In this age of digital publishing and exciting advances in AI and machine learning technology, it is important to keep in mind that your research is being read not only by human readers but also by machines.
Publishers in the STEM fields are using machine learning technology to develop innovative tools and features on their platforms to meet the needs of academic and industry researchers, practitioners, and students. For instance, just as Amazon and Netflix make recommendations to users based on their buying choices and search history, publishers are using algorithms to mine the content on their platforms (such as ScienceDirect and SpringerLink) to make customized recommendations based on the individual user’s activity. These features are designed to help researchers stay up to date with content relevant to their current research problems. Publishers are also developing uses for machine learning in the areas of manuscript peer review, plagiarism detection, scholarly social networking and career development, and information retrieval to answer specific research questions.
Search algorithms are becoming more and more sophisticated, designed to surface laser-focused, relevant results for both text and image queries. You can assist publishers in preparing your article for search engine optimization (SEO) to ensure that your article content is found and ranked highly by search engines such as Google and Google Scholar.
Tips for increasing discoverability: Below are tips for optimizing four essential structural components of your article. A key strategy for ensuring that the machines reading your article will pick up specific, relevant content targeted toward your intended audience is to treat these components as separate items that are able to function independently from each other and the main text.
1. Title. Give your article a concise, specific, no-frills title that succinctly conveys the article’s main subject. Include keywords in your title.
2. Abstract. Limit your abstract to 250 words or less. (Be sure to consult your target journal’s submission guideline for specific requirements.) Include keywords in the abstract, and make sure to include them in the first few sentences, as these are the sentences that are displayed in search engine results.
3. Keywords. Carefully choose 4–8 keywords that are specific to your article content and to your research field and subfield and list them following the abstract. (Again, consult your target journal for specific requirements.)
- Include keywords in the title but do not include the same words in your keyword list. The words in your title are already tagged and treated as key. Keywords should supplement the words in your title.
- Keywords in your list can be one word or groups of words, such as “high-density lipoprotein” and “unbiased nonlocal means filtering.”
- Include keywords in the headings and subheadings of your article.
4. Figure and table legends. Figures and tables should be able to be understood without the help of the body of the manuscript. Many readers will peruse the figures and tables to quickly see results before deciding to read further. Follow the three tips below to ensure your figures and tables are fully discoverable and usable by both human and machine readers.
- a. Provide a legend that includes keywords for each of your figures and tables. Do not embed the legend in the figure itself.
- b. The first sentence in your figure legend should concisely describe what it is being depicted and should identify the type of image shown, for example, bar graph, flow chart, scatterplot, sonogram, spectrogram, chromatogram, or map. This is especially important for the benefit of sight-impaired readers as well as machine readers.
Consider the examples below, which are created for this article but modeled after the first sentences of legends from actual manuscripts.
Poor: - Fig. 1. Study area and locations of the observation sites.
This figure would not be discoverable in targeted searches because the language in the legend is too general.
Good: - Fig. 1. A topographical map showing the research area and four observation stations for multifractal detrended fluctuation analysis (MFDFA) along the Yellow River.
Poor: - Fig. 1. Increased liver stiffness in a 21-year-old man.
Good: - Fig. 1. Magnetic resonance elastography images showing increased liver stiffness in a 21-year-old man who underwent atriopulmonary connection-Fontan palliation at age 4.
- c. Define all non-standard acronyms, initialisms, and symbols at first use in each figure or table and in each figure or table legend.
Following the simple strategies outlined above will increase your chances of having your article accepted for publication by your target journal and help maximize your article’s discoverability once published. As a thank you, your machine readers will make sure to share your research content with the appropriate humans.
Further Reading and Works Consulted:
- Evan, Ian. Harvard and beyond: 4 ways machine learning is making researchers more efficient. Elsevier Connect. April 18, 2018. https://www.elsevier.com/connect/harvard-and-beyond-4-ways-machine-learning-is-making-researchers-more-efficient. Accessed Jan. 21, 2019.
- Rowe, Kevin. How Search Engines Use Machine Learning: 9 Things We Know for Sure. Search Engine Journal. Feb. 23, 2018. https://www.searchenginejournal.com/how-search-engines-use-machine-learning/224451/. Accessed Jan. 21, 2019.
- Search Engine Optimization (SEO) for your article. Author Resources. Wiley. https://authorservices.wiley.com/author-resources/Journal-Authors/Prepare/writing-for-seo.html. Accessed Jan. 21, 2019.