Step 2: Metadata retrievalΒΆ

The step of metadata retrieval refers to the process of formulating and executing a search strategy, which involves operations related to obtaining search results from different sources, as well as loading, preparing, and deduplicating them. This step can be technically demanding due to limited functionality of data sources and data quality issues, and methodologically demanding because there are no simple recipes to formulate an effective search strategy.

Standalone literature reviews are typically based on a search strategy, which comprises the following elements:

  • The search scope can restrict the journals, disciplines (based on reasons related to their topics or source reputation), time, or publication type (grey vs. academic literature).

  • The search type and iterations can be exploratory or systematic, and may evolve throughout the search iterations.

  • The combination of search techniques, sources, and parameters. Search techniques can be database searches, citation searches (backward or forward snowballing), table-of-content searches, for example. Examples of sources are available here. For databases, the parameters (or search strings) can be developed based on a concept matrix (e.g., in line with the PICO framework), boolean operators, and search query translation tools.

  • The termination rule specifies the criteria for stopping the search process.

The methodological choices related to the search strategy should be documented for reporting purposes. This can be done in the data/paper.md or the readme.md.

Metadata retrieval is a high-level operation consisting of the following operations:

  • The search operation, which obtains search results automatically (using an API) or from a file provided by the user, and records information related to the search source (search parameters).

  • The load operation, which converts all search results (metadata) to the same format (BibTex) and add to the same dataset (data/record.bib file). Each record is linked to the origins that remain in the data/search directory.

  • The prep operation, which evaluates and improves the data quality of records. This is necessary to ensure adequate performance of deduplication algorithms and to reduce manual efforts for polishing reference sections. If manual preparation is required, records are set to the md_needs_manual_preparation state and the colrev status advises users on how to proceed.

  • The dedupe operation, which links records pointing to the same paper to create a dataset without duplicates. Specifically, duplicate records are merged in a way that preserves their links to the respective origin records and allows users to undo merges at any time (no simple removing of duplicates).

To start the retrieval operation, add search results to the data/search directory and run:

colrev retrieve

The design of the retrieval operations ensures that searches can be conducted in a highly iterative way. In particular, this pertains to an efficient flow of new records through the process and mechanisms to handle changes (e.g., when papers are retracted).

Operations