Salvato in:
| Autore principale: | |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2504.16856 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866916704006701056 |
|---|---|
| author | Shvets, Alexander |
| author_facet | Shvets, Alexander |
| contents | Most datasets for sentiment analysis lack context in which an opinion was expressed, often crucial for emotion understanding, and are mainly limited by a few emotion categories. Foundation large language models (LLMs) like GPT-4 suffer from over-predicting emotions and are too resource-intensive. We design an LLM-based data synthesis pipeline and leverage a large model, Mistral-7b, for the generation of training examples for more accessible, lightweight BERT-type encoder models. We focus on enlarging the semantic diversity of examples and propose grounding the generation into a corpus of narratives to produce non-repetitive story-character-centered utterances with unique contexts over 28 emotion classes. By running 700K inferences in 450 GPU hours, we contribute with the dataset of 100K contextual and also 300K context-less examples to cover both scenarios. We use it for fine-tuning pre-trained encoders, which results in several Emo Pillars models. We show that Emo Pillars models are highly adaptive to new domains when tuned to specific tasks such as GoEmotions, ISEAR, IEMOCAP, and EmoContext, reaching the SOTA performance on the first three. We also validate our dataset, conducting statistical analysis and human evaluation, and confirm the success of our measures in utterance diversification (although less for the neutral class) and context personalization, while pointing out the need for improved handling of out-of-taxonomy labels within the pipeline. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_16856 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Emo Pillars: Knowledge Distillation to Support Fine-Grained Context-Aware and Context-Less Emotion Classification Shvets, Alexander Computation and Language Most datasets for sentiment analysis lack context in which an opinion was expressed, often crucial for emotion understanding, and are mainly limited by a few emotion categories. Foundation large language models (LLMs) like GPT-4 suffer from over-predicting emotions and are too resource-intensive. We design an LLM-based data synthesis pipeline and leverage a large model, Mistral-7b, for the generation of training examples for more accessible, lightweight BERT-type encoder models. We focus on enlarging the semantic diversity of examples and propose grounding the generation into a corpus of narratives to produce non-repetitive story-character-centered utterances with unique contexts over 28 emotion classes. By running 700K inferences in 450 GPU hours, we contribute with the dataset of 100K contextual and also 300K context-less examples to cover both scenarios. We use it for fine-tuning pre-trained encoders, which results in several Emo Pillars models. We show that Emo Pillars models are highly adaptive to new domains when tuned to specific tasks such as GoEmotions, ISEAR, IEMOCAP, and EmoContext, reaching the SOTA performance on the first three. We also validate our dataset, conducting statistical analysis and human evaluation, and confirm the success of our measures in utterance diversification (although less for the neutral class) and context personalization, while pointing out the need for improved handling of out-of-taxonomy labels within the pipeline. |
| title | Emo Pillars: Knowledge Distillation to Support Fine-Grained Context-Aware and Context-Less Emotion Classification |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2504.16856 |