chrlukas/flattery_prediction_text

Flattery Prediction from Text

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

This model was finetuned to predict flattery in transcripts of English earning calls. It was introduced in This Paper Had the Smartest Reviewers -- Flattery Detection Utilising an Audio-Textual Transformer-Based Approach, which was published at INTERSPEECH 2024.

Model Details

Model Description

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

This is a fine-tuned variant of RoBERTa-base. It is trained using a dataset comprising single sentences uttered in business calls, which were labeled for flattery in a binary manner. The training set comprised 7167 sentences, 1878 sentences were used as development set. For more details, please refer to the paper, especially Sections 2 for the dataset, 3.1 for the training procedure and 4.1 for the results. The checkpoint provided here was trained using human gold-standard transcripts. It achieves Unweighed Average Recall (UAR) values of .8512 and .8865 on the development and test partition, respectively.

Model Sources

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

• —Repository: https://github.com/lc0197/flattery_from_speech
• —Paper: This Paper Had the Smartest Reviewers -- Flattery Detection Utilising an Audio-Textual Transformer-Based Approach (Interspeech 2024)

Uses

The following snippet illustrates the usage of the model.

from transformers import AutoTokenizer, AutoModelForSequenceClassification
from torch import sigmoid
import torch

# initialize model and tokenizer
checkpoint = "chrlukas/flattery_prediction_text"
tokenizer = AutoTokenizer.from_pretrained(checkpoint)
model = AutoModelForSequenceClassification.from_pretrained(checkpoint)
model.eval()

# predict flattery in a sentence
example = 'This is a great example!'    # should predict flattery
tokenized = tokenizer(example, return_tensors='pt')
with torch.no_grad():
  logits = model(**tokenized).logits
prediction = sigmoid(logits).item()
flattery = prediction >= 0.5
print(f'Flattery detected? {flattery}')

Bias, Risks, and Limitations

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

The model is trained on a highly-domain specific dataset sourced from earning calls, i.e., typically conversations between business analysts and CEOs of US-American companies. Hence, it can not be expected to generalize well to other domains and contexts.

Citation

<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->

BibTeX:

@inproceedings{christ24_interspeech,
  title     = {{This Paper Had the Smartest Reviewers - Flattery Detection Utilising an Audio-Textual Transformer-Based Approach}},
  author    = {Lukas Christ and Shahin Amiriparian and Friederike Hawighorst and Ann-Kathrin Schill and Angelo Boutalikakis and Lorenz Graf-Vlachy and Andreas König and Björn Schuller},
  year      = {2024},
  booktitle = {{Interspeech 2024}},
  pages     = {3530--3534},
  doi       = {10.21437/Interspeech.2024-87},
  issn      = {2958-1796},
}