The Human Value Detection at CLEF 2024 task consists of two sub-tasks: the first is to detect the presence or absence of each of these 19 values, while the second is to detect whether the value is attained or constrained.

Our system introduces a cascade model approach for the detection and stance classification of the predefined set of human values. It consists of two subsystems: one for detecting the presence of each human value and another for establishing the stance (if the sentence attains or constrains) of each human value. Each subsystem is designed and fine-tuned separately using a DeBERTa model as base.

Subsystem 1: Its primary function is to identify the presence of human values within sentences. By combining the 'attained' and 'constrained' labels to indicate an overall presence, it streamlines the multi-label classification task, simplifying it to a binary classification for each of the 19 human values (presence vs. absence).
Subsystem 2: it receives the outputs of subsystem 1 and classifies the stance towards each present human value in a binary classification (attained vs. constrained). This system transforms the sentences dataset into premise-hypothesis pairs, where each sentence is the premise, a value is the hypothesis, and the “attained” and “constrained” labels are the stance.

Given that subsystem 1 focuses on detecting the presence of human values in the text, and subsystem 2 focuses on the stances towards each detected human value, this cascade model approach improves the granularity of text classification.

This model is the responsible of the Subsystem 1 and accomplishes the first sub-task.

How to use

You can use this model using a text classification pipeline, as in the example:

python

from transformers import pipeline

model = "VictorYeste/deberta-based-human-value-detection"
tokenizer = "VictorYeste/deberta-based-human-value-detection"

values_detection = pipeline("text-classification", model=model, tokenizer=tokenizer, top_k=None)

values_detection("We would like to share this model with the research community.")

This returns the following:

[[{'label': 'Self-direction: thought', 'score': 0.02448045276105404},
  {'label': 'Stimulation', 'score': 0.01451807003468275},
  {'label': 'Universalism: concern', 'score': 0.006046739872545004},
  {'label': 'Self-direction: action', 'score': 0.004837467335164547},
  {'label': 'Benevolence: dependability', 'score': 0.001295178197324276},
  {'label': 'Benevolence: caring', 'score': 0.0009907316416501999},
  {'label': 'Conformity: interpersonal', 'score': 0.0004476217145565897},
  {'label': 'Security: societal', 'score': 0.00039295252645388246},
  {'label': 'Universalism: tolerance', 'score': 0.0003538706514518708},
  {'label': 'Power: dominance', 'score': 0.00016191638133022934},
  {'label': 'Power: resources', 'score': 0.0001522471575299278},
  {'label': 'Universalism: nature', 'score': 0.00014803129306528717},
  {'label': 'Humility', 'score': 0.0001100009903893806},
  {'label': 'Face', 'score': 9.083452459890395e-05},
  {'label': 'Conformity: rules', 'score': 8.524076838511974e-05},
  {'label': 'Achievement', 'score': 6.411433423636481e-05},
  {'label': 'Security: personal', 'score': 5.183048051549122e-05},
  {'label': 'Hedonism', 'score': 3.167059549014084e-05},
  {'label': 'Tradition', 'score': 2.4977327484521084e-05}]]

The model has been trained as a multi-label problem, so it can also be used to predict multiple labels as follows:

python

import torch
import numpy as np
import transformers

def multilabel_pipeline(text, model, tokenizer, id2label):
    # Code adapted from: https://github.com/NielsRogge/Transformers-Tutorials/blob/master/BERT/Fine_tuning_BERT_(and_friends)_for_multi_label_text_classification.ipynb
    """ Predicts the value probabilities (attained and constrained) for each sentence """
    encoding = tokenizer(text, return_tensors="pt")
    encoding = {k: v for k,v in encoding.items()}
    outputs = model(**encoding)
    logits = outputs.logits
    sigmoid = torch.nn.Sigmoid()
    probs = sigmoid(logits.squeeze().cpu())
    predictions = np.zeros(probs.shape)
    predictions[np.where(probs >= 0.5)] = 1
    predicted_labels = [id2label[idx] for idx, label in enumerate(predictions) if label == 1.0]
    return predicted_labels
     
values = ["Self-direction: thought", "Self-direction: action", "Stimulation",  "Hedonism", "Achievement", "Power: dominance", "Power: resources", "Face", "Security: personal", "Security: societal", "Tradition", "Conformity: rules", "Conformity: interpersonal", "Humility", "Benevolence: caring", "Benevolence: dependability", "Universalism: concern", "Universalism: nature", "Universalism: tolerance" ]
id2label = {idx:label for idx, label in enumerate(values)}
model_name = "VictorYeste/deberta-based-human-value-detection"
tokenizer = transformers.AutoTokenizer.from_pretrained(model_name)
model = transformers.AutoModelForSequenceClassification.from_pretrained(model_name)

Performance

This work proposes a system to resolve the challenge sub-tasks related to human values detection. Our approach uses cascade DeBERTa models, where the first detects the presence of each human value, and the second detects if the sentence attains or constrains the present human values in each sentence. The latter approach improves the effectiveness of the baseline at the test dataset by 4 on sub-task 1 and by 1 on sub-task 2. These models were trained on a subset of 44,758 sentences in English, validated on a subset of 14,904 sentences, and tested on a separate subset of 14,569 sentences.

This model has got the third place in the subtask 1 of Human Value Detection at CLEF 2024.

Limitations and bias

At the time of submission, no measures have been taken to estimate the bias embedded in the model, so it may not be safe for use in production.

License

The model is released under open license CC BY 4.0, available at https://creativecommons.org/licenses/by/4.0/legalcode.

BibTeX entry and citation information

@inproceedings{yeste2024philo,
  title={Philo of Alexandria at touch{\'e}: a cascade model approach to human value detection},
  author={Yeste, V{\'\i}ctor and Coll-Ardanuy, M and Rosso, Paolo},
  booktitle={Working Notes of the Conference and Labs of the Evaluation Forum (CLEF 2024). CEUR Workshop Proceedings, CEUR-WS. org},
  year={2024}
}