Network processing in the current Internet is at the entirety of the data packet, which is problematic when encountering network congestion. The newly proposed Internet service named Qualitative Communication changes the network processing paradigm to an even finer granularity, namely chunk level, which obsoletes many existing networking policies and schemes, especially the caching algorithms and cache replacement policies that have been extensively explored in Web Caching, Content Delivery Networks (CDN) or Information-Centric Networks (ICN). This paper outlines all the new factors that are brought by random linear network coding-based Qualitative Communication and proves the importance and necessity of considering them. A novel metric is proposed by taking these new factors into consideration. An optimization problem is formulated to maximize the metric value of all retained chunks in the local storage of network nodes under the constraint of storage limit. A cache replacement scheme that obtains the optimal result in a recursive manner is proposed correspondingly. With the help of the introduced intelligent cache replacement algorithm, the performance evaluations show remarkably reduced end-to-end latency compared to the existing schemes in various network scenarios.