Pith. sign in

REVIEW 3 major objections 6 minor 48 references

Uchaguzi-2022: A Dataset of Citizen Reports on the 2022 Kenyan Election

T0 review · 3 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper presents Uchaguzi-2022, a dataset of 14,169 categorized and geotagged citizen reports from the 2022 Kenyan General Election, and shows that few-shot large language models can categorize reports as well as volunteer annotators…

desk verdict A solid citizen-report corpus for African election monitoring, with benchmark numbers that need a distributional caveat. read the letter →

arxiv 2412.13098 v1 pith:DJKABYYV submitted 2024-12-17 cs.CL cs.SI

Roberto Mondini , Neema Kotonya , Robert L. Logan IV , Elizabeth M Olson , Angela Oduor Lungati , Daniel Duke Odongo , Tim Ombasa , Hemank Lamba
show 3 more authors
Aoife Cahill Joel R. Tetreault Alejandro Jaimes
This is my paper · ORCID
classification cs.CLcs.SI
keywords Uchaguzi-2022citizenreportselectionmonitoringKenya2022generaltextcategorizationgeotaggingfew-shotin-contextlearningAIforSocialGood
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper introduces Uchaguzi-2022, a dataset of 14,169 citizen reports submitted during Kenya's August 2022 general election, each labeled by volunteers with a topic, optional topic-specific tags, and geographic coordinates. It claims this is the first dataset of citizen-contributed election issues on the African continent that combines categorization with geotagging. Using this dataset, the paper asks whether language models can do the labeling work at scale that volunteers currently do by hand. The answer it defends is yes in large part: a few-shot prompted 70-billion-parameter language model reaches a micro-averaged F1 of 54.6 on topic classification, slightly above the 50.4 agreement between volunteer annotations and a single expert labeler, and few-shot models also beat a fine-tuned encoder on rare topic classes. The same model family outperforms a Wikipedia-trained named-entity recognizer at extracting Kenyan locations, and geocoding resolves coordinates for about 40 percent of reports that carry no explicit location label.

What carries the argument

The load-bearing object is the dataset itself: 14,169 volunteer-annotated reports plus a 500-report subset re-annotated by a single expert and used as the test set. The mechanism that carries the benchmark comparison is the pairing of two learning setups on the same test set: fully supervised fine-tuning of XLM-RoBERTa-base, and 25-shot in-context learning with large language models. Evaluation uses micro- and macro-averaged F1 for topic and tag prediction, with Cohen's kappa quantifying volunteer-expert agreement, and for geotagging it uses location-extraction exact match and ROUGE-L followed by geocoding coverage, accuracy at 161 km, and area under the curve. The few-shot prompts include topic and tag descriptions, which is what lets a model with no fine-tuning adapt to the taxonomy.

What would settle it

Re-annotate the 500-report sample with several independent experts who are not affiliated with the original annotators; if their pairwise agreement on topics is no higher than the current volunteer-expert kappa of 0.425, the human baseline against which models are compared is too unstable to support the claim that few-shot models are competitive.

Watch

Extended reading notes

Core claim

The central claim is that crowdsourced election monitoring can be made scalable without sacrificing classification quality. Uchaguzi-2022 contains 14,169 reports across ten topics, dominated by Opinions (66.6 percent), with tags such as Biometric voter registration issues and Counting irregularities, and coordinates for 98.4 percent of reports. On the 500-report expert-labeled test set, the best few-shot model, llama-3.1-70b, achieves a micro F1 of 54.6 and macro F1 of 39.7 on topic prediction, improving on the volunteer-expert agreement baseline of 50.4 and 48.6 respectively, while the fully fine-tuned XLM-RoBERTa-base reaches 49.4 and 27.9. This makes few-shot in-context learning, not fine-tuning, the strongest configuration for these tasks, especially on rare classes. For geotagging, the same few-shot models extract location mentions far more accurately than a Wikipedia-trained NER system, and combining them with OpenStreetMap geocoding obtains correct coordinates for 68.7 percent of reports with labeled locations and 38.2 percent of reports without them.

