Pattern Recognition of Effective Online Classified Advertisement
Kanyawee Pornsawangdee, Unchalisa Taetragool · 2019
Classified advertising is one of the most popular online advertising especially for second-hand goods. One of the main functions of online classifieds is to help sellers sell their items quickly and easily. Optimized advertising is then essential to maintain user engagement in the platform. In this research, the factors and characteristics of online classified ads including structured data and free text that may affect the effectiveness of the advertising in terms of the ability to close the sales are investigated. Initially, we collected 148K classified ads data in an automotive category that have the published date between January 1, 2018, and July 31, 2018, provided by Kaidee, the top Thai online marketplace. The data is then preprocessed by data cleaning and features extraction techniques to find the important features for using in the analysis. We divide the focusing features into 4 main groups: ad components, car attributes, price, and text features. We apply k-means clustering integrated with decision tree classification in modeling to classify the ability to close the sales on Kaidee platform within 30 days. The predictive efficiency of the models will be measured and compared using accuracy, precision, recall, and F-scores by using the 3-fold Cross Validation method. Finally, we analyze the important features that are related to closing the sales of each feature group.