datavaluepeople/Afxumo-toxicity-somaliland-SO

Model Card for Model ID

This model classifies text - specifically twitter posts - on if they contain afxumo (toxicity in english). It has been fine-tuned on a small sample (n=5910) of tweets, which are only slightly imbalanced (nafxumo=3314, nnot_afxumo=2596). These tweets have been collected on 3 March 2025, for tweets in February 2025. They were collected focused on the Somaliland conflict using queries. The language in the dataset is predominantly Somali, with some English interspersed. It is also slang-heavy as it is social media data.

Model Details

Model Description

<!-- Provide a longer summary of what this model is. -->

• —Developed by: datavaluepeople in conjunction with How to Build UP for a project to find toxic speech.
• —Language(s) (NLP): Somali, English
• —License: agpl-3.0
• —Finetuned from model: xlm-roberta-base-finetuned-somali

Uses

<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. --> This model is intended for use by (digital) peacebuilders and anyone working towards depolarising the digital public sphere. We have used it in conjunction with Phoenix, an ethical, open source, non-commercial platform designed to enable peacebuilders and mediators to conduct ethical and participatory social media listening in order to inform conflict transformation work. There is a lot of social media data, and this allows peacebuilders to quickly find instances of comments that discourage the dialogue to continue through intimidation of the other participants. The peacebuilder can then use this to inform programming, as well as conduct targeted interventions to mitigate that intimidation.

Direct Use

<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->

[More Information Needed]

Downstream Use [optional]

<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->

[More Information Needed]

Out-of-Scope Use

<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->

[More Information Needed]

Bias, Risks, and Limitations

<!-- This section is meant to convey both technical and sociotechnical limitations. -->

Due to disagreements between annotators on what does and doesn't constitute afxumo, this model was trained on a single annotator's opinion on what is or is not toxic. As such, this model will contain biases of that annotator, and will not give a complete view on toxicity in this context.

Recommendations

<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.

How to Get Started with the Model

Use the code below to get started with the model.

[More Information Needed]

Training Details

Training Data

<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->

[More Information Needed] Annotators were asked to annotate X/Twitter data with the following rubric. Due to annotation contstraints and low prevalence of some of the labels, all of the labels were coalesced into a single label (afxumo)

· Afxumo – Boost-ku ma ka kooban tahay hadal cay ah, xushmad darro ah, ama comment aan sax ahayn oo ku aadan qof ama koox gaara oo ku kalifta isticmaalaha baraha bulshada inuu ka baxo ama ka tago dooda socota. · Afxumo lafojebis ah (aad u xun)- Boost-ku ma ka koobanyahay hadal nacayb daran ah ama afxumo lafojebis ah oo ku aadan qof ama koox gaara oo ku kalifi karta istcimaalaha baraha bulshada inuu ka baxo ama ka tago dooda socota · Cunsuriyad – Boost-ku ma ka kooban yahay naceyb, nasab cay, iyo cunsuriyad lagu beegsanayo nasabka ama haybta qof ama koox gaara · Cay – Boost-ku ma ka kooban tahay aflagaado, kicin, ama faallo taban oo ku wajahan qof ama koox dad ah · Xadgudub Jinsi – Boost-ku ma ka kooban wax ku saabsan xadugudub jinsi oo lagu beegsanayo qof ama koox gaara sida ayad oo jinsiga ama xubnaha qofka wax laga sheegayo · Habaar– Boost-ku ma ka kooban tahay erayo habaar ah, inkaarid ah, ama hadal kale oo fisqi ah? · Hanjabaad – Boost-ku ma ka koobanyahay hanjabaad, wax yeelid, ama dhibaateyn qof ama koox gaara.

Training Procedure

<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->

Preprocessing [optional]

[More Information Needed]

Training Hyperparameters

• —Training regime: [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->

Speeds, Sizes, Times [optional]

<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->

[More Information Needed]

Evaluation

<!-- This section describes the evaluation protocols and provides the results. -->

Testing Data, Factors & Metrics

Testing Data

<!-- This should link to a Dataset Card if possible. -->

[More Information Needed]

Factors

<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->

[More Information Needed]

Metrics

<!-- These are the evaluation metrics being used, ideally with a description of why. -->

[More Information Needed]

Results

[More Information Needed]

Summary