Topic-based classification and identification of global trends for startup companies

Abstract

To foresee global economic trends, one needs to understand the present startup companies that soon may become new market leaders. In this paper, we explore textual descriptions of more than 250 thousand startups in the Crunchbase database. We analyze the 2009–2019 period by using topic modeling. We propose a novel classification of startup companies free from expert bias that contains 38 topics and quantifies the weight of each of these topics for all the startups. Taking the year of establishment and geographical location of the startups into account, we measure which topics were increasing or decreasing their share over time, and which of them were predominantly present in Europe, North America, or other regions. We find that the share of startups focused on data analytics, social platforms, and financial transfers, and time management has risen, while an opposite trend is observed for mobile gaming, online news, and online social networks as well as legal and professional services. We also identify strong regional differences in topic distribution, suggesting certain concentration of the startups. For example, sustainable agriculture is presented stronger in South America and Africa, while pharmaceutics, in North America and Europe. Furthermore, we explore which pairs of topics tend to co-occur more often together, quantify how multisectoral the startups are, and which startup classes attract more investments. Finally, we compare our classification to the one existing in the Crunchbase database, demonstrating how we improve it.

Categorising content of Paris Agreement NDCs through computational linguistics

“…To train the STM model we use the following covariates: NDC version ( rst or updated), conditionality (yes or no), emissions per capita, emissions per GDP and implied emission change to 2030. To choose the appropriate number of topics, we followed the earlier literature 32,33 to compare STM models with three to 50 topics on such criteria as prediction accuracy (held-out log-likelihood), the extent to which popular words from topics overlap (exclusivity), and the degree to which words from the same topic appear in the same texts (semantic coherence). Figure S6 of the Supplementary Information summarises the results indicating that 21 topics balance these three criteria best.…”

A topic modelling analysis of white papers in security token offerings: Which topic matters for funding?

“…This process reduces researcher bias because foreknowledge of document content does not affect the topic classifications (Zhang et al, 2021). The LDA topic model is widely used in patent content analysis (Wang et al, 2015;Zhang et al, 2021) and technology topics evaluation (Li et al, 2021;Wang et al, 2020;Savin et al, 2022aSavin et al, , 2022b In order to apply LDA to the STO white papers, we first pre-processed the corpus by 1) converting words to lowercase, 2) removing standard English stop words and punctuation, and 3) lemmatizing all the words by means of the Natural Language Toolkit 6 lemmatiser. We then analysed the distribution of terms with domain experts and filtered out generic terms that appeared in more than 60% of the white papers (Zhang et al, 2021).…”

Public expectations about the impact of COVID-19 on climate action by citizens and government

“…STM is a method to classify textual responses into distinct topics developed specifically for open-ended survey responses [ 12 ]. It has already been used in about 60 articles on a wide range of topics [ 13 ], including those published on public perceptions of climate change and air pollution [ 14 – 16 ], associations with carbon taxation and its fairness perception [ 17 ], people’s beliefs about others’ climate beliefs [ 18 ] classification of startup companies and identification of their global trends [ 19 ], and studies published on environmental innovation and societal transitions [ 20 ]. Applications of topic modelling have been even broader including, for example, tracing development of agricultural and water technologies over long time [ 21 ] and understanding public attitudes towards municipal solid waste sorting policy in China [ 22 ].…”