Research On Analysis Methods For Customer Sentiment And Purchase Intention Based On Multi-modal Learning

Along with China’s entering the era of digital economy,e-commerce is becoming one of the most important components of digital economy.Digital technologies,such as big data,cloud computing and artificial intelligence provide sufficient support for the development of e-commerce.New forms and models of e-commerce,such as social ecommerce,live streaming e-commerce and cross-border e-commerce,provide users with diversified and immersive consumption experience.Therefore,massive e-commerce platform data reflect consumers’ will and behavior in terms of multiple dimensions,and online product visible information gradually presents a trend of multi-modal information symbiosis.Rational use of multi-modal data of e-commerce platforms,fully mining and intelligent analysis of consumers’ purchase intentions,will help enterprises and businesses optimize marketing strategies and improve service quality which deepens the cognition of the target audience and increase the attention of consumers.Therefore,based on the characteristics of multi-modal data of e-commerce platform,a high-performance model is constructed to satisfy the needs of online activities,and obviously the model training is compatible with precision marketing practice.The research has presented reasonable research significance and application value towards precision targeting and e-commerce marketing.Under the background of the coexistence of multiple modalities of the ecommerce platform data,Many ’how’ have become the key problems for the e-commerce platforms to achieve the precision marketing such as:how to efficiently analyze consumers’purchase intentions,how to sort out the relationship between consumer groups and finishing dividing jobs,and how to discover potential consumer groups etc.Based on the theoretical foundations of multi-modal sentiment analysis,graph neural network clustering,online learning,group consistency,etc.,this paper fits with the research goals of precision marketing.In order to improve the insight and ability of precise marketing data analy sis and effectively solves the problems related to the analysis of consumers’ purchasing intention,this paper takes the analysis of consumer purchasing intention as the research method,constructs multi-modal sentiment analysis,multi-modal consumer clustering and potential consumer prediction models,and explores the analysis method system for multi-modal ecommerce platform data.The core work and significance of this paper are summarized as follows:Nowadays,the e-commerce platform user information presents the form of text,video,picture.audio etc.in the state of the multi-modal integration and co-occurrence.However,the classic consumer sentiment analysis method relies too much on text data information.Research on multi-modal information fusion,correlation between internal features of each modality and cross-modal features needs to be strengthened.To the problems above,this paper studies the feature extraction method of different modal commodity review data,and proposes a multi-modal sentiment analysis model based on multi-attention mechanism based on the embedded coding of multi-modal data.The model mines the feature correlation within each modality through self-modal attention mapping,and uses cross-modal attention mapping to mine the feature correlation between different modalities,and integrates the inter-modal correlations.Realize multi-modal information fusion.It completes the function of multi-modal sentiment classification based on ensemble deep learning.The multi-modal sentiment analysis model based on the multi-attention mechanism breaks through the limitation of the traditional sentiment analysis model that relies too much on text review data,gives full play to the complementary advantages of multiple modal information,and strengthens the ability of consumer sentiment analysis.In the auxiliary practice of precision marketing,the model established in this paper can be used to obtain the user experience analysis results of different attributes of the same product,and the difference analysis results of the attractiveness of products with similar market positioning but different brands to consumers.Reasonable clustering of e-commerce consumers is both the basis of market segmentation and brand positioning,and one of the key directions of e-commerce research.It is not only beneficial to deepening the understanding of target consumer groups but also enhance the ability to analyze consumer groups’ purchase intentions by improving the clustering effect of e-commerce consumers.When consumers use the e-commerce platform to shop,they also provide the e-commerce platform with rich multi-modal data,including shopping records,login logs,comments and other data.The relationship between consumers is an important basis for the clustering of consumer groups.In this paper,the relationship between consumers is represented by multi-modal graph structure data,and the multi-modal spectrum clustering method is used to achieve user clustering.On this basis,a multi-level multi-modal graph attention network model is constructed.The model combines graph neural network and multi-modal learning,which can extract different modal features of consumers at multiple levels,and has the ability to understand consumer class cluster topology.At the same time,it can realize the effective integration of different modalities and different levels of features,and further improve the effect of consumer clustering.In the auxiliary practice of precision marketing,using the multi-level and multi-modal attention network model of this paper,the consumers of the e-commerce platform can be clustered and divided,and the audience of the analyzed products can be analyzed in depth,including product audience analysis.,product positioning and comparative analysis of sales feedback,etc.E-commerce platforms update a large amount of multi-modal information every day(such as multi-modal data of new products,updated data of consumer behavior,etc.),For them,the massive data generated by platform merchants and consumers on a daily basis often contains rich trend information.In the face of these emerging large-scale multi-modal new data,the traditional static learning mode can only merge the new data with the existing data,and repeat the training of the model with the expanded overall data set,which will lead to some problems like:long model learning time and low efficiency.On the other hand,it is difficult for the static learning mode to realize the targeted model update adjustment for incremental data,which makes the model itself unable to efficiently deal with the problems of concept transfer and concept evolution in incremental data.In view of the above problems.this paper studies a multi-modal sentiment analysis model with online learning ability to meet the application scenario requirements of large-scale multi-modal data generated by ecommerce platforms every day.At the same time,it proposes an online multi-modal sentiment analysis model,which has an online self-modal mapping mechanism and an online cross-modal multi-attention mechanism to deal with the self-modal and cross-modal information fusion problems in the online learning process,respectively.When it is faced with the multi-modal information of online generation and fusion,an online integrated deep learning sentiment analysis method is proposed to ensure the online learning mode of sentiment classification,and provide an efficient solution for online sentiment analysis tasks.In the auxiliary practice of precision marketing,the application of the model proposed in this paper can dynamically analyze consumers’ emotions,so as to further calculate the changes of consumers’ purchasing tendency,and provide a powerful tool for dynamic tracking and analysis of consumer groups’ purchasing intentions.The purchase intention of consumer groups is directly related to the market potential of commodities and the formulation of marketing strategies.The analysis model based on individual users neither directly reflect the emotional preferences of consumer groups,nor can it effectively tap the purchase intention of potential consumer groups.In view of the above problems,this paper studies the purchase intention of potential consumer groups,aiming to solve the problem of predicting the distribution of potential consumer groups of goods under the condition of limited feedback information.This paper proposes a potential consumer group prediction method based on group consistency.It uses the imbalanced learning strategy to break through the constraints of traditional group consistency calculation,and realizes the consistency calculation between purchased consumer groups and potential consumer groups in various clusters.In terms of the auxiliary practice of precision marketing,this paper proposes to build a quantitative analysis tool for the potential consumer groups of commodities on the e-commerce platform,to provide sufficient data analysis solutions for the forecasting of the sales prospects of related commodities and the analysis of potential audiences of commodities in precision marketing.In terms of theory,the method proposed in this paper extends the sentiment analysis of individual consumers to the sentiment analysis of consumer groups.Through the consistent calculation results of subgroups within the cluster(subgroups of consumers who have shopped and subgroups of consumers who have not yet shopped in the same cluster),the distribution of potential consumers can be estimated,which effectively reduces the direct prediction of potential consumption from consumers who have shopped.biases and risks.In practice,the application of the model proposed in this paper can realize the prediction of the cluster attribution distribution and preference level distribution of potential consumer groups,which is of great significance in the prediction of commodity market potential and the formulation of marketing strategies.