Co-occurrence prediction of novel ideas in the annual Samahang Pisika ng Pilipinas conferences from 1982 to 2024

Abstract

Trends in the presentation of novel research ideas may be analyzed using a dynamic topic co-occurrence network. We formed such a network to analyze the technical papers in the yearly Samahang Pisika ng Pilipinas (SPP) conferences from 1982 to 2024 (43 years) based on the list of keywords culled from SPP papers from 2008 to 2024. Natural language processing and manual filtering were employed to generate a systematic topic list and a four-layer neural network (NN) was trained to predict co-occurring topics from 2020 to 2024 using 14 features calculated from the co-occurrence network from 1982 to 2019. We limit our prediction to nodes that are common between a network constructed from 2020 to 2024 and one that uses a 1982-2019 topic-data set. We found that NN prediction quality has a trade-off between recall and precision, and that the optimal threshold probability yields a precision of 0.346 and a recall of 0.339. Our approach yields insightful information about the temporal behavior of novel research topics in SPP conferences showing how Filipino researchers respond to emerging global trends.