R&D systems have a large number of types of documents required to document and control activities in this area. A number of problems have been identified that need to be solved in terms of collecting and consolidating data on R&D and creating reporting documentation based on it. First, the comparison of data from different sources. Second, changes in reporting formats and uncertainty about the data that should be available. Third, the issue of a flexible mechanism for storing data in multiple languages. The goal of the study is to improve the accuracy of comparing and grouping bibliographic records that occur when creating reporting documents. Experiments were conducted on entity recognition to process textual descriptions and match them with database objects. The results of the experiments with two approaches showed that using the word2vec model to compare the similarity between word combinations and existing database entities gives a value of 0.38 for the F1 metric. However, the second approach, based on the Levenshtein metric, reached a value of 0.9 for the F1 metric, so its use is advisable. The result of the research is a set of methods, specifications, and software components that have proven to be suitable for the tasks of comparing bibliographic records and generating reports.