Research On The Recommendation Method Of Coal Products Based On Convolution Matrix Decomposition

Coal is an important energy to support the national strategic development.In the coal trading market of Shanxi and other regions,it is often limited by many factors such as local policies,business development level and market service ability.Problems such as low degree of coal commodity standardization,too traditional supply and marketing mode,and unfavorable connection of coal production and transportation are gradually exposed.A large number of high-quality enterprises in the market are often faced with the embarrassing situation of urgent production and manufacturing tasks,which makes it difficult to buy high-quality,reasonable price and low freight coal products in a short time.Considering the price of coal commodity itself and the long-distance freight problem in the sales process,it is difficult to realize the standardization of coal trading,and it is difficult to formulate the ideal sales strategy combined with its own attributes in order to find the best buyer.With the continuous rise of modern online trading mode,the use of intelligent recommendation system can guide buyers to make a relatively reasonable purchase behavior through a series of decisions.This method may play a role in boosting the digital transformation of the coal industry.However,in the face of a large number of audience groups,there is often a scarcity of known users’ comments on goods,which will bring serious sparsity problem of goods scoring matrix.Based on the above analysis,the main research contents of this thesis are as follows:(1)Based on the idea of collaborative filtering,a model-based recommendation method TCRM is designed.Convolutional neural network is used to obtain the word embedding of relevant attributes from the text information of users and items as features,and the feature vector is used as the key factor of the recommendation model.By fusing the matrix decomposition method,the scoring matrix is reconstructed through user features and item feature vectors to generate the user’s rating of items,Combined with the similarity of features between users and features between projects,the recommendation of coal products in different dimensions is completed.(2)GCEM model is proposed to transform the convolution core of TCRM by multi-channel grouping expansion convolution,which increases the sliding receptive field of the filter,avoids the raster effect of expansion convolution in pixel computing,and retains the coherence of semantic information of attribute text to a large extent.(3)On the basis of GCEM model,according to the assumption that the observation error of actual scoring matrix and prediction matrix is based on Gaussian distribution,a more perfect DCPMF recommendation model is proposed through the objective optimization matrix decomposition with maximum a posteriori probability.On the basis of the overall framework,spark technology is introduced to compare the proposed method on the coal trading data set.The experimental results show that the proposed TCRM,GCEM and DCPMF models not only effectively mine auxiliary information,but also have better recommendation accuracy and RMSE value,and significantly improve the recommendation quality.