Research On Consumer Perceived Risk Measurement Based On Social Media Comments

In the research field of the consumer market,the theory of consumer perceived risk developed based on psychology is one of the essential theories of management,and perceived risk is an important factor affecting consumer behavior.The traditional measurement method of consumers’ perceived risk is mainly to design the scale of psychological structure,and data collection is usually carried out in the form of a questionnaire survey or interview.However,studies using traditional data acquisition methods have a sample size,timeliness,cycle,spatial span,and richness limitations.Therefore,the psychological perception of most consumers cannot be mastered,and some studies are faced with a repeatability crisis.With the popularity of the mobile Internet,social media has broken the boundary limits of time,space,and geographical location in the physical world and gradually penetrated people’s work and life.In particular,since the outbreak of COVID-19,social media such as Weibo,Douyin,Toutiao,and Watermelon videos have gradually evolved into a virtual channel for consumers to evaluate products,with the daily release of information on product evaluation exceeding millions.As social media is equal,multilateral,and open,consumers can convey their views,attitudes,and emotions through text,video,symbols,and other digital forms.Comments on social media can more intuitively reflect the true feelings of publishers and become the primary carrier for consumers to express their psychological cognition and emotions.Therefore,social media comments contain rich and measurable research data on consumers’ perceived risk.Under the research framework of psychology and consumer perceived risk theory,this paper obtains multi-source social media comments through an intelligent computer program and combs and reviews previous studies.On this basis,with the help of natural language processing based on machine learning method,the consumer comment text is processed to build the consumer perceived risk measurement process,which consists of three interrelated parts:The first part is the creation of a social media comment corpus.The establishment of social media comment corpus provides a data basis for accurately,comprehensively,and timely measuring the perceived risk of consumers of different commodities,overcomes the limitations of traditional questionnaires and interviews,and thus improves the comprehensiveness and richness of data.It is worth noting that different commodities have different characteristics and audiences,and the connotation and external performance reflected in comments vary greatly.Therefore,it is necessary to compare the measurement process and results of different commodities’ perceived risk.Therefore,this paper chooses two commodities electric vehicles and online games with considerable differences in product form,marketing mode,operation mode,and audience to construct a corpus to test whether the measurement framework designed in this study has good adaptability and expansibility.In addition,after data pretreatment,part of the data was randomly selected for manual annotation under the guidance of consumer perceived risk theory,forming the data basis required by the consumer perceived risk measurement framework.Secondly,the framework of consumer perceived risk measurement based on the machine learning methods is under-constructed.In this paper,three tasks of“thematic feature mining--semantic recognition of perceived risk--sentiment analysis” are constructed by referring to the cutting-edge achievements of Chinese natural language processing to achieve the measurement goal of consumers’ perceived risk.(1)Mining tasks of subject features of perceived risk.The purpose is to determine the classification labels based on the perceived risk characteristic attributes to serve as the label basis for the multi-label classification model to identify the multi-category perceived risk semantics in the comments in the second task.Specifically,to solve the compatibility problem between machine learning methods and consumer perceived risk theory,this study adopts the Labeled Dirichlet Allocation model in the feature mining of perceived risk subjects based on previously manually Labeled data.The Labeled LDA model identifies the categories of perceived risk topics of consumers of different commodities.Then,based on the previous research on consumer perceived risk of goods,the iterative matching process is used to verify and summarize to improve the compatibility of label setting and perceived risk theory.(2)Identification task of perceived risk semantics in consumer comment text.In order to identify topics that may cover multiple perceived risks from the semantic expression of any consumer review,a multi-label text classification model is designed in this study.This model integrates a convolutional neural network and a Long Short-Term Memory artificial neural network(CNN-LSTM).Based on the mining results of perceived risk topics in the first task,it analyzes and obtains the distribution results of the number of single or combined perceived risk comments and completes the complex semantic classification and recognition of different perceived risk topics.(3)Sentiment analysis task of the commentary text.Because the second task can only show consumers’ attention to or cognition of various perceived risks,it cannot provide insight and clarify consumers’ emotional positions on different perceived risks of goods.Therefore,in the third task,this study designed a sentiment analysis model(BERT-Bi LSTM-Attention),with dynamic text vectorization and bidirectional LSTM with the attention mechanism,to interpret and supplement the semantic recognition results of perceived risk in the second task to obtain the emotional tendency of each perceived risk in consumer psychology.Finally,based on the experimental results of extended research and management implications.