房屋租賃文本與法律條文相關性辨識之研究

The Relevance Identification Between Housing Rental Texts And Legal Provisions

近年來臺灣的房價市場由於各類因素而隨之水漲船高，因此許多人便選擇以租屋代替買房。而許多租屋族在簽訂房屋租賃合約時對於相關法規和細節的陌生，加上未詳閱或疏漏繁雜之合約內容的總總因素，都會進而引發糾紛。本研究將試圖開發一套線上房屋租賃辨識系統，讓房客能夠透過此系統辨識租賃契約是否有問題，並提示未來租屋行為、租屋過程若發生糾紛可用之法律條文，幫助租屋族減少對於法條與契約約定事項的不理解而產生的租賃糾紛。在本研究中，使用了五種模型並選取四種影響因子來進行模型訓練與評估，其中以distilbert-base-multilingual-cased模型並在未考慮任何影響因子下表現最佳macro f1分數為0.14、weighted f1分數為0.21，但各模型在本研究資料集之訓練效果普遍不佳，由研究結果顯示，租賃問題之相關法律條文辨識仍需要建立更大量且多元的語料庫。

In recent years, the housing market in Taiwan has seen soaring prices due to various factors, leading many people to choose renting over buying. However, many renters are unfamiliar with the relevant regulations and details when signing rental agreements. Combined with a lack of thorough review or overlooking complex contract terms, these factors often lead to disputes. This study aims to develop an online housing lease identification system that allows tenants to assess whether their lease contracts have issues. It will also provide information on legal provisions that can be used in case of disputes during the rental process, helping renters reduce misunderstandings about legal regulations and contract terms. In this research, five models were used, and four influencing factors were selected for model training and evaluation. Among them, the distilbert-base-multilingual-cased model performed the best with a macro F1 score of 0.14 and a weighted F1 score of 0.21, without considering any influencing factors. However, the overall performance of all models on this research dataset was generally poor. The results suggest that the identification of relevant legal provisions for lease issues still requires the establishment of a larger and more diverse corpus of text data.