New framework for automated article selection applied to a literature review of Enhanced Biological Phosphorus Removal

Abstract

Aims: Enhanced Biological Phosphorus Removal (EBPR) is a technology widely used in wastewater treatment to remove phosphorus (P) and prevent eutrophication. Establishing its operating efficiency and stability is an active research field that has generated almost 3000 publications in the last 40 years. Due to its size, including over 119 review articles, it is an example of a field where it becomes increasingly difficult to manually recognize its key research contributions, especially for non-experts or newcomers. Therefore, this work included two distinct but complementary objectives. First, to assemble for the first time a collection of bibliometric techniques into a framework for automating the article selection process when preparing a literature review (section 2). Second, to demonstrate it by applying it to the field of EBPR, producing a bibliometric analysis and a review of the key findings of EBPR research over time (section 3).

Findings: The joint analysis of citation networks, keywords, citation profiles, as well as of specific benchmarks for the identification of highly-cited publications revealed 12 research topics. Their content and evolution could be manually reviewed using a selection of articles consisting of approximately only 5% of the original set of publications. The largest topics addressed the identification of relevant microorganisms, the characterization of their metabolism, including denitrification and the competition between them (Clusters A-D). Emerging and influential topics, as determined by different citation indicators and temporal analysis, were related to volatile fatty acid production, P-recovery from waste activated sludge and aerobic granules for better process efficiency and stability (Clusters F-H).

Conclusions: The framework enabled key contributions in each of the constituent topics to be highlighted in a way that may have otherwise been biased by conventional citation-based ranking. Further, it reduced the need for manual input and a priori expertise compared to a traditional literature review. Hence, in an era of accelerated production of information and publications, this work contributed to the way that we are able to use computer-aided approaches to curate information and manage knowledge.

Fig 1. Difference between bibliographic coupling and co-citation.

Bibliographic coupling links papers P₁ and P₂ that cite a common reference P₃. Co-citation links papers P₄ and P₅ that both appear in the reference list of P₃.

Fig 2. Logic tree for the citation-profile-based classification of scientific publications.

Fig 3. Examples of delayed-peak (top row) and multi-peak (bottom row) citation profiles.

Fig 4. Composition of each citation quartile in terms of the six classes of citation profiles, where quartiles are determined based on the frequency of total citation counts.

The arrow indicates the direction of increasing number of citations.

Fig 5. Composition of citation profiles and fraction of total publications across five evenly-spaced periods of time between 1975 and 2010.

The composition of citation profiles (bars) was normalized against the number of publications in the corresponding time period, whereas the publication volume (line) was normalized with respect to the total number of publications.

Fig 6. Profile of the mean and median number of citations received by papers published in a given year.

Bars indicate the number of papers published in a given year.

Fig 7. Chord diagram of the flow of citations between the identified clusters, excluding intra-cluster citations.

Outer-most arch: sum of all links to and from a given cluster. Second outer-most arch: inward links, representing citations received from publications in other clusters. Third outer-most arch: outward links, representing citations given to publications in other clusters.

Fig 8. Boxplot of the publication year distribution for each cluster, indicating the median (band), interquartile range (box), as well as the minimum and maximum (whiskers) values.

The inter-quartile range indicates the time period during which 50% of papers in a cluster were published.