Forecasting Emerging Trends from Scientific Literature

Forecasting Emerging Trends from Scientific Literature Kartik Asooja, Georgeta Bordea, Gabriela Vulcu, Paul Buitelaar Insight Centre for Data Analytics NUI Galway, Ireland firstname.lastname@insight-centre.org Abstract Text analysis methods for the automatic identification of emerging technologies by analyzing the scientific publications, are gaining attention because of their socio-economic impact. The approaches so far have been mainly focused on retrospective analysis by mapping scientific topic evolution over time. We propose regression based approaches to predict future keyword distribution. The prediction is based on historical data of the keywords, which in our case, are LREC conference proceedings. Considering the insufficient number of data points available from LREC proceedings, we do not employ standard time series forecasting methods. We form a dataset by extracting the keywords from previous year proceedings and quantify their yearly relevance using tf-idf scores. This dataset additionally contains ranked lists of related keywords and experts for each keyword. Keywords: Trend analysis, Regression, Emerging trends, Keyword extraction 1. Introduction The rapidly changing landscape of scientific research calls for automated text analysis methods that reliably monitor and identify emerging trends from scientific publications. In recent years, emerging technologies draw an increased interest from research and development departments, aca- demic publishing companies, and policy makers, because of their high potential for socio-economic impact. Another area where early identification of growing trends plays an import role is for organizing scientific dissemination events. A broad overview of emergent trends is needed for confer- ence organisers to be able to identify hot topics for call for papers and recruit reviewers that are experts in these areas. In social science, an emergent technology is defined in terms of novelty, fast growth, coherence within a commu- nity of practice, and prominent impact (Rotolo et al., 2015). But a certain level of uncertainty and ambiguity is associ- ated with emerging technologies, as their future impact is unclear and there is little consensus on their terminologies and meaning. The method proposed in this work models the growth pattern of a scientific topic to predict its distri- bution in the future. Although approaches based on topic models have received a lot of attention, they still require a considerable amount of manual work and expert knowl- edge to identify a collective topic for human-readable la- bels. Therefore, in this work we rely on previous work on keyphrase extraction to identify scientific topics (Bordea et al., 2013). Recent analysis of keyphrase creation dynamics shows that they are frequently used in scientific discourse to signal novelty (Adar and Datta, 2015). Solutions based on citation analysis are less appropriate for contemporary analysis of emergent trends, as this type of data is less robust for recent documents and can not be ap- plied as soon as documents become available. Where the majority of previous approaches apply retrospective analy- sis, by mapping the evolution of scientific topics over time (Zhou et al., 2006; He et al., 2009; Bolelli et al., 2009), in this paper, we address the problem of actually predicting temporal distribution of the keywords at some …

Content truncated to 500 words