and Bing Liu's list. In the second layer, EvoDAG, a genetic programming system, evolves a nonlinear function that combines all the decision values. On the official test data the system reached macro-recall 0.649 for English (sixth place) and 0.477 for Arabic (fourth place). The work shows that weakly labelled data, once deduplicated and partitioned, measurably helps a language-light classifier.

## Key points
- **Distant-supervision data.** About 220M raw English tweets were reduced to roughly 11M labelled examples. More than 130K Arabic tweets were reduced to about 16K.
  - Tweets containing both positive and negative lexicon words were discarded.
  - The Arabic lexicon was a machine translation of the English one, produced with the Bing translator.
- **Partitioning and input selection.** The DS corpus was split into parts of 30K tweets, each with its own binary B4MSA classifier.
  - For each tweet, the top k classifiers are chosen dynamically by a vocabulary-affinity score.
  - The English model received 50 EvoDAG inputs: 6 from two HL classifiers and 44 from DS classifiers.
- **Value of the DS data.** Training-stage macro-recall in English was 0.680 with the DS data and 0.652 without it (configuration 2-HL, train/4), a gap of nearly 3%. Using 3.5M DS tweets instead of 11M gave almost the same score (0.679).
- **Arabic model.** It had only 5 EvoDAG inputs: one HL classifier with three outputs and two DS classifiers of about 8K tweets each. Its macro-recall dropped from 0.642 in training to 0.477 on the gold standard.
- **Cost of the configuration search.** Exhaustively evaluating every B4MSA configuration was estimated at about 15 years, at roughly 10 minutes per configuration. Hyper-heuristics found a competitive setup within a few hours.

## Interesting ideas
- Deduplication by coarsening: apply aggressive normalisations (case, URLs, users, numbers, punctuation) and drop any tweet whose normalised form has already appeared, which shrinks noisy corpora while keeping vocabulary variety.
- Per-input choice of sub-models by vocabulary affinity: each classifier scores how much of a given text's vocabulary it has seen, and only the most familiar ones are consulted.
- Treating preprocessing and tokenization choices as a combinatorial search scored by held-out performance, instead of fixing them by hand.
- A stacking layer that sees only raw decision values, with no labels saying which model or class each value came from, so it learns the agreement structure blindly.
- Seeding a low-resource language by machine-translating a lexicon and reusing the same labelling rules. The Arabic score drop also warns how far such transfer can fall short.

## For Mova

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

- **I1** (learn): Silver-data pools could be normalised the same way before training, so that templated near-copies do not inflate agreement counts or evaluation sets.
- **I2** (learn): Parsers or repair-rule sets trained on different domains could be routed by lexical coverage of the sentence, and the coverage score could be logged as an explainable reason for each choice.
- **I3** (tokens): Tokenizer options such as URL/user/number grouping and handling of emoticons or diacritics could be tuned by search against tagging and LAS scores on a dev treebank.
- **I4** (learn): When merging witness parsers, a combiner that sees only anonymised arc scores would show whether agreement alone, without trust in any named parser, predicts correctness.
- **I5** (world): Translated lexical resources could be used to bootstrap verb-class or value lexicons, but only with a separate gold check, since training-stage scores can hide large losses.

*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

Also on the byline: [Centro de Investigación en Ciencias de Información Geoespacial](https://ror.org/017rjh121) · [Centro de Investigación e Innovación en Tecnologías de la Información y Comunicación](https://ror.org/04rb41y32).

📚 Shelves (2): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works

🧭 Topics (2): [😊 Sentiment & emotion](https://md.research.elamur.ai/free/en/topics/sentiment-emotion/) · 972 works · [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works

## Cite

Cite as: Sabino Miranda-Jiménez, Mario Graff, Eric Sadit Tellez and Daniela Moctezuma. 2017. INGEOTEC at SemEval 2017 Task 4: A B4MSA Ensemble based on Genetic Programming for Twitter Sentiment Analysis. Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017).

