Research On Methods For Suggestion Mining From Social Network

In the new normal era,we can see the large change of digital transformation,customer behaviors and business adjustments.To success in this era knowing customer need is very important for the business direction.However,these changes impacted to the businesses on two opposite ways;Positive way,if the businesses can adjust and improve themselves to serve customer demand on time,their business will achieve a leap growth and make a huge profit.In the opposite way,if the businesses cannot understand customer needs or understand them too slow,their business will lose the big chance or fail.Moreover,nowadays most of people are not only use social media and shopping online but they also review the products or services on the network.However,the enormous texts and reviews on the social media have many kinds of propose such as to report their problems,to complain,to admire or to give the suggestion.This research studies about big data from the network especially the suggestion mining.I expect that if the business owner can select the suggestion of the customers faster than normal,the businesses will survive and still can grow up even though the whole world economic still slowdown.So that to use the knowledge from Information Technology can help this process to decrease the human resources,expenses and times.This research studies about the method for suggestion mining:(1)Suggestion Classification(2)Suggestion Types Classification.My experiment collected the customer reviews 2,561 sentences and compared suggestions classifications performances of Decision tree,Na(?)ve Bayes and Support Vector Machine.My experimental results shown SVM with polynomial kernel is the best classifier by analysis term along with frequency of couple terms tagged in vector,with the Precision,Recall and F-Measure are equal to 85.75%,93.62% and 89.51%respectively.For Suggestion type classification,I use macro averaging to measurement the performance with the Precision and Recall are equal to 94.94%and 94.94% respectively.Results show that our suggestion mining framework has good performance for disambiguation suggestions sentences and can reduce time consumption to read all customer reviews.