--- title: "LSIS at SemEval-2017 Task 4 — paper by Htait et al. (2017)" description: "The paper describes the LSIS entry to SemEval-2017 Task 4, subtask A, which asks systems to label… 2 references, cited by 2 papers here." authors: ["Amal Htait", "Sébastien Fournier", "Patrice Bellot"] year: 2017 venue: "Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017)" source: "https://aclanthology.org/S17-2120/" doi: "10.18653/v1/s17-2120" 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/htait-2017-lsis-semeval-2017-task-4/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/htait-2017-lsis-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/) / LSIS at SemEval-2017 Task 4: Using Adapted Sentiment Similarity Seed Words For English and Arabic Tweet Polarity Classification # LSIS at SemEval-2017 Task 4: Using Adapted Sentiment Similarity Seed Words For English and Arabic Tweet Polarity Classification **Authors:** [Amal Htait](https://md.research.elamur.ai/free/en/authors/amal-htait/), [Sébastien Fournier](https://md.research.elamur.ai/free/en/authors/sebastien-fournier/) and [Patrice Bellot](https://md.research.elamur.ai/free/en/authors/patrice-bellot/) · **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/S17-2120/) · [PDF](https://aclanthology.org/S17-2120.pdf) · [DOI 10.18653/v1/s17-2120](https://doi.org/10.18653/v1/s17-2120) · [OpenAlex](https://openalex.org/W2751434455) **Citations:** 17 citations (OpenAlex, 2026-10-14) · [2](https://md.research.elamur.ai/free/en/papers/htait-2017-lsis-semeval-2017-task-4/cited-by/) in our library, of the 149,439 papers whose bibliographies we read ## Summary The paper describes the LSIS entry to SemEval-2017 Task 4, subtask A, which asks systems to label English and Arabic tweets as positive, negative or neutral. The system is unsupervised and lexicon-based. It extends the Turney and Littman idea of scoring a word by how much closer it is to positive seed words than to negative ones. The authors argue that the classic seeds, such as "superior", rarely appear on Twitter. They therefore use seed lists built from tweets: frequent words from labelled tweet corpora, with neutral items removed. Closeness is measured as cosine similarity between word2vec vectors. The English vectors come from a public Twitter model, and the authors trained an Arabic Twitter model themselves. Word scores are summed over a tweet, and fixed thresholds decide its label. On SemEval 2013 and 2014 data, the adapted seeds roughly double macro-averaged F1 compared with Turney's seeds. This shows that matching seed vocabulary to the target domain strongly affects similarity-based sentiment lexicons. ## Key points - **English seeds:** reused from the authors' earlier work. They are the most frequent words in positive and negative Sentiment140 tweets, filtered by hand to drop neutral words. In Sentiment140, "Superior" occurs 42 times, while "Nice" occurs 23,563 times. - **Arabic seeds:** taken from the most common words in two labelled corpora, the Arabic Sentiment Tweets Dataset and the Twitter data-set for Arabic Sentiment Analysis. They were filtered against Mohammad et al.'s (2016) list of 240 Modern Standard Arabic polarity words, giving 20 positive and 20 negative seeds that are meant to work across dialects. - **Embeddings and tools:** - English uses the Godin et al. Twitter word2vec model, trained on 400M tweets with 400 dimensions. - Arabic uses a new word2vec model trained on 42M tweets from archived Twitter streams. - Similarities are computed with gensim. - **Pipeline:** - Preprocessing strips URLs, usernames, stop words and numbers out of every tweet but retains five frequent emoticons. - Elongated words are normalised (e.g. "coooool" becomes "cool"). - Each word gets a score from its similarity to the positive versus negative seeds, and the word scores are summed. - A tweet is positive if the sum is above 1, negative if it is below −2, and neutral otherwise. These thresholds were tuned on SemEval 2013/2014. - **Results, Turney seeds → adapted seeds:** - 2013: AvgF1 0.262→0.564, AvgR 0.381→0.571, Acc 0.480→0.508. - 2014: AvgF1 0.303→0.589, AvgR 0.383→0.553, Acc 0.511→0.552. ## Interesting ideas - Seed words should be chosen by how often they actually occur in the target genre, since rare seeds give embedding similarities little to anchor on; this applies to any seed-driven lexicon induction. - A cheap recipe for building seeds is to take the most frequent words per label in an annotated corpus and then remove neutral ones with an existing small lexicon instead of by hand. - The paper compares classic and adapted seeds while keeping the rest of the pipeline