2. Background

The interconnection among CTI, TF and TRM activities is identified by means of the abundance of reference literature. In the 90s, Porter et al. proposed a method, called technology opportunities analysis (TOA), which used ST&I data and bibliometrics with the purpose of identifying and assessing the implications of emerging scientific areas and new research technologies [7]. Following this path, Lee and Jeong used bibliometric data, co-word analysis, to generate a strategic diagram to be used for the analysis of the development trends of a specific technology domain [8]. Similarly, Lee et al. proposed a new TRM methodology to increase roadmapping effectiveness to support effective decision-making in new product and technology planning processes. The data source was patents and the method was founded on keyword-based product–technology maps, from which objective and quantitative information can be derived [9].

Latest efforts in this field are focused on the integration of more complex statistical methods and (semi)automatization proposals. In this regard, works can be found such as that proposed by Zhang et al. [10], in which a TRM composing method is described where data inputs are raw science textual data sources. The method seeks to identify macro-trends for R&D decision makers and is primarily based on a clustering-based topic identification model, a multiple science data sources integration model, and a semi-automated fuzzy set-based TRM composing model with expert aid. With similar goals, Joung and Kim propose technical keywordbased analysis of patents to monitor emerging technologies [11]. The approach includes the automatic selection of keywords and the identification of the relatedness among them. This task is based on the analysis of a technical keyword-context matrix, which is obtained by means of text-mining tools and techniques.

However, when it comes to introduce a consistent forecasting method based on ST&I data, there is a lack of time series analysis (TSA) methods. In terms of statistical methods, the most common approach for forecasting the future evolution of a technology based on bibliometric data is growth curve analysis (see [12] for further discussion). When it comes to combine scientometrics and TF, the inclusion of specific time series models is hardly encountered within the reference literature (see for example [13, 14]). What is more, the time series commonly take the frequency of generic items, such as patents or articles, as indicators without going down to a lower level, such as keywords, which provide richer information about the technology or field that is being analyzed. This kind of strategy is roughly chosen by Park and Jun [15] within the patent analysis field. Here, time series regression and clustering techniques are combined to construct a technological trend model of identified clusters, and that furthermore, these clusters are described by means of top keywords.

The following section describes the proposed approach, which is based on the combination of methods and techniques discussed here, in an attempt to identify an optimal combination of the most representative ones.