Load-bearing premise

The benchmark comparison assumes that the 500-report expert-annotated test set is a reliable gold standard, but the expert belongs to the same organization as the volunteers and agrees with them only moderately on topics, so any noise in the expert labels would shift the model rankings.

Editorial extensions

If this is right

  • Because the best few-shot model slightly exceeds the human agreement baseline on micro F1, the paper argues it could generate silver annotations for the roughly 86,000 incoming reports that volunteers never labeled.
  • Few-shot in-context learning outperforms fine-tuning on underrepresented topics such as Irrelevant Report and Political Rallies, so rare-class performance is not a blocker for automation.
  • Both approaches fail entirely on Polling Station Administration and Staffing Issues, so fully automatic categorization still needs human review for these safety-relevant topics.
  • ICL location extraction combined with OpenStreetMap geocoding recovers coordinates for about 40 percent of reports without explicit location labels, indicating that missing locations on the platform are partly an annotation gap rather than an absence of location mentions.
  • The topic taxonomy is Kenyan-election-agnostic, so the categorization approach transfers across regions, while the tag taxonomy is Kenya-specific and would require redefinition.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An implication the paper leaves implicit is that the moderate volunteer-expert agreement of kappa 0.425 reflects genuine label subjectivity, so a practical pipeline should treat model outputs as a triage that surfaces confident and uncertain cases for volunteer review rather than as a replacement for human labeling.
  • The dataset's code-switched English/Swahili texts and long-tailed topic distribution could serve as a lightweight stress test for multilingual models in low-resource civic settings, going beyond the paper's own benchmark.
  • The paper leaves unquantified the volunteer-hours saved by automation; a natural next experiment would measure the end-to-end cost of a human-plus-model workflow against a human-only one using the same 500-report test set.
  • Because only 2.4 percent of reports carry a location text field while 98.4 percent have coordinates, the silver-geocoding result suggests that retrofitting existing archives with location names and edges could improve map-based analyses of past elections.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper introduces Uchaguzi-2022, a dataset of 14,169 citizen reports on the 2022 Kenyan general election, collected via the Ushahidi platform. Each report is annotated by volunteers with a topic, topic-specific tags, a title, and geographic coordinates; the paper analyzes the dataset's coverage, language, geography, and temporal trends. The authors then benchmark models for automated topic classification, tag prediction, location extraction, and geocoding, comparing a fine-tuned XLM-RoBERTa model with few-shot in-context learning using gpt-4o, mixtral-8x7b, and llama-3.1-70b. The main empirical claims are that few-shot models are competitive with fully fine-tuned models on categorization, that llama-3.1-70b achieves micro-F1 54.6 on topic prediction versus a volunteer-expert agreement baseline of 50.4, and that ICL approaches substantially outperform a Wikipedia-trained NER model on location extraction and geocoding. The dataset and code are released under a licensing agreement.

Significance. If the benchmark results are robust, this paper makes a meaningful contribution to low-resource NLP and AI-for-social-good. The dataset fills a clear gap: it provides citizen-contributed election-issue reports with fine-grained topic tags and geotags for an African election, whereas existing election datasets focus mainly on sentiment or are based on news articles rather than citizen reports. The authors are transparent about the moderate inter-annotator agreement (Cohen's kappa 0.425 for topics), which is a strength. The paper also ships code and a data-access mechanism, supporting reproducibility. The central empirical conclusion—that few-shot LLMs can match or exceed the volunteer-expert agreement level—is plausible but depends on the construction of the expert-labeled test set, which is the main risk to the paper's claims.

