| Traditionally,the most common methods of collecting affective variables(vocabularies)for products in the Kansei engineering design procedure are collecting affective variables(vocabularies)from the literature reviews,discussing or interviews.However,since the number of subjects is limited,and it can be difficult for participants to rate the affective variables due to the unfamiliar vocabularies in the questionnaires which will lead to inaccuracy.In addition,the result can be influenced by the ability of the moderator in the discussing group.Consider the background in this era of ecommerce network sales,both methods has the disadvantage of time consuming and subjectivity.On the other hand,the seller online usually also be the producer and the designer who have realized the importance of the feedback from the customer.However,they cannot extract precisely and valuable from such large size of amount of rates.Therefore,by taking a type of new national standard electric bicycle products as a carrier,this study bring a new thinking,a new method of Kansei engineering by using text mining.The study by writing a web crawler with an open-sourced programming language---python and its several function modules.The research collected several thousand of e-bicycle product reviews from the different web store pages but the same product type.After the basic procedures of data cleaning,after using ‘jieba’ segment tools and statistical analysis of words,a word list can be generated from the e-bicycle customer reviews.Then with a highly experienced e-bicycle design and manufacturing engineer,an e-bicycle product professional word list can we select and construct finally,which considered the specific Chinese language features not only its synonyms and antonyms of words,but also the dialects of different regions,even mistakes.The list covered affective vocabulary,the pieces name of the e-bicycle product,the interaction vocabulary,the scenarios,the specific users,service and price around,and so on.By selecting the word dimension from the professional word list,all the customer reviews were transformed into a multi-dimensional vector group or matrix of 34*3,964 with the idea of "word bag model" which had been adjust by the properties of the task,hence finished the vectorization of the text.With the help of the SOM Toolbox toolkit on the MATLAB platform,the SOM neural network algorithm was implemented successfully.The algorithm helped project the high-dimensional comment data onto a two-dimensional six-lattice SOM plane of 13*13 neurons and form a vision of clusters.After extracting the comment samples of the winning nodes in different clusters on the SOM plane,the cluster division and the clustering effect of the algorithm could be assessed by judging the difference of the nodes based on word frequency statistics.The research on interactive experience processing was taken as an example to run the SOM neural network in five dimensions of "affective vocabulary","basic driving experience","hand interaction experience","foot interaction experience" and "ride interaction experience".By locating four clusters associated with interactive experience and affective vocabulary and analyzing the samples extracted from the four clusters,we can get the correlation between affective variables and specific product design attributes in the process of four kinds of experience,and to a certain extent,we can dig out the potential user needs.And we can make sure that the method indeed has its guiding significance and practical application value for the design of online products. |
PDF Full Text Request