Detection of Reference Topics and Suggestions using Latent Dirichlet Allocation (LDA)

Pelatihan Aplikasi Teknologi Informasi (PATI) is an activity of training required for new students in Universitas Muhammadiyah Malang (UMM) to provide knowledge and training on UMM or information technology concerned about general technology. At the end of the training, the students give the conclusions and suggestions to PATI. During this event, the training Committee gave less concern in term of the inference from students to provide a material evaluation. The primary factor originated from the commenting processes which should be performed one by one. Therefore, the comprehensive method should be implemented by modelling using Latent Dirichlet Allocation (LDA) in order to facilitate the Committee to undertake an analysis of the conclusions and suggestions. LDA is a “generative probabilistic model” of a collection of composites made up of parts. In terms of topic modeling, the composites are documents and the parts are words and/or phrases (n-grams). Conclusions and suggestions are taken as many as 1025 data from PATI 2016/2017. Based on such research, modelling of LDA identifies the 7 topics in the overall data. The process of analysis is done by external details each comment contains what topics. The evaluation is done by testing 250 data to determine the results of the conformity between the results of the analysis of the system as well as actual results obtained from respondents. The test results obtained accuracy of 83.6%.