Research On Demand Forecasting And Decision Optimization Of Remanufactured Products Based On Online Reviews

With the rapid deterioration of the global climate and the excessive consumption of resources,environmental protection has become a highly concerning topic.Only by promoting green and low-carbon development in a cost-effective manner and ensuring that green manufacturing is promoted in manufacturing sector can we contribute to achieving carbon peaking and carbon neutrality.To advance the goal of carbon neutrality,remanufacturing is an effective method to transform recycled products into brand new products by reusing good parts while replacing worn out ones,which has great potential to increase economic benefits,save energy consumption,and reduce negative environmental impacts.However,due to the unobservable nature of the remanufacturing process,there is a high degree of uncertainty in its demand.Especially in the field of supply chain management,demand uncertainty seriously affects the normal operation of the entire supply chain system and has a significant impact on enterprises’ procurement,inventory management,scheduling,capacity management,and assortment planning.Therefore,intelligent technologies need to be applied to provide accurate and effective information for customer demand forecasting and to solve the problem of ambiguity and uncertainty of customer demand data.Based on this,we start from the rich structured and unstructured data of online retailers,take online consumer reviews(OCRs)as the main data source,and employ a multidimensional fine-grained sentiment analysis approach to identify the most appropriate and precise model for enterprise demand forecasting by comparing the performance of multiple classical forecasting models.It further provides theoretical support and reference basis for production and operation management of enterprises and policy design of the government.First,a framework of "hierarchical embedding models" is proposed to obtain twodimensional embedding representations of attribute-oriented OCRs and emotionoriented OCRs.In terms of the attribute dimension,two different types of attributes are modeled: general attributes and unique attributes.For general attributes,a hierarchical framework for transforming OCRs into attribute locality maps is proposed,facilitating consumers to simplify their choices by leaping from specific technical attributes to abstract heuristic attributes,thus extracting perceptions and preferences for higherlevel attributes embedded in these reviews and minimizing the cognitive burden when assessing their quality level.For the unique attributes that distinguish remanufactured products from new products,it was verified that consumers are enthusiastic about green products due to these attributes.In terms of the emotional dimension,a three-level hierarchy of positive and negative emotions was investigated.Further,in order to quantify consumers’ perceived importance of different product attributes within the same product category,a consumer preference model was developed to visualize the fine-grained attribute differences between inter-brand and intra-brand products in terms of two indicators,the importance ranking of attributes and the ranking of attributes to be improved,to reflect consumers’ preferences.To evaluate the effectiveness of the proposed model,we collected datasets on remanufactured and new cell phones from Amazon,the largest online retailer,to investigate the specific factors affecting the demand for remanufactured products by comparing statistical-based methods,machine learning and deep learning methods,and to validate the performance of the Transformer model on demand prediction.Second,an optimal integrated recycling,manufacturing and remanufacturing planning problem oriented to the results of the demand forecasting model is solved,and a multi-period hybrid manufacturing/remanufacturing model is developed,which considers the differences in carbon emissions from manufacturing,remanufacturing for different quality levels of cores and transportation simultaneously,where the carbon trading price differs between the manufacturing and remanufacturing phases.To make the variability of cores closer to reality,four commonly used probability distribution functions(uniform distribution,Erlang-truncated exponential distribution,triangular distribution and gamma distribution)are used to characterize the quality respectively while considering the stochastic yield rate,based on which the optimal interval range of core quality levels is investigated and proved to be reasonable in real case studies.These models are also used to construct carbon emission cost functions with general applicability in the remanufacturing stage to study the impact of differentiated carbon emissions of cores with different quality levels on corporate decisions.As well as,reasonable subsidy/penalty policies are proposed according to the cores of different quality levels to provide a reference basis for the design of optimal government remanufacturing policies.Finally,genetic algorithm and particle swarm optimization algorithm are used to compare and verify the validity and credibility of the proposed model.Finally,a two-level hybrid manufacturing and remanufacturing system that comprises multiple retailers and a single manufacturer is investigated under recycling with a finite number of times,taking into account multiple uncertainties in return rate,quality of cores,yield rate,and demand simultaneously.Two different carbon policies,namely cap-and-trade policy and carbon tax policy,are compared and analyzed to obtain the optimal remanufacturing quantity,manufacturing quantity,yield rate,and expected average total cost under each policy,and to guide manufacturers on how it is profitable to recycle selectively based on the quality level of cores.Numerical experiments and parametric sensitivity analysis are conducted based on a more favorable cap-and-trade policy,and finally,a subsidy model under a cap-and-trade policy is developed.It provides a reference and reference for policy makers to optimize the design and selection of regulatory policies(i.e.,carbon tax,cap-and-trade,and subsidy policies)to maximize social welfare;it provides a realistic path for enterprises to intuitively understand how to weigh the economic benefits and environmental costs under carbon emission policies and subsidy policies,and to take specific countermeasures in time according to different policies to adjust production and remanufacturing planning to avoid unnecessary losses.This paper provides different insights for consumers,companies and governments respectively.For consumers,the difficulty of identifying a product’s strengths and weaknesses is simplified.Consumers can not only get a more direct macro view of a product through more intuitive high-dimensional attributes as a way to understand its quality level,but also make it easier to compare different products with fine-grained attributes through structured preference models.For enterprises,it can provide managers and technicians with different perspectives on intra-brand and inter-brand products,as well as guidance for remanufacturing feasibility analysis and product redesign,and also provide an effective reference basis for the recycling,production and remanufacturing practices of hybrid manufacturing/remanufacturing enterprises.For governments,it can provide reference and reference for optimal design and selection of regulatory policies under the trade-off between social welfare and environmental impact,so that it can compare different decision-making strengths and develop reasonable incentive and penalty strategies to improve regulatory effectiveness.