WRAPresentations

Introducing WRAPresentations, a cutting-edge sentence-transformers model that leverages the power of a 768-dimensional dense vector space to map tweets according to the four classes Reason, Statement, Notification and None. This powerful model is tailored for argument mining on Twitter, derived from the BERTweet-base architecture initially pre-trained on Twitter data. Through fine-tuning with the TACO dataset, WRAPresentations is effectively in Weaving Relevant Argument Properties (WRAP) into the embedding space.

Class Semantics

The TACO framework revolves around the two key elements of an argument, as defined by the Cambridge Dictionary. It encodes inference as a guess that you make or an opinion that you form based on the information that you have, and it also leverages the definition of information as facts or details about a person, company, product, etc..

WRAPresentations, to some degree, captures the semantics of these critical components in its embedding space.

Consequently, it has also learned the class semantics, where inferences and information can be aggregated in relation to these distinct classes containing these components:

• Statement, which refers to unique cases where only the inference is presented as something that someone says or writes officially, or an action done to express an opinion.
• Reason, which represents a full argument where the inference is based on direct information mentioned in the tweet, such as a source-reference or quotation, and thus reveals the author’s motivation to try to understand and to make judgments based on practical facts.
• Notification, which refers to a tweet that limits itself to providing information, such as media channels promoting their latest articles.
• None, a tweet that provides neither inference nor information.

In its entirety, WRAPresentations encodes the following hierarchy for tweets:

<div align="center"> <img src="https://github.com/TomatenMarc/public-images/raw/main/Component_Space_WRAP.svg" alt="Component Space" width="100%"> </div>

Class Semantic Transfer to Embeddings

Observing the tweet distribution within the embedding space of WRAPresentations, we noted that extended pre-training via contrastive learning led to denser emergence of the expected class sectors compared to the embeddings of BERTweet, as shown in the following figure. <div align="center"> <img src="https://github.com/TomatenMarc/public-images/raw/main/sector_purity_coordinates.svg" alt="Argument Tree" width="100%"> </div>

Usage (Sentence-Transformers)

Using this model becomes easy when you have sentence-transformers installed:

pip install -U sentence-transformers

Then you can use the model to generate tweet representations like this:

from sentence_transformers import SentenceTransformer

tweets = ["This is an example #tweet", "Each tweet is converted"]

model = SentenceTransformer("TomatenMarc/WRAPresentations")
embeddings = model.encode(tweets)
print(embeddings)

<a href="https://github.com/VinAIResearch/BERTweet/blob/master/TweetNormalizer.py"> <blockquote style="border-left: 5px solid grey; background-color: #f0f5ff; padding: 10px;"> Notice: The tweets need to undergo preprocessing following the specifications for BERTweet-base. </blockquote> </a>

Usage (HuggingFace Transformers)

Without sentence-transformers, you can use the model like this: First, you pass your input through the transformer model, then you have to apply the right pooling-operation on top of the contextualized word embeddings.

from transformers import AutoTokenizer, AutoModel
import torch


# Mean Pooling - Take attention mask into account for correct averaging
def mean_pooling(model_output, attention_mask):
    token_embeddings = model_output[0]  # First element of model_output contains all token embeddings
    input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)


# Tweets we want embeddings for
tweets = ["This is an example #tweet", "Each tweet is converted"]

# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained("TomatenMarc/WRAPresentations")
model = AutoModel.from_pretrained("TomatenMarc/WRAPresentations")

# Tokenize sentences
encoded_input = tokenizer(tweets, padding=True, truncation=True, return_tensors="pt")

# Compute token embeddings
with torch.no_grad():
    model_output = model(**encoded_input)

# Perform pooling. In this case, mean pooling.
sentence_embeddings = mean_pooling(model_output, encoded_input["attention_mask"])

print("Sentence embeddings:")
print(sentence_embeddings)

Furthermore, the WRAPresentations model is a highly suitable embedding component for AutoModelForSequenceClassification, enabling fine-tuning of tweet classification tasks specifically for the four classes: Reason, Statement, Notification, and None. The categorization of Reason and Statement as argument classes and Notification and None as non-argument classes is implicitly learned during the fine-tuning process. This setup facilitates efficient identification and analysis of argumentative content and non-argumentative content in tweets.

