classla/multilingual-IPTC-news-topic-classifier

Multilingual IPTC Media Topic Classifier

News topic classification model based on xlm-roberta-large and fine-tuned on a news corpus in 4 languages (Croatian, Slovenian, Catalan and Greek), annotated with the top-level IPTC Media Topic NewsCodes labels.

The model can be used for classification into topic labels from the IPTC NewsCodes schema and can be applied to any news text in a language, supported by the xlm-roberta-large.

Based on a manually-annotated test set (in Croatian, Slovenian, Catalan and Greek), the model achieves macro-F1 score of 0.746, micro-F1 score of 0.734, and accuracy of 0.734, and outperforms the GPT-4o model (version gpt-4o-2024-05-13) used in a zero-shot setting. If we use only labels that are predicted with a confidence score equal or higher than 0.90, the model achieves micro-F1 and macro-F1 of 0.80.

Intended use and limitations

For reliable results, the classifier should be applied to documents of sufficient length (the rule of thumb is at least 75 words).

Use example:

from transformers import pipeline

# Load a multi-class classification pipeline - if the model runs on CPU, comment out "device"
classifier = pipeline("text-classification", model="classla/multilingual-IPTC-news-topic-classifier", device=0, max_length=512, truncation=True)

# Example texts to classify
texts = [
    """Slovenian handball team makes it to Paris Olympics semifinal Lille, 8 August - Slovenia defeated Norway 33:28 in the Olympic men's handball tournament in Lille late on Wednesday to advance to the semifinal where they will face Denmark on Friday evening. This is the best result the team has so far achieved at the Olympic Games and one of the best performances in the history of Slovenia's team sports squads.""",
    """Moment dog sparks house fire after chewing power bank An indoor monitoring camera shows the moment a dog unintentionally caused a house fire after chewing on a portable lithium-ion battery power bank. In the video released by Tulsa Fire Department in Oklahoma, two dogs and a cat can be seen in the living room before a spark started the fire that spread within minutes. Tulsa Fire Department public information officer Andy Little said the pets escaped through a dog door, and according to local media the family was also evacuated safely. "Had there not been a dog door, they very well could have passed away," he told CBS affiliate KOTV."""]

# Classify the texts
results = classifier(texts)

# Output the results
for result in results:
    print(result)

## Output
## {'label': 'sport', 'score': 0.9985264539718628}
## {'label': 'disaster, accident and emergency incident', 'score': 0.9957459568977356}

IPTC Media Topic categories

The classifier uses the top-level of the IPTC Media Topic NewsCodes schema, consisting of 17 labels.

List of labels

labels_list=['education', 'human interest', 'society', 'sport', 'crime, law and justice',
'disaster, accident and emergency incident', 'arts, culture, entertainment and media', 'politics',
'economy, business and finance', 'lifestyle and leisure', 'science and technology',
'health', 'labour', 'religion', 'weather', 'environment', 'conflict, war and peace'],

labels_map={0: 'education', 1: 'human interest', 2: 'society', 3: 'sport', 4: 'crime, law and justice',
5: 'disaster, accident and emergency incident', 6: 'arts, culture, entertainment and media',
7: 'politics', 8: 'economy, business and finance', 9: 'lifestyle and leisure', 10: 'science and technology',
11: 'health', 12: 'labour', 13: 'religion', 14: 'weather', 15: 'environment', 16: 'conflict, war and peace'}

Description of labels

The descriptions of the labels are based on the descriptions provided in the IPTC Media Topic NewsCodes schema and enriched with information which specific subtopics belong to the top-level topics, based on the IPTC Media Topic label hierarchy.

Label	Description
disaster, accident and emergency incident	Man-made or natural events resulting in injuries, death or damage, e.g., explosions, transport accidents, famine, drowning, natural disasters, emergency planning and response.
human interest	News about life and behavior of royalty and celebrities, news about obtaining awards, ceremonies (graduation, wedding, funeral, celebration of launching something), birthdays and anniversaries, and news about silly or stupid human errors.
politics	News about local, regional, national and international exercise of power, including news about election, fundamental rights, government, non-governmental organisations, political crises, non-violent international relations, public employees, government policies.
education	All aspects of furthering knowledge, formally or informally, including news about schools, curricula, grading, remote learning, teachers and students.
crime, law and justice	News about committed crime and illegal activities, the system of courts, law and law enforcement (e.g., judges, lawyers, trials, punishments of offenders).
economy, business and finance	News about companies, products and services, any kind of industries, national economy, international trading, banks, (crypto)currency, business and trade societies, economic trends and indicators (inflation, employment statistics, GDP, mortgages, ...), international economic institutions, utilities (electricity, heating, waste management, water supply).
conflict, war and peace	News about terrorism, wars, wars victims, cyber warfare, civil unrest (demonstrations, riots, rebellions), peace talks and other peace activities.
arts, culture, entertainment and media	News about cinema, dance, fashion, hairstyle, jewellery, festivals, literature, music, theatre, TV shows, painting, photography, woodworking, art exhibitions, libraries and museums, language, cultural heritage, news media, radio and television, social media, influencers, and disinformation.
labour	News about employment, employment legislation, employees and employers, commuting, parental leave, volunteering, wages, social security, labour market, retirement, unemployment, unions.
weather	News about weather forecasts, weather phenomena and weather warning.
religion	News about religions, cults, religious conflicts, relations between religion and government, churches, religious holidays and festivals, religious leaders and rituals, and religious texts.
society	News about social interactions (e.g., networking), demographic analyses, population census, discrimination, efforts for inclusion and equity, emigration and immigration, communities of people and minorities (LGBTQ, older people, children, indigenous people, etc.), homelessness, poverty, societal problems (addictions, bullying), ethical issues (suicide, euthanasia, sexual behavior) and social services and charity, relationships (dating, divorce, marriage), family (family planning, adoption, abortion, contraception, pregnancy, parenting).
health	News about diseases, injuries, mental health problems, health treatments, diets, vaccines, drugs, government health care, hospitals, medical staff, health insurance.
environment	News about climate change, energy saving, sustainability, pollution, population growth, natural resources, forests, mountains, bodies of water, ecosystem, animals, flowers and plants.
lifestyle and leisure	News about hobbies, clubs and societies, games, lottery, enthusiasm about food or drinks, car/motorcycle lovers, public holidays, leisure venues (amusement parks, cafes, bars, restaurants, etc.), exercise and fitness, outdoor recreational activities (e.g., fishing, hunting), travel and tourism, mental well-being, parties, maintaining and decorating house and garden.
science and technology	News about natural sciences and social sciences, mathematics, technology and engineering, scientific institutions, scientific research, scientific publications and innovation.
sport	News about sports that can be executed in competitions, e.g., basketball, football, swimming, athletics, chess, dog racing, diving, golf, gymnastics, martial arts, climbing, etc.; sport achievements, sport events, sport organisation, sport venues (stadiums, gymnasiums, ...), referees, coaches, sport clubs, drug use in sport.

