Short Text Classification Using Contextual Analysis

Micro blogging tools provide a real time service for the public to express opinions, broadcast news and information offer an opportunity comment respond such output. Word usage in social media is continually evolving. bloggers may use different sets of words describe specific event they new (i.e. neither exist training dataset nor informal or formal dictionaries) contexts. Dynamically capturing their potential meaning from context can help reflect relationship media, which then be useful solving various problems, like classification task. Different approaches have been proposed this regard, one them Contextual Analysis. This paper focuses on examining approach grouping short texts (tweets) talking about same into category. A transparent method text multi-class categorization presented. It uses Analysis capture most important detect similar In order test efficacy these areas, study evaluates performance other well known methods, as Naïve Bayes, Support Vector Machines, K-Nearest Neighbors Convolutional Neural Networks. On average, experiments’ results show that effectively categorize tweets groups, with high f1-measure score f1>97.09% f1>95.27%, imbalanced classes number experiments, respectively. However, baseline negatively influenced by dataset. The Networks produces best among algorithms f1>97.74% all 1.73% 2.72% higher than lowest Naive Bayes Neighbors, respectively, but does not meet requirements transparency results.