Assessing trustworthiness of social media posts is increasingly important, as the number of online users and activities grows. Current deploying assessment systems measure post trustworthiness as credibility. However, they measure the credibility of all posts, indiscriminately. The credibility concept was intended for news types of posts. Labeling other types of posts with credibility scores may confuse the users. Previous notable works envisioned filtering out non-newsworthy posts before credibility assessment as a key factor towards a more efficient credibility system. Thus, we propose to implement a topic-based supervised learning approach that uses Term Frequency-Interim Document Frequency (TF-IDF) and cosine similarity for filtering out the posts that do not need credibility assessment. Our experimental results show that about 70% of the proposed filtering suggestions are agreed by the users. Such results support the notion of newsworthiness, introduced in the pioneering work of credibility assessment. The topic-based supervised learning approach is shown to provide a viable social network filter.