Table of Links
2 Related Work and 2.1 Technology Convergence Approaches
2.2 Technology Convergence Measurements
2.3 Technology Convergence Models
4 Method and 4.1 Proximity Indices
4.2 Interpolation and Fitting Data
5 Results and Discussion and 5.1 Overall Results
5.3 Limitations and Future Works
5 Results and Discussion
This section delves into the details of the results and their insights, followed by a case study illustrating the effectiveness of our proposed method in two cybersecurity topics. Finally, we discuss the limitations and future work.
5.1 Overall Results
The following tables provide detailed forecasting accuracy for each method, presenting data specific to index types, algorithms, and forecasting horizons. In these tables, Index 1 and Index 2 represent citations from one technology to another and vice versa. Index 3 and 4 denote collaboration based on incremental and non-incremental h-indices, respectively, while Index 5 pertains to common keywords. The colors brown, violet, and blue indicate the best forecasting accuracy for horizons of 3, 6, and 12 months, respectively.
Overall, no single forecasting method consistently outperformed others across all tasks. For local forecasting, exponential smoothing and Theta algorithms demonstrated superior performance, while random forest outperformed linear regression and light gradient boosting machine (LGBM) in other forecasting types. In contrast, transfer learning methods exhibited comparatively poorer performance, potentially due to their computationally intensive nature and associated training limitations.
Different index types exhibited diverse forecasting outcomes. Citation indices, often flat, were straightforward to predict, resulting in a median SMAPEs of 0 across all forecasting horizons (see Table 4). Conversely, collaboration indices posed greater challenges, with the most accurate predictions stemming from randomized clustering forecasting using random forests (refer to Table 2). Keyword indices saw reasonable forecasting, with algorithmic clustering via random forests providing the most accurate results (Table 3).
This paper is available on arxiv under CC BY 4.0 DEED license.
Authors:
(1) Alessandro Tavazz, Cyber-Defence Campus, armasuisse Science and Technology, Building I, EPFL Innovation Park, 1015, Lausanne, Switzerland, Institute of Mathematics, EPFL, 1015, Lausanne, Switzerland and a Corresponding author (tavazale@gmail.com);
(2) Dimitri Percia David, Cyber-Defence Campus, armasuisse Science and Technology, Building I, EPFL Innovation Park, 1015, Lausanne, Switzerland and Institute of Entrepreneurship & Management, University of Applied Sciences of Western Switzerland (HES-SO Valais-Wallis), Techno-Pole 1, Le Foyer, 3960, Sierre, Switzerland;
(3) Julian Jang-Jaccard, Cyber-Defence Campus, armasuisse Science and Technology, Building I, EPFL Innovation Park, 1015, Lausanne, Switzerland;
(4) Alain Mermoud, Cyber-Defence Campus, armasuisse Science and Technology, Building I, EPFL Innovation Park, 1015, Lausanne, Switzerland.