Training

The WRAPresentations model underwent fine-tuning with 1,219 golden tweets from the TACO dataset, covering six topics. Five topics were chosen for optimization, representing 925 tweets (75.88%) covering #brexit (33.3%), #got (17%), #lotrrop (18.8%), #squidgame (17.1%), and #twittertakeover (13.8%). The model used a stratified 60/40 split for training/testing on optimization data. Additionally, 294 golden tweets (24.12%) related to the topic of #abortion were chosen as the holdout-set for final evaluation.

Before fine-tuning, we built a copy of the dataset by creating an augmentation of each tweet. The augmentation consisted of replacing all the topic words and entities in a tweet replaced, and then randomly masking 10% of the words in a tweet, which were then matched using BERTweet-base as a fill-mask model. We chose to omit 10% of the words because this resulted in the smallest possible average cosine distance between the tweets and their augmentations of 0.08, which is close to total dissimilarity, making augmentation during extended pre-training itself a regulating factor prior to any overfitting with the later test data. During fine-tuning, we formed pairs by matching each tweet with all remaining tweets in the same data split (training, testing, holdout) with similar or dissimilar class labels. For the training and testing set during the fine-tuning process, we utilized the augmentations, and for the holdout tweets, we used their original text to test the fine-tuning process and the usefulness of the augmentations towards real tweets. For all pairs, we chose the largest possible set so that both similar and dissimilar pairs are equally represented while covering all tweets of the respective data split. This process created 307,470 pairs for training and 136,530 pairs for testing. An additional 86,142 pairs were used for final evaluation with the holdout data.

The model was trained with the parameters:

DataLoader:

torch.utils.data.dataloader.DataLoader of length 5065 with parameters:

{'batch_size': 32, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}

Loss:

sentence_transformers.losses.ContrastiveLoss.ContrastiveLoss with parameters:

{'distance_metric': 'SiameseDistanceMetric.COSINE_DISTANCE', 'margin': 0.5, 'size_average': True}

Parameters of the fit()-Method:

{
    "epochs": 5,
    "evaluation_steps": 1000,
    "evaluator": "sentence_transformers.evaluation.BinaryClassificationEvaluator.BinaryClassificationEvaluator",
    "max_grad_norm": 1,
    "optimizer_class": "<class 'torch.optim.adamw.AdamW'>",
    "optimizer_params": {
        "lr": 4e-05
    },
    "scheduler": "WarmupLinear",
    "steps_per_epoch": null,
    "warmup_steps": 2533,
    "weight_decay": 0.01
}

Evaluation Results

Following the standard protocol for cross-topic evaluation for argument mining, we evaluated the WRAPresentation model using the BinaryClassificationEvaluator showing:

An evaluation was conducted on previously unseen data from the holdout topic #abortion, resulting in the model achieving a sophisticated macro-F1 score of 74.08%. The recall, which stands at 84.05%, indicates the model's ability to capture subtle tweet patterns and class-specific features for Reason, Statement, Notification, and None. Despite having a lower precision of 66.22%, the model's primary focus is on prioritizing recall to capture relevant instances. Fine-tuning precision can be addressed in a subsequent classification phase, when using this model for AutoModelForSequenceClassification. On the one hand, the baseline model (vinai/bertweet-base) exhibited an outstanding recall of 99.89%. However, this high recall came at the cost of precision, which was relatively low at 50.08%. This precision-recall trade-off suggests a potential issue with overfitting. On the other hand, the extended baseline (vinai/bertweet-base with extended pre-training on augmented training data) showed improved precision at 66.51% and a commendable recall of 82.55%, resulting in an F1 score of 73.67%. Comparatively, for WRAPresentations, the precision was slightly lower than that of the extended baseline but excelled in terms of recall, showing a 2.5% increase. This improvement could be attributed to the benefits of contrastive learning, enhancing the model's ability to generalize, which is evident in the form of enhanced recall performance. However, WRAPresentations demonstrated its ability to effectively distinguish between tweets of the argument framework, capturing intra-class semantics while discerning inter-class semantics. This is indicated by its better F1 score of 74.08%, showcasing a superior balance between recall and precision. As a result, WRAPresentations proves to be more suitable for argument mining on Twitter, as it achieves a more reliable performance in identifying relevant instances in the data.

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 128, 'do_lower_case': False}) with Transformer model: RobertaModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False})
)

Environmental Impact

• Hardware Type: A100 PCIe 40GB
• Hours used: 2h
• Cloud Provider: Google Cloud Platform
• Compute Region: asia-southeast1 (Singapore)
• Carbon Emitted: 0.21kg CO2

Licensing

WRAPresentations © 2023 is licensed under CC BY-NC-SA 4.0.