Swedish Sentiment Violence

🚀 Swedish BERT Models for Sentiment Analysis

Recorded Future, in collaboration with AI Sweden, presents two Swedish language models tailored for sentiment analysis. These models are built upon the KB/bert-base-swedish-cased model and fine - tuned to handle multi - label sentiment analysis tasks.

These models have been optimized for detecting sentiments related to fear and violence. They output three floating - point numbers corresponding to the labels "Negative", "Weak sentiment", and "Strong Sentiment" at their respective indices.

The models were trained on Swedish conversational data gathered from diverse internet sources and forums. They are exclusively trained on Swedish data and support inference only for Swedish input texts. Inference metrics for non - Swedish inputs are undefined, and such inputs are regarded as out - of - domain data.

The current models are supported by Transformers version >= 4.3.3 and Torch version 1.8.0. Compatibility with older versions has not been verified.

✨ Features

• Sentiment Focus: Fine - tuned for fear and violence sentiment analysis.
• Swedish - Specific: Trained solely on Swedish data for accurate Swedish text inference.
• Multi - Label Output: Provides three - label sentiment output.

📦 Installation

The models can be imported from the transformers library.

Swedish - Sentiment - Fear

from transformers import BertForSequenceClassification, BertTokenizerFast

tokenizer = BertTokenizerFast.from_pretrained("RecordedFuture/Swedish-Sentiment-Fear")
classifier_fear = BertForSequenceClassification.from_pretrained("RecordedFuture/Swedish-Sentiment-Fear")

Swedish - Sentiment - Violence

from transformers import BertForSequenceClassification, BertTokenizerFast

tokenizer = BertTokenizerFast.from_pretrained("RecordedFuture/Swedish-Sentiment-Violence")
classifier_violence = BertForSequenceClassification.from_pretrained("RecordedFuture/Swedish-Sentiment-Violence")

💻 Usage Examples

Basic Usage

After initializing the model and tokenizer, you can use the model for inference.

# For Swedish - Sentiment - Fear
from transformers import BertForSequenceClassification, BertTokenizerFast

tokenizer = BertTokenizerFast.from_pretrained("RecordedFuture/Swedish-Sentiment-Fear")
classifier_fear = BertForSequenceClassification.from_pretrained("RecordedFuture/Swedish-Sentiment-Fear")

# Prepare input text
input_text = "Your Swedish text here"
inputs = tokenizer(input_text, return_tensors='pt')

# Perform inference
outputs = classifier_fear(**inputs)

Advanced Usage

The models are optimized for specific sentiment analysis tasks. You can use them for more complex scenarios by fine - tuning the input pre - processing or post - processing steps.

# For Swedish - Sentiment - Violence
from transformers import BertForSequenceClassification, BertTokenizerFast

tokenizer = BertTokenizerFast.from_pretrained("RecordedFuture/Swedish-Sentiment-Violence")
classifier_violence = BertForSequenceClassification.from_pretrained("RecordedFuture/Swedish-Sentiment-Violence")

# Prepare input text
input_text = "Your Swedish text here"
inputs = tokenizer(input_text, return_tensors='pt')

# Perform inference
outputs = classifier_violence(**inputs)

# Post - processing example
import torch
probabilities = torch.softmax(outputs.logits, dim = 1)

📚 Documentation

Sentiment Definitions

Swedish - Sentiment - Fear

Strong Sentiment

Texts that hold an expressive emphasis on fear and/ or anxiety.

Weak Sentiment

Texts that express fear and/ or anxiety in a neutral way.

Swedish - Sentiment - Violence

Strong Sentiment

Texts that reference highly violent acts or hold an aggressive tone.

Weak Sentiment

Texts that include general violent statements that do not fall under the strong sentiment.

Verification Metrics

Swedish - Sentiment - Fear

Swedish - Sentiment - Violence

📄 License

This project is released under the MIT license.