Women’s Suit Recommendation And Fitting Based On Random Forest Algorithm

With the prevalence of online garment consumption trends,major e-commerce platforms generate a large amount of garment purchase data every day.Due to the increasing number of users and types of garment,the sparseness and complexity of user data have caused a lot of obstacles to the optimization of garment recommendation models.Existing garment recommendation models are more sensitive to data loss,and have lower error tolerance to data sets.Existing garment recommendation models have poor processing effects on discrete data.They can only provide users with graphic information related to the recommendation results which users cannot try.They will reduce the user ’ s shopping experience and generate a lot of regretful consumption.Therefore,in accordance with the development of machine learning technology,big data analysis technology,artificial intelligence technology and industrial Internet collaborative technology,researching an intelligent recommendation model integrating garment recommendation has become the development trend of garment recommendation systems.In order to make up for the shortcomings of the existing garment recommendation model,solve the problems of complex clothing data,clothing data overload,low recommendation accuracy,and single recommendation results,the following research was made.Using women’s suits as an example:Firstly,the research status of garment recommendation technology was reviewed.A recommendation model based on random forest algorithm and combining style quantification of women’s suits,style similarity recommendation,and image fitting of recommendation results was proposed.The basic knowledge of random forest algorithm,clothing recommendation technology and image fitting technology was introduced in detail.Secondly,completed the construction of the female suit feature system and realized the quantification of the female suit style.Collected some product information about women’s clothing on JD.com and Taobao.com,combined with literature analysis and expert interviews,summarized the common style labels of women’s suits into five categories: business,casual,commuter,sweet and street,and summarized the characteristics of women’s suits as HSV color feature,style feature,pattern feature and decoration feature.Analyzed the content of each feature module,and explained the value assignment method and value standard of each feature.Completed the preliminary preparation for the recommended model construction.Then,a female suit recommendation model was constructed and the fitting training was completed.According to 8 experts’ scoring and color extraction,the suit sample set was marked.The feature importance score was calculated with the aid of the Relief-F algorithm.The feature was selected and the comparison experiment was designed.The random forest algorithm was used to establish associations between the features in the samples and the style labels.Build a classification model.Calculate the similarity between user goals and samples through decision-making leaf nodes,sorted the samples and formed a recommendation list.Collected image fitting training samples and completed the retraining of virtual image fitting models for female suits.This study used 250 women ’ s suit samples collected from e-commerce platforms,combined with fitting models for verification,and summarized 16 features（hue,saturation,lightness,length,silhouette,sleeve length,sleeve type,placket,material,solid color,text,geometry,nature,stitching,folds and open lines）as the optimal feature set.The accuracy of the model for style classification reaches 85.3%,which is significantly improved when compares with not preferred feature set and other classifiers.The accuracy rate of model recommendation is83.3% through user survey of 30 consumers,and the similarity of different results and goals in the recommendation list shows a decreasing trend.Research shows that the suit recommendation system based on the random forest algorithm can basically meet the customers’ needs for style recommendation,effectively solve the problems of clothing data overload,clothing data complexity,and single recommendation results,and improve the accuracy of clothing recommendation.