推文中的宣稱檢測

Abstract

隨著網際網路技術的進步，假信息的議題日益嚴重。在眾多處理假信息的方法中，事實查核被視為最值得信任的手段。然而，傳統的事實查核通過人力驗證信息真偽的過程可能耗時過長，無法趕上信息產生的速度，因此事實查核的自動化成為了一個極具價值的目標。自動化事實查核通常分為四個階段，包括宣稱檢測、已驗證的宣稱檢索、證據檢索、與宣稱驗證。

本研究著重於解決推文中的宣稱檢測問題。宣稱檢測的目標是確定哪部分內容是宣稱，而一個宣稱可被定義為「宣稱某事是事實」，驗證這些宣稱的真偽即為事實查核的過程。在這項研究中，我們首先提出了過去在推文中進行宣稱檢測時兩個未被充分處理的問題：推文的混亂格式與任務定義，並提出了相應的方法來解決這些問題。此外，我們也透過實驗去驗證我提出的方法的有效性，並總結了實驗中的啟示與發現。

With the advancement of Internet technology, the issue of misinformation is increasingly serious. Among the many methods of dealing with misinformation, fact-checking is considered the most trustworthy means. However, the traditional fact-checking process, which verifies the authenticity of information manually, may be too time-consuming to keep up with the speed of information generation. Thus, the automation of fact-checking has become a highly valuable goal. Automated fact-checking is generally divided into four stages, including claim detection, verified claim retrieval, evidence retrieval, and claim verification.

This study focuses on solving claim detection in tweets. The goal of claim detection is to determine which part of the content is a claim, and a claim can be defined as "an assertion of something as a fact". Verifying the truth of these claims is the process of fact-checking. In this research, we first propose two issues that have not been adequately addressed in claim detection within tweets : tweet's noisy format and task definition, and propose corresponding methods to solve these problems. In addition, we also validate the effectiveness of our proposed methods through experiments and summarize the insights and findings from the experiments.