The accurate forecasting of exceptional growth in research areas has been an extremely difficult problem to solve. In a previous study we introduced an approach to forecasting which research clusters in a global model of the scientific literature would have an annual growth rate of 8% annually over a 3-year period. In this study we (a) introduce a much more robust method of creating and updating global models of research, (b) introduce new indicators based on author publication patterns, (c) test a much larger set (81) of indicators to forecast exceptional growth, and (d) expand the forecast horizon from 3 to 4 years. Forecast accuracy increased dramatically (threat score increased from 20 to 32) from our previous study. Most of this gain is surprisingly due to the advances in model robustness rather than the indicators used for forecasting. We also provide evidence that most indicators (including popular network indicators) do not improve the ability to forecast growth in research above the baseline provided by indicators associated with the vitality of a research cluster.

Recent large-scale bibliometric models have largely been based on direct citation, and several recent studies have explored augmenting direct citation with other citation-based or textual characteristics. In this study we compare clustering results from direct citation, extended direct citation, a textual relatedness measure, and several citation-text hybrid measures using a set of nine million documents. Three different accuracy measures are employed, one based on references in authoritative documents, one using textual relatedness, and the last using document pairs linked by grants. We find that a hybrid relatedness measure based equally on direct citation and PubMed-related article scores gives more accurate clusters (in the aggregate) than the other relatedness measures tested. We also show that the differences in cluster contents between the different models are even larger than the differences in accuracy, suggesting that the textual and citation logics are complementary. Finally, we show that for the hybrid measure based on direct citation and related article scores, the larger clusters are more oriented toward textual relatedness, while the smaller clusters are more oriented toward citation-based relatedness.

There are many different relatedness measures, based for instance on citation relations or textual similarity, that can be used to cluster scientific publications. We propose a principled methodology for evaluating the accuracy of clustering solutions obtained using these relatedness measures. We formally show that the proposed methodology has an important consistency property. The empirical analyses that we present are based on publications in the fields of cell biology, condensed matter physics, and economics. Using the BM25 text-based relatedness measure as the evaluation criterion, we find that bibliographic coupling relations yield more accurate clustering solutions than direct citation relations and cocitation relations. The so-called extended direct citation approach performs similarly to or slightly better than bibliographic coupling in terms of the accuracy of the resulting clustering solutions. The other way around, using a citation-based relatedness measure as evaluation criterion, BM25 turns out to yield more accurate clustering solutions than other text-based relatedness measures.