Pemanfaatan text summarization dengan Support Vector Machine dan K-nearest neighbor pada analisis sentimen untuk mempermudah pengguna membaca review game STEAM

Abstract

Today the development of the game is increasing and in line with the growth of the players. Usually, these players who are often called players have a special platform to see the latest game developments. One that is often targeted is Steam, where the platform provides complete information such as reviews, prices, release dates, and so on for users who want to buy games. Usually before buying a game the user will see a review first. The number of reviews on Steam makes it difficult for users to find information. From these problems, text summarization was carried out to summarize information and sentiment analysis to assess the value of the game. In order to get a good summary of the information, it is necessary to go through several data processing processes. The game review data collection process is obtained through the available Steam API. Once collected, preprocessing will be carried out to overcome the varied and inconsistent data that can affect the training process. Preprocessing includes Tokenization, Stopwords Removal, and Stemming. The text summarization process for feature to vector uses TF-IDF and Sentiment Score to get the main sentence before the training process using SVM is carried out. The classification process uses the KNN method which compares each game review data whether the data is closer to positive or negative, thus helping users when viewing game information becomes shorter and easier. Measurement of the success of this method in answering problems by testing data with the Confusion Matrix and surveying Steam users. The use of text summarization for each game review has little role in improving the results of sentiment analysis, because the method is not suitable and the game review data is in the form of an abstract. The accuracy of sentiment analysis is still better when text summarization is not carried out on the data. A total of 50 people were asked to provide statements regarding the results of sentiment analysis and text summarization. The results obtained by 40 out of 50 users said the application helped read game reviews and 10 others did not.