major comments (3)
  1. [Section 2.4 and Section 4.1.1, Table 5] The expert-labeled test set of 500 reports is not a random sample of Uchaguzi-2022. The sampling strategy in Section 2.4 deliberately oversamples rare topics (e.g., Irrelevant Report, Political Rallies) and undersamples the dominant Opinions topic, which is 66.6% of the dataset but only about 20% of the test set. Because micro-F1 is a prevalence-weighted metric, the reported values in Table 5 and the resulting comparison between fine-tuned and few-shot models, as well as the comparison against the human baseline, are specific to the stratified distribution and may not reflect performance on the natural data distribution. The authors should report the test set class distribution, provide reweighted or naturally-distributed results (e.g., by evaluating on a random sample or by weighting per-class F1), and discuss whether the conclusion that few-shot models are competitive with fully fine-tuned models remains valid under the natural distribution.
  2. [Section 4.2.2 and Appendix C] The silver geocoding evaluation set of 2,681 reports is constructed by using zero-shot gpt-4o prompting to filter the set of reports that 'likely include a location mention.' Since gpt-4o is also one of the models evaluated on this silver set (Table 10), the selection process is not independent of the evaluated systems and may bias the geocoding results in favor of gpt-4o and similar models. The authors should either construct the silver set using a method that is independent of the evaluated models (e.g., a rule-based heuristic or human annotation on a random subset) or provide a sensitivity analysis showing that the ranking of models does not change when a different filtering approach is used. Without this, the claim that 'ICL methods are still able to correctly geocode around 40% of reports without explicit location annotations' may not generalize to the broader population of unannotated reports.
  3. [Section 2.4 and Section 4.1.1] The human baseline in Table 5 is computed as agreement between volunteer annotations and a single expert annotator, but the expert is also employed by Ushahidi, the same organization that trains and coordinates the volunteers. The moderate Cohen's kappa of 0.425 indicates that the gold standard itself carries substantial subjectivity. The paper should explicitly frame this baseline as agreement with one expert's opinion rather than as a ground-truth benchmark, and it should provide per-class human agreement to help interpret the macro-F1 numbers. This is particularly important because the conclusion that llama-3.1-70b is 'competitive with the human baseline' is based on micro-F1 exceeding 50.4, while its macro-F1 (39.7) is well below the human baseline (48.6).
minor comments (6)
  1. [Section 4.1.1, paragraph after Table 5] The phrase 'with lower macro-averaged F1 but higher overall accuracy' is ambiguous; the table reports micro-F1 and macro-F1, not accuracy. Please clarify whether 'overall accuracy' refers to micro-F1 or to a separate accuracy metric, and if the latter, report it.
  2. [Table 5] The row 'Human annotation (baseline)' does not specify how the F1 scores are computed from volunteer and expert annotations. Please state explicitly which annotation set is treated as predictions and which as ground truth, and whether the reported values are micro/macro F1 over the same label set used for the models.
  3. [Section 3.1] The sentence 'For each task we omit labels with < 20 observations' should specify the resulting label sets, since this threshold directly affects the tag prediction tasks and the set of classes in Table 7.
  4. [Section 4.2.1, Table 8] The column '% Empty Pred.' is not defined in the text. Please state explicitly that it is the percentage of samples for which the system returned no location, and indicate whether the denominator is the 147-sample evaluation set.
  5. [Appendix C, Figure C2] The example outputs in the location extraction prompt use inconsistent JSON keys: some use 'label' and others use 'location'. Please standardize the schema to avoid confusion in reproducing the prompt.
  6. [Section 1, contribution 1] The claim that this is 'the first dataset of citizen-contributed election issues in the African continent that contains this associated metadata' would benefit from a more thorough related-work comparison, including earlier Ushahidi deployments in other African elections and any other crowdsourced election-monitoring datasets from Africa.

Circularity Check

1 steps flagged · score 3.0 of 10

Silver geocoding test set is self-selected by the evaluated model, while the main categorization benchmarks are externally scored.

  1. fitted input called prediction [Section 3.2 (Automating Geotagging, silver dataset construction), Section 4.2.2 (Geocoding, Table 10), Appendix C]
    "We evaluate geocoding on a total of 2,818 samples: the 147 samples used for the location extraction evaluation, and, additionally, a silver dataset of 2,681 reports that do not have annotated locations but have associated coordinates and likely include a location mention. ... We use zero-shot prompting with gpt-4o (Appendix C) to identify and filter out reports that do not mention locations. We choose gpt-4o as it has the highest recall for location extraction (§4.2.1)."

