Model Card for raicrits/topicChangeDetector_v1

This model analyses the input text and provides an answer whether in the text there is a change of topic or not (resp. TOPPICCHANGE, SAMETOPIC).

Model Details

Model Description

Developed by: Alberto Messina ([email protected])
Model type: BERT for Sequence Classification
Language(s) (NLP): Italian
License: TBD
Finetuned from model: https://huggingface.co/xlm-roberta-base

Model Sources [optional]

Repository: N/A
Paper [optional]: N/A
Demo [optional]: N/A

Uses

The model should be used giving as input a short paragraph of text taken from a news programme or article in Italian about which it is requested to get an answer about whether or not it contains a change of topic. The model has been trained to detect topic changes without apriori knowledge of possible points of separation (e.g., paragraphs or speaker turns). For this reason it tends to be sensitive to the amount of text supposed to belong to either of the two subsequent topics, and therefore performs better when the sought for topic change occurs approximately in the middle of the input. To reduce the impact of this issue, it is suggested to use the model on a sequence of partially overlapping pieces of text taken from the document to be analysed, and to further process the results sequence to consolidate a decision.

Direct Use

TBA

Out-of-Scope Use

The model should not be used as a general purpose topic change detector, i.e. on text which is not originated from news programme transcription or siilar content.

Bias, Risks, and Limitations

The training dataset is made up of automatic transcriptions from RAI Italian newscasts, therefore there is an intrinsic bias in the kind of topics that can be tracked for change.

How to Get Started with the Model

Use the code below to get started with the model.

TBA

Training Details

Training Data

TBA

Training Procedure

Preprocessing [optional]

TBA

Training Hyperparameters

Training regime: Mixed Precision

Evaluation

TBA

Testing Data, Factors & Metrics

Testing Data

TBA

Metrics

TBA

Results

TBA

Summary

TBA

Environmental Impact

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

Hardware Type: 2 NVIDIA A100/40Gb
Hours used: 2
Cloud Provider: Private Infrastructure
Carbon Emitted: 0.22 kg CO2 eq.

Glossary [optional]

TBA

More Information [optional]

The development of this model is partially supported by H2020 Project AI4Media - A European Excellence Centre for Media, Society and Democracy (Grant nr. 951911) - http://ai4media.eu

Model Card Authors [optional]

Alberto Messina

Model Card Contact

[email protected]