Consumer Intention Mining On Time-Sync Comments

Internet technology has brought people a convenient way to interact with information.With the development of technology,as a new text communication method,time-sync comment has been accepted and used by the majority of netizens.Time-sync comment refers to the text message that appears directly on the video.It can be displayed on the video in the way of scrolling,staying,and more special effects.It is a short comment sent by the video viewer.With the wide application of time-sync comments,the demand of mining texts containing consumption intention is also increasing.Due to the time-sync comments' characteristics of extremely short text length and serious lack of semantics,the existing consumer intention mining methods cannot achieve better results in the field of time-sync comments.Combining the textual characteristics of time-sync comments,this thesis researches consumer intention mining methods that are more suitable for time-sync comments environments.Specifically,the following research works have been completed:1.The acquisition and analysis of time-sync comments data.Time-sync comment,as the main text form in video media,is rarely studied in previous research.According to the current research status in this field,this thesis firstly collects the text of the time-sync comments,which mainly uses two methods,namely the method of calling the official API of the video website and the method of simulating the login-in page.The next step is to analyze the obtained data,summarize the text features of time-sync comments field,and apply these data features to the subsequent research.2.The identification of time-sync comments consumption intention.Considering that the text of the time-sync comments is concise,and some structures are usually omitted,it is difficult for existing text-based consumption intention recognition methods to achieve ideal results.Therefore,this thesis proposes a template-based consumption intention recognition method.The method firstly uses the template extraction method based on Prefix Span algorithm to generate consumption intention templates automatically,then the text matching method is applied to convert the time-sync comments into the template feature vectors with the help of the consumption intention templates generated previously.Next,the template vectors transformed from theprevious step are inputed to train the consumption intention recognition model.Finally,the consumption intention of the time-sync comments under this model is identified.Experiments show that the method can effectively improve the speed and accuracy of consumption intention recognition in time-sync comments compared with the traditional text feature based classification model.3.The identification of time-sync comments consumption intention objects.For those time-sync comments with determinate consumption intention,this thesis proposes a method of TF-SI-IDF(Term Frequency–Syntactic Relations & Interactive Information-Inverse Document Frequency)for identifying consumer intention objectsby combining TF-IDF(Term Frequency–Inverse Document Frequency),syntactic relations and interactive information to extract consumer intention objects in the time-sync comments.In the extraction of explicit consumption intention objects,compared with the original consumption intention object extraction method,the extraction method proposed in this thesis has a better extraction effect by combining syntactic features and interactive information of time-sync comments.In the extraction of implicit consumption intention objects,this thesis firstly uses the interactive information of the time-sync comments to supplement the text with implicit consumption intention.Then the method proposed in this thesis is used to extract complementary words.Experiments show that this method can perform the extraction of consumption intention objects well,which provides a new idea for the extraction of implicit consumption intention objects.