    The 2,681-sample 'silver' geocoding test set is defined as the subset of reports that gpt-4o's zero-shot prompt judges to contain a location mention. gpt-4o is then one of the models evaluated on Table 10 for geocoding this same set. Because the set was filtered by gpt-4o's own location-extraction ability, reports on which gpt-4o would extract no location (and thus cannot geocode) are excluded by construction. The reported Acc@161km and AUC for gpt-4o are therefore self-selected estimates, not unbiased measures on 'reports without explicit location annotations.' The same filter also conditions the comparison for the other ICL models, since all are scored only on reports that gpt-4o deemed to contain locations.

full rationale

The primary categorization benchmarks (Tables 5-7) are evaluated against expert-annotated labels that are independent of the trained models, so the central 'few-shot models are competitive with fully fine-tuned models' claim is not circular. The 500-report test set is topic-stratified, which can affect prevalence-weighted metrics such as micro-F1, but this is a distributional concern rather than a circularity. The one partially circular step is the silver geocoding evaluation: the 2,681-report set is constructed by gpt-4o filtering, and gpt-4o is then evaluated on it, inflating its geocoding numbers. This affects a secondary result (geocoding without labeled location), not the main categorization claim, so the overall circularity score is low-to-moderate. The gold geocoding evaluation on the 147 manually verified location samples remains independent.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The paper's claims rest on the volunteer annotation pipeline, the single-expert gold standard, and the gpt-4o-filtered silver geocoding set, plus hand-chosen modeling thresholds (25-shot, label cutoff 20, 500-sample test). No invented physical or theoretical entities are introduced; the dataset itself is a genuine new resource, not a postulated entity.

free parameters (4)
  • Minimum label frequency threshold for tag tasks = 20
    Tags with fewer than 20 observations are omitted from training and evaluation (Section 3.1). This hand-chosen cutoff changes the label space and therefore the macro-F1 comparability across tasks and models.
  • Number of in-context examples for few-shot learning = 25
    The few-shot setting uses 25 examples per prompt (Section 3.1); the count and the particular sampled examples are hand-chosen, and no sampling seeds are reported, so results may vary across draws.
  • Expert-annotated gold-standard test size and sampling rule = 500 (100 per topic or 10%, whichever is smaller)
    The 500-sample test set is built by a stratified rule (Sections 2.4 and 3.1). The resulting class distribution is not proportional to the corpus (Opinions is 66.6% of the corpus but a smaller share of the test set), so micro-F1 is not a direct deployment-accuracy estimate.
  • Geocoding accuracy radius = 161 km
    Acc@161km is adopted from prior geocoding evaluation (Jurgens et al., 2015), not fitted to this data; it affects the reported accuracy and AUC readings in Tables 9 and 10.
assumptions (5)
  • domain assumption Volunteer annotations in the 14,169-report corpus are accurate enough to serve as training labels.
    Moderate topic agreement between volunteers and the expert (kappa 0.425) and 98.4% coordinate coverage are treated as sufficient signal for supervised training (Section 2.4). If volunteer labels are dominated by noise, the supervised benchmark numbers lose meaning.
  • domain assumption The single expert annotator's labels constitute a valid gold standard.
    The expert is also employed by Ushahidi and no inter-expert reliability is measured (Sections 2.4 and 4.1). The paper's benchmark comparisons assume these labels are a fair reference.
  • domain assumption Reported coordinates locate the reported event, not the reporting citizen.
    Asserted in Section 8 and used implicitly in the geocoding evaluation. If coordinates instead track reporters, geocoding accuracy against annotations is mis-measured.
  • domain assumption fastText language identification is reliable enough for the language distribution analysis.
    The claim that 98.0% of reports are English (Section 2.3) comes from fasttext; code-switching is acknowledged, so language proportions are approximate.
  • ad hoc to paper gpt-4o zero-shot filtering reliably identifies reports that mention locations, and the filtered reports form a valid silver geocoding test set.
    Introduced in Section 3.2 to expand the geocoding evaluation to 2,681 samples. The filter is produced by one of the evaluated models, so absolute results on this set are partly self-selected.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Uchaguzi-2022: A Dataset of Citizen Reports on the 2022 Kenyan Election." pith.science (2026). https://pith.science/paper/DJKABYYV

