A scientific research topic trend prediction model based on multi‐LSTM and graph convolutional network

Abstract

In recent years, with the increase of social investment in scientific research, the number of research results in various fields has increased significantly. Accurately and effectively predicting the trends of future research topics can help researchers discover future research hotspots.However, due to the increasingly close correlation between various research themes, there is a certain dependency relationship between a large number of research themes. Viewing a single research theme in isolation and using traditional sequence problem processing methods cannot effectively explore the spatial dependencies between these research themes. To simultaneously capture the spatial dependencies and temporal changes between research topics, we propose a deep neural network-based research topic hotness prediction algorithm, a spatiotemporal convolutional network model. Our model combines a graph convolutional neural network (GCN) and Temporal Convolutional Network (TCN), specifically, GCNs are used to learn the spatial dependencies of research topics a and use space dependence to strengthen spatial characteristics. TCN is used to learn the dynamics of research topics' trends. Optimization is based on the calculation of weighted losses based on time distance. Compared with the current mainstream sequence prediction models and similar spatiotemporal models on the paper datasets, experiments show that, in research topic prediction tasks, our model can effectively capture spatiotemporal relationships and the predictions outperform state-of-art baselines.

Predicting the Trending Research Topics by Deep Neural Network Based Content Analysis

“…In [16], scientific research topic trend prediction problem is modeled as prediction of token distributions over publication observations. Papers as well as research fields are represented with one hot encoding of tokens.…”

“…Among the related work, there are studies that use neural network based solutions such as those given in [25], [13], [15] and [16]. However, there are basic differences in problem definition and data modeling.…”

“…In [15], topics are represented with feature vectors where several of the features, such as Web of Science categories, are possibly extracted from external resources. In [14] and [16], the problem modeling differs from our work such that the papers and fields are represented by one hot encoding of the topics and the influence among the fields is used for predicting the topic distributions. On the other hand, in our approach we directly focus on predicting the trend of the keyword by using only the publication history of the venues.…”

“…Experiments presented in [9] similarly utilize the last 5 years as an observation window. The problem definition and model outputs are similar in [10] and [16], they detect keyword set as conference topics for the next year for all conferences in the experiments. Unfortunately this modeling is not compatible with our experiments.…”

“…Academic research topic trend prediction problem has been studied earlier, but mostly through more conventional data mining and machine learning approaches [10], [12]. There are recent efforts focusing on research topic prediction [13]- [15] and topic trend prediction [16], but they rely on handcrafted features and additional information such as a semantic network, citation information or influence among research fields and venues. There are also related studies in the literature to detect hot topics [17] rather than trend prediction.…”