Summarization is a central activity in the field of Natural Language Processing that makes it possible to extract essential information from a massive amount of text with minimal preservation of key meaning. But in the case of the Uzbek language, summarization is a weak field because it is an agglutinative language with no high-quality annotated corpora. The traditional approach towards extractive summarization, i.e., Term Frequency-Inverse Document Frequency (TF-IDF), is not effective in capturing linguistic complexities in Uzbek, leading to weak coherence as well as relevance in a summary. Here, we propose a new approach towards summarization that is based on TF-IDF with word importance probability estimation, calculation of variance, as well as detection of skewness in order to dynamically decide and choose the key portions of text. Compared with methods that are traditionally based on frequency alone, our approach estimates mathematical expectation of words’ importance, measures words' distribution variance, as well as employs symmetry-based examination in order to determine whether key information is at the starting, in between, or towards ending portion of the text. Experimental experiments have confirmed that the proposed approach is more contextually sound and pertinent in its summaries as compared with baseline models of TF-IDF. The research is a contribution towards enhancing low-resource NLP in Uzbek as well as towards a scalable statistical approach that can be adopted in highly inflected languages.