Exploring popular topic models

Abstract

Information from micro-blogging site such as Twitter is a huge repository of data. A lot of research is happening on sentiments, discovering patterns and prediction. One challenging task is dividing this humongous unstructured data into clusters. Several topic modeling methods are proposed by researchers. This paper presents a brief summary of topic modeling methods LDA, LSI and NMF and their applications. Experiments are conducted on the Twitter based datasets created using tweets on keywords Cauvery river, Lokpal bill and Rahul Gandhi. Paper covers a brief discussion on evaluating the accuracy of topics formed using perplexity, log-likelihood and topic coherence measures. Best topics formed are then fed to the Logistic regression model. The model created is showing better accuracy with LDA.

A Seed-Guided Latent Dirichlet Allocation Approach to Predict the Personality of Online Users Using the PEN Model

“…The outperformance of LDA compare to NMF also can be seen in Moreno et al [38], who focused on predicting the Big 5 traits. Our literature review also showed that standard LDA yielded better results compare to LSA in nonpersonality topic modeling study [55]. However, as expected, our experiment indicated that the seed-guided model performed better than nonseeded topic models where the coherence scores of SLDA and GNMF are more than 50 in all the analyses disclosed in Table 4.…”

“…Prior to evaluating the topics generated by SLDA, we used three nonseeded models, namely LDA [40], NMF [55], and LSA [55] as well as a seed-guided model called GNMF [42] to make comparisons with our proposed model. We also conducted an analysis to determine the sensitivity of seed words against the performance of SLDA and GNMF.…”

“…We also conducted an analysis to determine the sensitivity of seed words against the performance of SLDA and GNMF. The LDA, NMF and LSA were chosen due to their popularity in topic modeling and affective computing [31,38,42,55]. The parameter settings for LDA and NMF, respectively, followed the same settings applied in SLDA and GNMF [42], whereas the settings applied in [55] were adopted for LSA.…”

“…The LDA, NMF and LSA were chosen due to their popularity in topic modeling and affective computing [31,38,42,55]. The parameter settings for LDA and NMF, respectively, followed the same settings applied in SLDA and GNMF [42], whereas the settings applied in [55] were adopted for LSA. For GNMF, we followed the settings applied in [42] and evaluated the performance based on a number of high probable seed words in S as the vocabulary played an integral part in the seeded algorithms.…”

Pengamatan Tren Ulasan Hotel Menggunakan Pemodelan Topik Berbasis Latent Dirichlet Allocation

“…Dalam analisis sentimen dan topic modeling pada data Twitter yang berisi "#IndiaFightsCorona" untuk menganalisis opini publik, LDA berkinerja lebih baik untuk sentimen positif, dan untuk sentimen negatif, LSA berkinerja lebih baik. Eksperimen dilakukan pada kumpulan data berbasis Twitter yang dibuat menggunakan twit dengan kata kunci sungai Cauvery, tagihan Lokpal dan Rahul Gandhi menggunakan metode permodelan topik LDA, LSI dan NMF yang mengevaluasi keakuratan topik yang dibentuk dengan menggunakan langkah-langkah confussion, kemungkinan log, dan koherensi topik dimana topik terbaik yang terbentuk kemudian diumpankan ke model regresi Logistik dan diketahui model yang dibuat menunjukkan akurasi yang lebih baik dengan LDA [6].…”