Research And Application Of Mobile Phone Evaluation Model Based On Attribute Clustering

With the rapid development of the ecommerce market,the online shopping is increasingly favored by consumers.The judgment can be made by analyzing the performances of phone's appearance,price,function and other aspects according to the review information on the e-commerce platform,when consumers purchase a mobile phone.In addition,mobile phone manufacturers can also take advantage of these reviews to understand the concern tendencies of the consumers,and providing strategic solutions for the next generation phone manufacturing.The mobile phone evaluation model in this thesis is based on commentary corpus and can provides reference for consumers or manufacturers by analyzing the commentary corpus to reflect the consumer's tendency towards a certain feature of the mobile phone.The existed theories about evaluation model,the attribute evaluation structure,the attribute clustering and related technologies are studied.Firstly,crawler rules of octopus data collector are designed and 5 brands mobile phone reviews are collected from JD website.These reviews are seen as a reference material,and the reviews about honor phone named as the commentary material of a single phone are used to detect the model evaluation corpus.Secondly,the words of reference corpus and the commentary material of a single phone are segmented and stop words in them are removed by present Chinese Natural Language Processing technology.Thirdly,The CBOW(Continuous Bag-of-Words)model is trained by pre-processed reference corpus,and the attribute words and their word vectors related to the feature words of the mobile phone are obtained from the model,which are respectively clustered by Kmeans,spectral clustering and SOM(Self-organizing Maps),and selects the best clustering result to determine the attribute evaluation structure of the mobile phone.Finally,based on the pre-prcessing results of the reference corpus and the commentary material of a single phone,the attribute words about honor mobile phone reviews are extracted,the evaluation model of a certain mobile phone is simulated and visually displays by combining the mobile phone attribute evaluation structure with the evaluation model.According to the thought of TOPSIS(Technique for Order Preference by Similarity to an Ideal Solution),the evaluation method of phone is established by calculating the proximity of the attribute word set of the commentary material of a single phone and the attribute word set of reference corpus,which means the phones' feature tendency.Due to the non-numerical data,the way to calculate Jaccard similarity coefficient is used as the method to calculate the phones' feature tendency.Because of the diversity of Chinese word descriptions,the evaluation model based on semantic similarity is proposed to dynamically adjust the model according to the semantic similarity threshold.The experiments demonstrate that the average precision rate,average recall rate,average F-measure and accuracy rate of attribute word clustered by SOM reach 84.4%,80.63%,82.47% and 80% respectively,which prove that the rationality of attribute evaluation structure is established by Kmeans attribute clustering.The evaluation method based semantic similarity can more flexibly reflect the degree of mobile phone feature propensity of different semantic levels,and provide consumers with reference to different accuracy ranges.