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{mirandajimenez2017ingeotec,
  title = {{INGEOTEC at SemEval 2017 Task 4: A B4MSA Ensemble based on Genetic Programming for Twitter Sentiment Analysis}},
  author = {Sabino Miranda-Jiménez, Mario Graff, Eric Sadit Tellez and Daniela Moctezuma},
  booktitle = {Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017)},
  year = {2017},
  doi = {10.18653/v1/s17-2130},
  url = {https://doi.org/10.18653/v1/s17-2130}
}
```

- [← previous Alto: Rapid Prototyping for Parsing and Translation](https://md.research.elamur.ai/free/en/papers/gontrum-2017-alto-rapid-prototyping-parsing-translation/)
- [next → A Layered Language Model based Hybrid Approach to Automatic Full Diacritization of Arabic](https://md.research.elamur.ai/free/en/papers/badrashiny-2017-layered-language-model-based-hybrid/)
- [↑ back to papers of 2017](https://md.research.elamur.ai/free/en/papers/2017/page/5/)

## References

12 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Steven Bird, Ewan Klein, and Edward Loper. 2009. Natural Language Processing with Python. O’Reilly Media. [find](https://scholar.google.com/scholar?q=Steven+Bird%2C+Ewan+Klein%2C+and+Edward+Loper%2E+2009%2E+Natural+Language+Processing+with+Python%2E+O%E2%80%99Reilly+Media%2E)
2. Igor Mozetič, Miha Grčar, and Jasmina Smailović. 2016. Multilingual twitter sentiment classification: The role of human annotators. PloS one 11(5):e0155036. [↗](https://doi.org/10.1371/journal.pone.0155036)
3. Jorge Carrillo de Albornoz, Laura Plaza, and Pablo Gervás. 2012. Sentisense: An easily scalable concept-based affective lexicon for sentiment analysis. In Proceedings of LREC 2012. pages 3562– 3567. [find](https://scholar.google.com/scholar?q=Jorge+Carrillo+de+Albornoz%2C+Laura+Plaza%2C+and+Pablo+Gerva%CC%81s%2E+2012%2E+Sentisense%3A+An+easily+scalable+concept%2Dbased+affective+lexicon+for+sentiment+analysis%2E+In+Proceedings+of+LREC+2012%2E+pages+3562%E2%80%93+3567%2E)
4. Preslav Nakov, Alan Ritter, Sara Rosenthal, Veselin Stoyanov, and Fabrizio Sebastiani. 2016. [SemEval-2016 task 4: Sentiment analysis in Twitter](https://md.research.elamur.ai/free/en/papers/nakov-2016-semeval-2016-task-4-sentiment/). In Proceedings of the 10th International Workshop on Semantic Evaluation. Association for Computational Linguistics, San Diego, California, SemEval ’16.
5. Jan Deriu, Maurice Gonzenbach, Fatih Uzdilli, Aurelien Lucchi, Valeria De Luca, and Martin Jaggi. 2016. Swisscheese at semeval-2016 task 4: Sentiment classification using an ensemble of convolutional neural networks with distant supervision. In Proceedings of the 10th International Workshop on Semantic Evaluation (SemEval-2016). Association for Computational Linguistics, San Diego, California, pages 1124–1128. http://www.aclweb.org/anthology/S16-1173. [↗](https://aclanthology.org/S16-1173/)
6. NRC. 2017. Syrian tweets arabic sentiment analysis dataset. [http://saifmohammad.com/](http://saifmohammad.com/) WebPages/ArabicSA.html. Accessed 17-Feb-2017. [↗](http://saifmohammad.com/)
7. SemEval. 2017. Semeval-2017: Sentiment analysis task 4. [http://alt.qcri.org/](http://alt.qcri.org/) semeval2017/task4/. Accessed 17-Feb-2017. [↗](http://alt.qcri.org/)
8. Alec Go, Richa Bhayani, and Lei Huang. 2009. Twitter sentiment classification using distant supervision. CS224N Project Report, Stanford 1(12). [find](https://scholar.google.com/scholar?q=Alec+Go%2C+Richa+Bhayani%2C+and+Lei+Huang%2E+2009%2E+Twitter+sentiment+classification+using+distant+supervision%2E+CS224N+Project+Report%2C+Stanford+1%2812%29%2E)
9. Eric S. Tellez, Sabino Miranda-Jimnez, Mario Graff, Daniela Moctezuma, Oscar S. Siordia, and Elio A. Villaseor. 2017. A case study of spanish text transformations for twitter sentiment analysis. Expert Systems with Applications 81:457 – 471. [↗](https://doi.org/10.1016/j.eswa.2017.03.071)
10. M. Graff, E. S. Tellez, S. Miranda-Jiménez, and H. J. Escalante. 2016. Evodag: A semantic genetic programming python library. In 2016 IEEE International Autumn Meeting on Power, Electronics and Computing (ROPEC). pages 1–6. https://doi.org/10.1109/ROPEC.2016.7830633. [↗](https://doi.org/10.1109/ropec.2016.7830633)
11. Eric Sadit Tellez, Sabino Miranda-Jiménez, Mario Graff, Daniela Moctezuma, Ranyart Rodrigo Suárez, and Oscar Sánchez Siordia. 2016. A simple approach to multilingual polarity classification in twitter. CoRR abs/1612.05270. http://arxiv.org/abs/1612.05270. [↗](https://arxiv.org/abs/1612.05270)
12. Mario Graff, Eric S. Tellez, Hugo Jair Escalante, and Sabino Miranda-Jiménez. 2017. Semantic Genetic Programming for Sentiment Analysis. In Oliver Schtze, Leonardo Trujillo, Pierrick Legrand, and Yazmin Maldonado, editors, NEO 2015, Springer [↗](https://doi.org/10.1007/978-3-319-44003-3_2)

*Updated 1 November 2026*