identical, which makes a clean control that isolates the effect of a single resource. - The decision thresholds are asymmetric (above 1 for positive, below −2 for negative) and were tuned on earlier shared-task data, which accounts for a skew in scores toward the positive side. - Collapsing elongated spellings to their base form before lookup lets informal text reuse a standard vocabulary. ## For Mova How these ideas could serve Mova, our small open language model — its files, stages and tests: - **I1** (world): When compiling evaluative values or attitude lexicons for the knowledge layer, pick anchor words that are frequent in the genres Mova reads (tales, school texts) rather than generic dictionary terms. - **I2** (learn): The same frequency-plus-filter step could bootstrap candidate lists (e.g., polarity of verbs or adjectives) from silver-labelled data, with an existing resource acting as the gatekeeper. - **I3** (learn): Evaluate every swapped-in lexicon or rule set against the old one on a frozen pipeline, so that gains can be attributed explainably to that component. - **I4** (frames): When polarity or attitude scores feed speech-act or irony judgements, calibrate separate cut-offs per class on held-out data instead of assuming that zero is neutral. - **I5** (tokens): Add an explainable elongation-normalisation step in the tokenizer, keeping the original form as a feature, so that informal input still reaches correct lemmas. *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 - [Centre National de la Recherche Scientifique](https://md.research.elamur.ai/free/en/labs/centre-national-de-la-recherche-scientifique/) — Research institute, Paris, France · 169 works · 62,488 citations - [Aix-Marseille Université](https://md.research.elamur.ai/free/en/universities/aix-marseille-universite/) — University, Marseille, France · 25 works · 39,241 citations - [OpenEdition Center](https://md.research.elamur.ai/free/en/labs/openedition-center/) — Research institute, Marseille, France · 2 works · 42 citations Also on the byline: [Arts et Métiers](https://ror.org/018pp1107). 📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works 🧭 Topics (3): [English](https://md.research.elamur.ai/free/en/topics/english/) · 2,379 works · [😊 Sentiment & emotion](https://md.research.elamur.ai/free/en/topics/sentiment-emotion/) · 1,035 works · [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 3,066 works ## Cite Cite as: Amal Htait, Sébastien Fournier and Patrice Bellot. 2017. LSIS at SemEval-2017 Task 4: Using Adapted Sentiment Similarity Seed Words For English and Arabic Tweet Polarity Classification. 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{htait2017lsis, title = {{LSIS at SemEval-2017 Task 4: Using Adapted Sentiment Similarity Seed Words For English and Arabic Tweet Polarity Classification}}, author = {Amal Htait, Sébastien Fournier and Patrice Bellot}, booktitle = {Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017)}, year = {2017}, doi = {10.18653/v1/s17-2120}, url = {https://doi.org/10.18653/v1/s17-2120} } ``` - [← previous Unsupervised AMR-Dependency Parse Alignment](https://md.research.elamur.ai/free/en/papers/chen-2017-unsupervised-amr-dependency-parse-alignment/) - [next → RIGOTRIO at SemEval-2017 Task 9: Combining Machine Learning and Grammar Engineering for AMR Parsing and Generation](https://md.research.elamur.ai/free/en/papers/gruzitis-2017-rigotrio-semeval-2017-task-9/) - [↑ back to papers of 2017](https://md.research.elamur.ai/free/en/papers/2017/page/5/) ## References 2 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Erik Cambria, Robert Speer, Catherine Havasi, and Amir Hussain. 2010. Senticnet: A publicly available semantic resource for opinion mining. In AAAI fall symposium: commonsense knowledge. volume 10. Acc 0.658 0.521 Acc 0.581 0.445 [find](https://scholar.google.com/scholar?q=Erik+Cambria%2C+Robert+Speer%2C+Catherine+Havasi%2C+and+Amir+Hussain%2E+2010%2E+Senticnet%3A+A+publicly+available+semantic+resource+for+opinion+mining%2E+In+AAAI+fall+symposium%3A+commonsense+knowledge%2E+volume+10%2E+Ac) 2. Samhaa R El-Beltagy and Ahmed Ali. 2013. Open issues in the sentiment analysis of arabic social media: A case study. In Innovations in information technology (iit), 2013 9th international conference on. IEEE, pages 215–220. *Table 5: The results at semEval2017 Task 4 subtask A - for English and Arabic Languages.* Ramy Eskander and Owen Rambow. 2015. [Slsa: A sentiment lexicon for standard arabic](https://md.research.elamur.ai/free/en/papers/eskander-2015-slsa-sentiment-lexicon-standard-arabic/). In EMNLP. pages 2545–2550. *Updated 2 November 2026*