Topic Based Brand Image Awareness And Emotion Analysis

Brand image mining is the key step in the brand image strategy, and is the first step to shape the brand image and develop the brand communication strategy. It has far-reaching significance for brand image building and brand competition. With the rapid development of Internet technology, user can contact and access to information anytime and anywhere, and create a lot of user generated content. Brand spread has be exposed to more sensitive, convenient, free high-speed transmission but uncertainty space. The traditional brand image investigation cannot meet the needs of the brand image mining in the diversity and timeliness of the sample, and in the analysis method.Under this environment, mass fragmentation of user-generated topic data provide a wealth of viable data resources and new research ideas for brand image mining. The research on brand image mining driven by topic data is the strategic basis of brand image communication in the new communication environment.The brand image generated by users includes three aspects: cognition, emotion and behavior. This paper only focuses on the cognition and emotion. First of all, we give the method of topic recognition for the brand image mining, according to different analysis objective to trade-off data scope. Secondly, we design mining methods for brand image cognition and emotion. Employ these methods on massive and fragmented user generated topic data could obtain brand image cognition of users and the individual emotion and group emotional state of users. The detailed researches are as follows:(1) Topic identification method for brand image mining. The data scope for brand image mining is different due to the different task. In this paper, we give different topic identification method. First, collect brand topic based on keywords search. Second,identify hot topic for high timeliness task. In this paper,we predict the hot topics based on curves classification model. Build heat curves using the degree of spread and the degree of attention focused. Preprocess the hot topics' heat curves, eliminate the negative effects for similarity measurement bring by original dimension. Classify the rich and changeful curves, extract their common characteristics and behavior rules,make it present a relatively clear regularity. Employ the weighted-vote scheme to predict a topic being trend or not. Based on keywords search topics and based on hot curves identify hot topics can satisfy the general requirements of brand image mining in data preparing.(2) Brand image cognition analysis method based on topics. The user's perception of the brand image refers to the user's overall impression of the brand(including function, service, utility, etc.). It is the foundation of brand development.First, extract initial recognition labels based on language rules. Second, aggregate recognition labels by synonym dictionary using Jaccard similarity. Finally, the importance of recognition labels can be calculated by TFMF model. Based on the obtained important recognition labels can understand consumers' overall perception,the most concerned characteristics and the unique characteristic comparing to the competitors of the brand.(3) Brand image individual emotion analysis method based on topics. There is an old saying: 'Psychological offense is the best of tactics'. Sentiment is a strong weapon in brand image communication. The analysis of the brand image cannot be separated from the individual emotional state. The effective extraction of sentiment label in user generated content is the basis of individual emotion analysis. New emerging words, hot words shifting, fragmentation and large dimension and sparse features generated from large amounts of most commonly used Chinese words are the main obstacles to emotional analysis of Chinese. In this paper,a novel method is proposed to solve the above problems. Construct emoticons dictionary, then automatically obtain the sentiment labels of training set to solve the problem of massive data annotation. This can save the human and financial costs of training labels, and has high objectivity.Modified G2 test was introduced for feature selection. This can preserve the features of strong classification ability without over filtering, and eliminate the interference of invalid features as much as possible. The multistage judgment sampling strategy was employed to ensure the diversity of base classifiers. The weighted majority voting was applied to combine the ensemble classifier. This approach is a swarm intelligence approach. The experiments demonstrate that the method can quickly and efficiently obtain emotional labels, retain strong distinguish features, more important, achieve higher accuracy. It is a competitive method for sentiment classification of Chinese tweets.(4) Brand image group sentiment analysis method based on topics. Sentiment is a most important role in the consumption experience. Understand the users' group sentiment state and the evolution logic of the brand image can help enterprises and users to understand the evolution of the brand image. Construct group sentiment measuring model, gather group sentiment based on individual emotion. Establish the group emotional time series data of different granularity. Through the analysis of group sentiment time-series data, understand brand group sentiment evolution. Analysis the property of hot topics issue, understand the affecting that hot topics made to brand groups sentiment evolution. Through the experiment and the study of the case can be found that different stages of brand life cycle will lead to the different evolution states of brand sentiment. The hot issues will accelerate the evolution of emotional process.The different attributes of hot issues will affect sentiment evolution direction. The handling strategy of hot issues will deepen or eliminate the impact of the hot issues.