Ekstraksi Ciri Produktivitas Dinamis untuk Prediksi Topik Pakar dengan Model Discrete Choice

Abstract

Recommendation of active or productive experts is indispensable in supporting collaborations. Activities of publication and citation indicate expert productivity. An expert can be inferred to have an interest in a subject through productivity in that particular topic. Since an expert can change interests over time, the contribution of this paper is a Discrete Choice Model (DCM) based on topic productivities to predict the primary interests of the experts. DCM uses features extracted from bibliographic data of citation relation and title-abstract texts. Before extracting productivity features and dynamicity features to represent interest changes, title clustering with KMeans++ is used to identify research topics. There are six productivity features and five dynamicity values for each productivity feature to demonstrate the expert behavior. Therefore, a clustered topic as a research interest is represented as an expert choice with 30 extracted features in the proposed method. The experiments used multinomial logistic regression for DCM and a log-likelihood indicator for the fitted models of the features. The resulted DCM models showed that productive behavior of the experts by doing many publications and receiving many citations effected to the precision of topic prediction by 80%. Some features were better for predicting primary interests of the expert. It was demonstrated with a lower precision value of 60% by using features that represent the expert behavior of only doing publication or only getting citation.

Identifying collaboration dynamics of bipartite author-topic networks with the influences of interest changes

“…From the KMeans ++ results, less cluster number was suggested although the Silhouette scores were not satisfactory with less than 0.5. 1 3 Therefore, we cautiously continued on other clustering scenarios which leaded to better results of higher Silhouette score (> 0.5) by only using title texts and made a smaller feature matrix (Purwitasari et al 2018a).…”

Ekstraksi Fitur Produktivitas Dinamis berdasarkan Topik Artikel Ilmiah untuk Klasterisasi Peneliti

“…Kata-kata yang didapatkan dari informasi tekstual diklasterisasi untuk menghasilkan klaster topik penelitian. Salah satu penelitian terdahulu telah melakukan klasterisasi kata-kata yang didapatkan dari informasi tekstual artikel ilmiah menjadi beberapa klaster topik dengan menggunakan metode K-Means [11]. Pada penelitian lainnya, dilakukan klasterisasi kata-kata menggunakan metode Latent Dirichlet Allocation (LDA) [12] dengan sumber data artikel ilmiah peneliti.…”