Training data

The model was fine-tuned on the training split of the EMMediaTopic 1.0 dataset consisting of 15,000 news in four languages (Croatian, Slovenian, Catalan and Greek). The news texts were extracted from the MaCoCu-Genre web corpora based on the "News" genre label, predicted with the X-GENRE classifier. The training dataset was automatically annotated with the IPTC Media Topic labels by the GPT-4o model (yielding 0.72 micro-F1 and 0.73 macro-F1 on the test dataset).

The code for the development and evaluation of the model is available on this GitHub repository.

Label distribution in the training dataset:

labels	count	proportion
sport	2300	0.153333
arts, culture, entertainment and media	2117	0.141133
politics	2018	0.134533
economy, business and finance	1670	0.111333
human interest	1152	0.0768
education	990	0.066
crime, law and justice	884	0.0589333
health	675	0.045
disaster, accident and emergency incident	610	0.0406667
society	481	0.0320667
environment	472	0.0314667
lifestyle and leisure	346	0.0230667
science and technology	340	0.0226667
conflict, war and peace	311	0.0207333
labour	288	0.0192
religion	258	0.0172
weather	88	0.00586667

Performance

The model was evaluated on a manually-annotated test set in four languages (Croatian, Slovenian, Catalan and Greek), consisting of 1,129 instances. The test set contains similar amounts of texts from the four languages and is more or less balanced across labels.

The model was shown to achieve micro-F1 score of 0.734, and macro-F1 score of 0.746. The results for the entire test set and per language:

	Micro-F1	Macro-F1	Accuracy	No. of instances
All (combined)	0.734278	0.745864	0.734278	1129
Croatian	0.728522	0.733725	0.728522	291
Catalan	0.715356	0.722304	0.715356	267
Slovenian	0.758865	0.764784	0.758865	282
Greek	0.733564	0.747129	0.733564	289

Performance per label:

	precision	recall	f1-score	support
arts, culture, entertainment and media	0.602151	0.875	0.713376	64
conflict, war and peace	0.611111	0.916667	0.733333	36
crime, law and justice	0.861538	0.811594	0.835821	69
disaster, accident and emergency incident	0.691176	0.886792	0.77686	53
economy, business and finance	0.779221	0.508475	0.615385	118
education	0.847458	0.735294	0.787402	68
environment	0.589041	0.754386	0.661538	57
health	0.79661	0.79661	0.79661	59
human interest	0.552239	0.672727	0.606557	55
labour	0.855072	0.830986	0.842857	71
lifestyle and leisure	0.773585	0.476744	0.589928	86
politics	0.568182	0.735294	0.641026	68
religion	0.842105	0.941176	0.888889	51
science and technology	0.637681	0.8	0.709677	55
society	0.918033	0.5	0.647399	112
sport	0.824324	0.968254	0.890511	63
weather	0.953488	0.931818	0.942529	44

For downstream tasks, we advise you to use only labels that were predicted with confidence score higher or equal to 0.90 which further improves the performance.

When we remove instances predicted with lower confidence (229 instances - 20%), the model yields micro-F1 of 0.798 and macro-F1 of 0.80.

	Micro-F1	Macro-F1	Accuracy
All (combined)	0.797777	0.802403	0.797777
Croatian	0.773504	0.772084	0.773504
Catalan	0.811224	0.806885	0.811224
Slovenian	0.805085	0.804491	0.805085
Greek	0.803419	0.809598	0.803419

Fine-tuning hyperparameters

Fine-tuning was performed with simpletransformers. Beforehand, a brief hyperparameter optimization was performed and the presumed optimal hyperparameters are:

model_args = ClassificationArgs()

model_args ={
             "num_train_epochs": 5,
             "learning_rate": 8e-06,
             "train_batch_size": 32,
             "max_seq_length": 512,
             }   
      

Citation

Paper with the details on the model is currently in submission. If you use the model, please cite this paper:

@article{kuzman2024llmteacherstudent,
      title={{LLM Teacher-Student Framework for Text Classification With No Manually Annotated Data: A Case Study in IPTC News Topic Classification}}, 
      author={Kuzman, Taja and Ljube{\v{s}}i{\'c}, Nikola},
      journal={arXiv preprint arXiv:2411.19638},
      year={2024}
}

Funding

This work was supported by the Slovenian Research and Innovation Agency research project Embeddings-based techniques for Media Monitoring Applications (L2-50070, co-funded by the Kliping d.o.o. agency).