(1)Discussion based on experimental results.Based on the experimental results of this study,the theme and emotional tendency of the perceived risk of electric vehicles and online games are deeply analyzed.In addition,the co-occurrence network of perceived risk topics is generated by using the analysis method of a social network,revealing the combination of perceived risk topics that are frequently co-occurrence in social media comments.(2)Empirical research based on experimental results.The experimental results of each perceived risk were matched with significant events of the same period.Furthermore,the timeline-based dynamic comment volume changes and emotional fluctuations were explored from empirical research to test the timeliness and effectiveness of consumer perceived risk measurement in social media comments.(3)Comparative analysis based on experimental results.Horizontally,the experimental results of different commodities are compared;Longitudinally,the results of existing consumer perceived risk studies are compared.It further demonstrates the feasibility and innovation of this study.(4)Management implications from the perspective of government and enterprises based on the experimental results.Given the government and enterprises,this paper analyzes the experimental results to give corresponding management enlightenment to the government and enterprises and expounds on the management practice significance of this study.The innovation of this study mainly includes the following aspects:First,compared with the data acquisition method of traditional psychometric research,this study innovatively took large-scale social media comment data as the driving force,which enriched the research perspective of consumer perceived risk theory.In this paper,after its creation,the corpus of social media comments on electric vehicles and online game products related to consumers’ perceived risks is manually annotated for times,trained,validated,and tested by machine learning models.It can provide new data support for future relevant research.Second,according to system theory,this study creatively designed,constructed,and validated a consumer perceived risk measurement framework.This framework consists of a three-task sequence: “Topic mining,Perceived Risk Semantic Identification,Sentiment Analysis.” The first and second tasks of the measurement framework can be iterable to ensure the recognization of the comment corpus and optimization of the matching between Perceived Risk topics(granularity)and the corpus.By this mechanism,the dimensions of consumers’ perceived risks preset by referring to previous studies are constantly verified,modified,and adapted to the objective results mined from comments to generate new perceived risk topics(granularity),thus ensuring that the process of machine learning mining and identification is placed in the interpretable category of relevant theoretical studies.At the same time,the newly discovered topics of perceived risk in the corpus can be quickly reviewed into theories for testing and supplemented with empirical or revision of existing theories,ensuring the compatibility of machine learning and perceived risk theories in the research field of consumer perceived risk measurement.More importantly,the verification,modification,and adaptation process not only ensure the measurement framework’s ability to discover new perceived risk topics but also can easily generalize the measurement of the perceived risk of different commodities.Thirdly,the sentiment analysis task is innovatively incorporated into the perceived risk measurement framework.It breaks through the technical problem that only the number of consumer comments on various risk topics can be quantified in the semantic recognition of consumers’ perceived risks,realizes the measurement of perceived risk attention and cognition,and identifies the emotional tendency of consumers to each perceived risk.Specifically,introducing the dynamic vector transformation model and attention mechanism into the sentiment analysis model designed in this study improved the effect of sentiment analysis.Based on sentiment analysis results,each dimension of perceived risk and granularity of sentiment orientation were defined.In addition,for the identification results of various dimensions and granularity of perceived risks,the support,neutral,and opposition positions of consumers are obtained,which provides methods and technical support for judging the overall cognitive strength of consumers on various perceived risks.In general,based on the innovative idea of “retrospective-style”,this paper reexamines and discovers research problems,excavates research value,and uses research results to discover new changes in time,helping to test,revise and supplement the existing research basis of consumer perceived risk measurement.The framework of consumer perceived risk measurement based on social media comment data proposed in this paper can not only obtain multi-dimensional and fine-grained consumer perceived risk perception and the emotional tendency of different commodities but also capture the fluctuation caused by time and events.The measurement framework has adaptability,timeliness,and extendibility,which can be a reference for cutting-edge research and data analysis in the current consumer market.To some extent,the research of this paper promotes the crossover research of computer science,psychology,and management in consumer perceived risk theory.The research results also support the adjustment and implementation of the government’s industrial strategy to promote sustainable development and provide a reference for the strategic planning and marketing strategy formulation of relevant industries and enterprises.