Tables in documents serve as valuable sources of information, and there has been a growing interest in research on information extraction from them. This paper delves into the methods for information extraction from text data we refer to as catalogs, bearing resemblance to tables. A catalog is a collection of text data where data of specific types (categories) are arranged roughly in a predefined order, such as a product list composed of product name, manufacturer, and production date. By tokenizing the text data, catalogs can be transformed into tables. However, catalogs are not always created with explicit table-like structures in mind, which may lead to potential discrepancies in data alignment in the resulting tables. Moreover, catalogs lack information like table headers, which have been commonly used for interpreting tables. In this paper, we propose a method to extract data of specific type from datasets comprising multiple catalogs. Distinctively different from conventional table-based information extraction methods, our method does not rely on external knowledge or labeled data, but exclusively on the estimated distribution of types of data within the dataset.  We have validated the efficacy of our approach using synthetically generated data and product data from e-commerce sites.