The paper outlines a series of experiments aiming to test whether the vector space text analysis model could be successfully and reliably used to automatically determine the level of competence of a reviewer to evaluate a specific paper. The model corresponds to an implicit method of describing papers and reviewers' competences that does not require users to perform any additional actions. The similarity factors between papers and reviewers are calculated based on content analysis of publications' abstracts. The document collection consists of all submitted papers and reviewers' previous publications. The latter are retrieved from external Internet resources -- Semantic Scholar and DBLP. The analysis is performed on real datasets obtained from CompSysTech series of conferences for a period of 5 years (2014--2018). During that period, CompSysTech relied on both an explicit method of describing papers and reviewers' competences (selecting keywords from a predefined topic taxonomy) and a real human self-assessment (evaluating own level of expertise). Sharing the same datasets allows a fair comparison between the different methods of calculating paper-reviewer similarities, and an objective evaluation of their accuracy in respect to the real human judgement. Experimental results clearly reveal that the vector space model could be successfully used for automatic assignment of reviewers to papers.

This article focuses on the importance of the precise calculation of similarity factors between papers and reviewers for performing a fair and accurate automatic assignment of reviewers to papers. It suggests that papers and reviewers? competences should be described by taxonomy of keywords so that the implied hierarchical structure allows similarity measures to take into account not only the number of exactly matching keywords, but in case of non-matching ones to calculate how semantically close they are. The paper also suggests a similarity measure derived from the well-known and widely-used Dice's coefficient, but adapted in a way it could be also applied between sets whose elements are semantically related to each other (as concepts in taxonomy are). It allows a non-zero similarity factor to be accurately calculated between a paper and a reviewer even if they do not share any keyword in common.

Content-based image retrieval methods in present days are used in modern social media and search engines. They give techniques for analyzing, organizing, processing and searching through million images uploaded daily in the internet. For that reason, searching is the most critical process in CBIR. It should be accurate and complete in reasonable amount of time. Proper metadata should be extracted from the images and used to meet the performance requirements. From the other hand, to meet the subsequent processing this metadata should be indexed and stored in appropriate way.This paper describes some of the most popular image extraction and analysis systems known as Content Based Image Retrieval Systems (CBIR). It reveals how examined different CBIR systems work and how closer are generated similarity results. At the end, information is summarized, and conclusions are made regarding existing solutions and methods used in these applications.