@misc{pith2026241213098,
  author       = {Pith},
  title        = {Pith review of: Uchaguzi-2022: A Dataset of Citizen Reports on the 2022 Kenyan Election},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DJKABYYV}},
  note         = {Machine review of arXiv:2412.13098}
}
read the original abstract

Online reporting platforms have enabled citizens around the world to collectively share their opinions and report in real time on events impacting their local communities. Systematically organizing (e.g., categorizing by attributes) and geotagging large amounts of crowdsourced information is crucial to ensuring that accurate and meaningful insights can be drawn from this data and used by policy makers to bring about positive change. These tasks, however, typically require extensive manual annotation efforts. In this paper we present Uchaguzi-2022, a dataset of 14k categorized and geotagged citizen reports related to the 2022 Kenyan General Election containing mentions of election-related issues such as official misconduct, vote count irregularities, and acts of violence. We use this dataset to investigate whether language models can assist in scalably categorizing and geotagging reports, thus highlighting its potential application in the AI for Social Good space.

Figures

Figures reproduced from arXiv: 2412.13098 by the authors.

Figure 1
Figure 1. Annotated report example. Each input mes￾sage (top) is categorized and geolocated (bottom). their communities. Furthermore, the systematic organization of large amounts of crowdsourced information enables reporting agencies to draw meaningful insights from this data and share these with the relevant policy makers to enact positive changes (Shayo, 2021). Unfortunately, categoriz￾ing and geotagging large amounts of da… view at source ↗
Figure 2
Figure 2. Distributions of title and message lengths. Field % Non-empty title 97.9 topic 100.0 tags 85.1 location 2.4 coordinates 98.4 [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Reports per capita. Scale is per 100K citi￾zens. Kenya (Muaka, 2011), we study the language dis￾tribution in the reports with fasttext (Joulin et al., 2017). According to this model, 98.0% of reports are in English, 1.3% are in Swahili, and the remain￾ing 0.7% are other languages. Among the reports detected as English, we observe the presence of English-Swahili code-switching and show exam￾ples in [PITH_FULL_IMAGE:… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Report counts over time (left) and words characterizing different phases of the election (right). (this event also corresponds to the spike in reports in late July). In contrast, the characteristic elec￾tion day tokens appear to relate to preliminary re￾sults and turno…
Figure 6
Figure 6. Figure 6: Location extraction. We retrieve all location mentions (green) in the title or text of a report. the volunteer-labeled data into train (80%) and val￾idation (10%) sets, and use the expert-labeled data as our test set (which was sub-sampled from the remaining 10%). In t…
Figure 7
Figure 7. Figure 7: Examples of inaccurate topic prediction. across both metrics [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: Examples of inaccurate tag prediction. Mi￾nor edits to the original texts for readability. models achieve the highest performance in both exact match and ROUGE-L, with llama-3.1-70b outperforming the other models. We observe a large gap in performance between the mBERT…
Figure 9
Figure 9. Figure 9: Example of failed geocoding. coordinates are mostly inaccurate. The results for samples without labeled location ( [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 29 canonical work pages

  1. [1]

    Per Sekse Aarvik. 2015. Uchaguzi: An analysis of the crowdsourced election monitoring in kenya 2013. Master's thesis, The University of Bergen

    work page 2015

  2. [2]

    Hadi Abdine, Yanzhu Guo, Virgile Rennard, and Michalis Vazirgiannis. 2022. https://aclanthology.org/2022.politicalnlp-1.9 Political communities on T witter: Case study of the 2022 F rench presidential election . In Proceedings of the LREC 2022 workshop on Natural Language Processing for Political Sciences, pages 62--71, Marseille, France. European Languag...

    work page 2022

  3. [3]

    Anton Abilov, Yiqing Hua, Hana Matatov, Ofra Amir, and Mor Naaman. 2021. https://ojs.aaai.org/index.php/ICWSM/article/view/18113 Voterfraud2020: a multi-modal dataset of election fraud claims on twitter . In Proceedings of the Fifteenth International AAAI Conference on Web and Social Media, ICWSM 2021, held virtually, June 7-10, 2021 , pages 901--912. AAAI Press

    work page 2021

  4. [4]

