Building Effective Features based on Automatic Learning for Smart Search

Product title quality greatly affects natural language processing problems performed on their data domain. This research work proposes some suitable features for Vietnamese product titles. The provided product title will be considered and in turn features will be extracted through three versions: original, processed, and unmarked. Features are divided into four groups: Binary Features; Feature group based on frequency of occurrence of words and repeating words; Feature groups based on dictionaries; Feature groups based on word2vec. The effectiveness of each feature group as well as the combination between them are evaluated through the title classification problem. The experimental results confirmed the importance of the proposed features.