Study on the Deep Learning Product Classification Based on the Motivation of Consumers

Abstract

New Product Development (NPD) is actually a complex area involving strategy, management, research and development, production, marketing and decision-making, technology and the market need to be closely integrated. Due to the dynamic and competitive market environment, the compatibility of new products, that is, the consistency between a new product and the values of consumers, is a dynamic and complex system. Compatibility is related to the consumer’s experience, lifestyle, religious beliefs, and prior knowledge of the product item. According to the process of product development, the first step is often to define the nature and function of the product, which is actually a process of new product positioning and classification in the process of product innovation. In general, the traditional process of product classification only focuses on the product and the market, and it is also the process that the designer deduces the product on the basis of successive generations. This method has become the bottleneck or restriction factor of raising productivity and standardizing production in the practical production which needs innovation constantly. We’ve learned that to better understand something, we need to better categorize it. In recent years, the method of artificial intelligence technology has been widely used in product classification, identification, search and other fields, which is in line with our technical requirements. And the use of machine learning to solve the classification problem in product classification has been a widespread concern of researchers, they believe that digital, intelligent and networked means to enable us to find new solutions. In this context, this paper presents a fast and effective product classification method based on deep learning technology, the deep learning-based Motivation process framework, which embeds human-based motivational thinking into machine learning. It’s a new kind of experiment. This framework consists of three parts: target customer modeling method based on deep learning technology; customer feature closed loop based on Motivation process framework; Weighted fusion partially outputs an iterative classification result that combines a consumer perspective with a producer perspective. We use the consumer information reasoning method and the weighted fusion module to test the deep learning-based Motivation process framework method on Cars. The experimental data show that this method can improve the performance of new product classification. This paper introduces the consumer motivation analysis into the traditional deep learning method for the first time, and finds that it has a strong application prospect. Based on this fact, this paper proposes a framework of classification algorithm based on deep learning technology, which integrates relevant design and human psychology methods. In order to improve the traditional classification algorithm which only inputted the customer’s Past purchase traces Past purchase Library-CNN, the original PPL-CNN was optimized by Motivation process framework multi-neural network fusion to improve the overall performance of the network. Firstly, the image data of the target user is preprocessed and characterized to be transformed into feature vector. For example, Pearson product-moment Correlation Coefficient was chosen to evaluate the correlation between the interests and expectations of target consumers, thus making up for the limitations of having to enter and use data from large databases. The target consumer modeling module is then used to capture consumer interest, which is then fed to subsequent Motivation process framework modules. The target consumer modeling module uses image retrieval technology to model the target consumer’s past purchase behavior, explores the relationship between the consumer’s purchase history information and the new product information, and consummates the target consumer’s personalized modeling. In this experiment, the image data representing the user’s expectation is used to extract the feature information from the user’s motivation. The image is then further feature extracted from the serialized data by the convolutional neural network, so that more dimensional information can be used for classification, and finally the output is combed and converged through the fully connected layer. According to the product characteristics of the producer, the paper proposes a multi-neural network-based feature fusion method for the classification of motive requirements. Because of the bi-directional coupling of the data features, the neural network model designed based on this method can better fit the data, on the basis of the original, a two-way fit mechanism between consumer and producer is added to better deal with this kind of problems. The experimental results show that the model based on this method can effectively extract the multi-dimensional features of User requirements, compared with other comparative models, the performance of the model takes into account the optimal ranking of consumers and producers, thus producing a more comprehensive classification result. This approach and technical framework will influence the future development of NPD.