    Khadijat Oluwatoyin Ajao. 2017. https://repository.up.ac.za/handle/2263/65543 Citizen journalism and conflict in Africa : the Ushahidi Platform in Kenya 's 2008 post-election violence . Thesis, University of Pretoria. Accepted: 2018-07-16T07:53:39Z

    work page 2017

  5. [5]

    Cynthia Amol, Lilian Wanzare, and James Obuhuma. 2024. Politikweli: A swahili-english code-switched twitter political misinformation classification dataset. In Speech and Language Technologies for Low-Resource Languages, pages 3--17, Cham. Springer Nature Switzerland

    work page 2024

  6. [6]

    Mateusz Baran, Mateusz W \'o jcik, Piotr Kolebski, Micha Bernaczyk, Krzysztof Rajda, Lukasz Augustyniak, and Tomasz Kajdanowicz. 2022. https://aclanthology.org/2022.politicalnlp-1.5 Electoral agitation dataset: The use case of the P olish election . In Proceedings of the LREC 2022 workshop on Natural Language Processing for Political Sciences, pages 32--3...

    work page 2022

  7. [7]

    Adam Bermingham and Alan Smeaton. 2011. https://aclanthology.org/W11-3702 On using T witter to monitor political sentiment and predict election results . In Proceedings of the Workshop on Sentiment Analysis where AI meets Psychology ( SAAIP 2011) , pages 2--10, Chiang Mai, Thailand. Asian Federation of Natural Language Processing

    work page 2011

  8. [8]

    Ankita Bhaumik, Andy Bernhardt, Gregorios Katsios, Ning Sa, and Tomek Strzalkowski. 2023. https://doi.org/10.18653/v1/2023.wassa-1.38 Adapting emotion detection to analyze influence campaigns on social media . In Proceedings of the 13th Workshop on Computational Approaches to Subjectivity, Sentiment, & Social Media Analysis , pages 441--451, Toronto, Cana...

    doi:10.18653/v1/2023.wassa-1.38 2023

Show all 48 references
  1. [9]

    Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert - Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jef...

    2020

  2. [10]

    Abhinav Chinta, Jingyu Zhang, Alexandra DeLucia, Mark Dredze, and Anna L. Buczak. 2021. https://doi.org/10.18653/v1/2021.wnut-1.44 Study of manifestation of civil unrest on T witter . In Proceedings of the Seventh Workshop on Noisy User-generated Text (W-NUT 2021), pages 396--...

    2021 doi

  3. [11]

    Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzm \'a n, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. https://doi.org/10.18653/v1/2020.acl-main.747 Unsupervised cross-lingual representation learning ...

    2020 doi

  4. [12]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. https://doi.org/10.18653/v1/N19-1423 BERT : Pre-training of deep bidirectional transformers for language understanding . In Proceedings of the 2019 Conference of the North A merican Chapter of the Associat...

    2019 doi

  5. [13]

    Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al - Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, Anirudh Goyal, Anthony Hartshorn, Aobo Yang, Archi Mitra, Archie Sravankumar, Artem Korenev, Arthur Hinsvark, Arun Rao, Aston Z...

    2024 doi arXiv

  6. [14]

    Nils Constantin Hellwig, Markus Bink, Thomas Schmidt, Jakob Fehle, and Christian Wolff. 2023. https://aclanthology.org/2023.icnlsp-1.9 Transformer-based analysis of sentiment towards G erman political parties on T witter during the 2021 election year . In Proceedings of the 6t...

    2023

  7. [15]

    Albert Q. Jiang, Alexandre Sablayrolles, Antoine Roux, Arthur Mensch, Blanche Savary, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Emma Bou Hanna, Florian Bressand, Gianna Lengyel, Guillaume Bour, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne...

    2024 arXiv

  8. [16]

    Armand Joulin, Edouard Grave, Piotr Bojanowski, and Tomas Mikolov. 2017. https://aclanthology.org/E17-2068 Bag of tricks for efficient text classification . In Proceedings of the 15th Conference of the E uropean Chapter of the Association for Computational Linguistics: Volume ...

