Predicting the Information Need for Domestic Violence Survivors Based on the Fine-Tuned Large Language Model.

ABSTRACT

Women with domestic violence experiences often refuse to seek help face-to-face due to embarrassment. They begin to share their emotions and seek help from online health communities. Understanding and responding to these posts can be crucial in providing timely support to the victims. We proposed a fine-tuned large language model (LLM) capable of accurately predicting the informational need based on the content of postings. We fine-tuned the LAMMA2-7B-chat model based on the guidance of identifying the information need and a dataset comprising 273 posts from Reddit, which are manually annotated by domain experts. Furthermore, we evaluated the performance of our model using a random sample of 15 posts, and 66.6% were accurately predicted. The results demonstrate that our model can rapidly capture the information needs expressed in the posts, enabling healthcare providers to provide timely and useful support based on our predictions.