---
license: other
language:
- it
pipeline_tag: text-classification
widget:
- text: >-
    Ripartire la parola d'ordine, al governo chiediamo di accelerare la campagna
    sui vaccini e di lavorare a un cronoprogramma delle riaperture. Dobbiamo
    dare una prospettiva di rinascita a tutti gli italiani, dall'opposizione
    ancora all'attacco del governo, gli italiani sono esausti di fare sacrifici
    che non portano a nulla. Sono quattro le persone indagate dalla Procura di
    Roma per le minacce via mail al ministro della Salute. Tra ottobre del 2020
    e il gennaio del 2021 avrebbero inviato al ministro dei messaggi dal
    contenuto gravemente minaccioso. Al ministro la solidarietà di tutto il
    mondo politico e a causa della pandemia si assottigliano i redditi delle
    famiglie italiane. Aumenta anche la pressione fiscale. Lo rileva l'Istat.
- text: >-
    Le terapie intensive hanno superato la soglia del 30% di riempimento. La
    lotta al virus e anche lotta alle fake news, prosegue la collaborazione tra
    ministero della Salute e Twitter quando si cercano notizie sul Covid del
    Social rimanda le pagine del ministero, includendo anche le ultime
    informazioni sui vaccini. COVID-19 è stato l'hashtag più twittato a livello
    globale nel 2020. La poltrona negata da Erdogan ad Ursula von der Leyen, lo
    avete sentito? Fa ancora discutere dentro e fuori dal Parlamento europeo:
    Marco Clementi. 
- text: >-
    I bambini che soffrono di autismo hanno gli stessi diritti di tutti gli
    altri bambini sottolinea garante per l'infanzia, occorre dunque fare rete
    tra famiglia, scuola, pediatri e servizi sociali. Domani mattina alle 705 su
    Rai Uno torna la nostra rubrica di approfondimento 7 giorni. L'anticipazione
    nel servizio.
metrics:
- accuracy
- precision
- recall
---
# Model Card for raicrits/topicChangeDetector_v1

<!-- Provide a quick summary of what the model is/does. -->

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

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



- **Developed by:** Alberto Messina (alberto.messina@rai.it)
- **Model type:** BERT for Sequence Classification
- **Language(s) (NLP):** Italian
- **License:** TBD
- **Finetuned from model:** https://huggingface.co/xlm-roberta-base

### Model Sources [optional]

<!-- Provide the basic links for the model. -->

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

## 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. -->
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

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

TBA

### Out-of-Scope Use

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

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

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

<!-- This should link to a Data 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. -->

TBA

### 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]

TBA


#### Training Hyperparameters

- **Training regime:** Mixed Precision 

## Evaluation

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

### Testing Data, Factors & Metrics

#### Testing Data

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

TBA

#### Metrics

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

TBA

### Results

TBA

#### Summary

TBA

## Environmental Impact

<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->

Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).

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

## Glossary [optional]

<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->

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

alberto.messina@rai.it