    2017

  9. [17]

    David Jurgens, Tyler Finethy, James McCorriston, Yi Tian Xu, and Derek Ruths. 2015. http://www.aaai.org/ocs/index.php/ICWSM/ICWSM15/paper/view/10584 Geolocation prediction in twitter using social networks: A critical analysis and review of current practice . In Proceedings of ...

    2015

  10. [18]

    KNBS. 2019. https://www.knbs.or.ke/2019-kenya-population-and-housing-census-results/ 2019 K enya population and housing census . Technical report, Kenya National Bureau of Statistics

    2019

  11. [19]

    Menglin Li and Kwan Hui Lim. 2024. https://arxiv.org/abs/2403.00786 Leveraging contrastive learning for few-shot geolocation of social posts . ArXiv preprint, abs/2403.00786

    2024 arXiv

  12. [20]

    Chin-Yew Lin. 2004. https://aclanthology.org/W04-1013 ROUGE : A package for automatic evaluation of summaries . In Text Summarization Branches Out, pages 74--81, Barcelona, Spain. Association for Computational Linguistics

    2004

  13. [21]

    Marina Litvak, Natalia Vanetik, Sagiv Talker, and Or Machlouf. 2022. https://aclanthology.org/2022.trac-1.9 Detection of negative campaign in israeli municipal elections . In Proceedings of the Third Workshop on Threat, Aggression and Cyberbullying (TRAC 2022), pages 68--74, G...

    2022

  14. [22]

    Ilya Loshchilov and Frank Hutter. 2019. https://openreview.net/forum?id=Bkg6RiCqY7 Decoupled weight decay regularization . In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019 . OpenReview.net

    2019

  15. [23]

    Mary McHugh. 2012. https://doi.org/10.11613/BM.2012.031 Interrater reliability: The kappa statistic . Biochemia medica : časopis Hrvatskoga društva medicinskih biokemičara / HDMB, 22:276--82

    2012 doi

  16. [24]

    Sovesh Mohapatra and Somesh Mohapatra. 2022. https://aclanthology.org/2022.nlp4dh-1.3 Sentiment is all you need to win US presidential elections . In Proceedings of the 2nd International Workshop on Natural Language Processing for Digital Humanities, pages 15--20, Taipei, Taiw...

    2022

  17. [25]

    Dumisani Moyo. 2009. https://doi.org/10.1080/14616700902797291 Citizen Journalism and the Parallel Market of Information in Zimbabwe 's 2008 Election . Journalism Studies, 10(4):551--567. Publisher: Routledge \_eprint: https://doi.org/10.1080/14616700902797291

    2009 doi

  18. [26]

    Leonard Muaka. 2011. https://api.semanticscholar.org/CorpusID:149109464 Language perceptions and identityamong kenyan speakers . In Proceedings of the 40th Annual Conference on African Linguistics

    2011

  19. [27]

    Martin N Ndlela. 2020. Social media algorithms, bots and elections in africa. Social media and elections in Africa, Volume 1: Theoretical perspectives and election campaigns, pages 13--37

    2020

  20. [28]

    Millicent Ochieng, Varun Gumma, Sunayana Sitaram, Jindong Wang, Vishrav Chaudhary, Keshet Ronen, Kalika Bali, and Jacki O'Neill. 2024. https://doi.org/10.48550/arXiv.2406.00343 Beyond metrics: Evaluating llms' effectiveness in culturally nuanced, low-resource real-world scenar...

    2024 doi arXiv

  21. [29]

    Brian Okoth. 2022. https://web.archive.org/web/20220731222119/https://standardmedia.co.ke/national/article/2001450960/chebukati-venezuelans-key-in-aug-9-polls-technology-roll-out-arrested-at-jkia Chebukati: Three V enezuelans key in A ug. 9 polls technology roll-out arrested a...

    2022

  22. [30]

    OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, Red Avila, Igor Babuschkin, Suchir Balaji, Valerie Balcom, Paul Baltescu, Haiming Bao, Mohammad Bavarian, Jeff ...

    2024 arXiv

  23. [31]

    Clionadh Raleigh, rew Linke, Håvard Hegre, and Joakim Karlsen. 2010. https://doi.org/10.1177/0022343310378914 Introducing acled: An armed conflict location and event dataset . Journal of Peace Research, 47(5):651--660

    2010 doi

  24. [32]

    Eric Sanders and Antal van den Bosch. 2020. https://aclanthology.org/2020.lrec-1.756 Optimising T witter-based political election prediction with relevance and S entiment filters . In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 6158--6165, Ma...

    2020

  25. [33]

    Thomas Schmidt, Jakob Fehle, Maximilian Weissenbacher, Jonathan Richter, Philipp Gottschalk, and Christian Wolff. 2022. https://aclanthology.org/2022.konvens-1.9 Sentiment analysis on T witter for the major G erman parties during the 2021 G erman federal election . In Proceedi...

    2022

  26. [34]

    Philipp Seeberger and Korbinian Riedhammer. 2022. https://doi.org/10.18653/v1/2022.nlp4pi-1.9 Enhancing crisis-related tweet classification with entity-masked language modeling and multi-task learning . In Proceedings of the Second Workshop on NLP for Positive Impact (NLP4PI),...

    2022 doi

  27. [35]

    Deodatus Patrick Shayo. 2021. Citizen participation in local government elections in the age of crowdsourcing: Explorations and considerations in tanzania. Program on Governance and Local Development Working Paper, (47)

    2021

  28. [36]

    Ralph Sundberg and Erik Melander. 2013. https://doi.org/10.1177/0022343313484347 Introducing the ucdp georeferenced event dataset . Journal of Peace Research, 50(4):523--532

    2013 doi

  29. [37]

    Simone Tedeschi, Valentino Maiorca, Niccol \`o Campolungo, Francesco Cecconi, and Roberto Navigli. 2021. https://doi.org/10.18653/v1/2021.findings-emnlp.215 W iki NE u R al: C ombined neural and knowledge-based silver data creation for multilingual NER . In Findings of the Ass...

    2021 doi

  30. [38]

    Erik Tjong Kim Sang and Johan Bos. 2012. https://aclanthology.org/W12-0607 Predicting the 2011 D utch senate election results with T witter . In Proceedings of the Workshop on Semantic Analysis in Social Media, pages 53--60, Avignon, France. Association for Computational Linguistics

    2012

  31. [39]

    Ushahidi. 2022 a . https://www.ushahidi.com/in-action/ushahidi-s-uchaguzi-platform-empowers-citizen-participation-in-2022-kenya-general-elections/ Uchaguzi platform

    2022

  32. [40]

    Ushahidi. 2022 b . https://docs.ushahidi.com/uchaguzi-support/digital-response-teams/digital-response-teams-overview Ushahidi's digital response teams

    2022

  33. [41]

    Ushahidi. 2022 c . https://docs.ushahidi.com/uchaguzi-support/digital-response-teams/publishing Ushahidi's publishing team

    2022

  34. [42]

    Ushahidi. 2024. https://www.ushahidi.com/about/blog/situational-report-3-an-analysis-of-the-sentiments-around-the-protests-in-kenya/ Report on sentiment around protests in kenya

    2024

  35. [43]

    Stephen Wan and C \'e cile Paris. 2015. https://doi.org/10.18653/v1/W15-3707 Ranking election issues through the lens of social media . In Proceedings of the 9th SIGHUM Workshop on Language Technology for Cultural Heritage, Social Sciences, and Humanities ( L a T e CH ) , page...

    2015 doi

  36. [44]

    Hao Wang, Dogan Can, Abe Kazemzadeh, Fran c ois Bar, and Shrikanth Narayanan. 2012. https://aclanthology.org/P12-3020 A system for real-time T witter sentiment analysis of 2012 U . S . presidential election cycle . In Proceedings of the ACL 2012 System Demonstrations , pages 1...

    2012

  37. [45]

    Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mari...

    2020 doi

  38. [46]

    Shi-Tao Yeh et al. 2002. Using trapezoidal rule for the area under a curve calculation. Proceedings of the 27th Annual SAS User Group International (SUGI’02) , pages 1--5

    2002

  39. [47]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  40. [48]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.