Research On Taobao Shopping Behavior Prediction And Commodity Recommendation Mechanism

Information overload has become an unavoidable problem in the era of big data.The emergence of recommendation systems has effectively alleviated information overload.The research and improvement of recommendation systems have received extensive attention.Various recommendation algorithms have their own adaptation fields.Based on different applications and problems,various improved recommendation algorithms emerge one after another.A single recommendation technique can no longer solve the problem,and various hybrid recommendation algorithms have emerged.This paper adopts a text classification model to classify items with low scores and avoid recommending products with low scores to users.Then input the user's behavior word vector mean and recommended commodity word vector,and push the product through the deep neural network-based Word2Vec recommendation model.The main research contents are as follows.The classification problem of the evaluation of goods belongs to the text classification process.The MDSMOTE algorithm adds the method of correcting the wrong samples to the MDSMOTE algorithm when generating new samples,which solves the problem of not merging the wrong samples when generating some new samples.The problem that the hyperplane is biased toward a few classes is a problem that has always existed in the traditional FSVM classification algorithm.In this problem,positive and negative penalty coefficients and fuzzy factors are introduced in FSVM.Combining MDSMOTE and FC-SVM to obtain the MDSMOTE+FC-SVM classification model makes the discrimination rate of the unbalanced data set higher.Experiments show that the model has higher accuracy.Train user behavior word vectors using the Word2Vec model.The Word2Vec word vector training field has a wide range of excellent applications.In this paper,the technology is cited.Through the Word2Vec framework,each user is treated as an article.The user purchases items in order of time,and the items are treated as words through Word2Vec.Get the user vector mean.Using the CBOW model to implement Word2Vec,while optimizing the Hierarchical Softmax layer,reduces the probability calculation speed of each word.Then,the input layer and the hidden layer and the model structure are adjusted to reduce the time complexity of the model training.A commodity recommendation model based on Word2Vec word vector training model is constructed.The user behavior word vector mean and the recommended item vector are taken as inputs,and a deep neural network is used in the middle,and the result is a score of the recommended item.Then,the recommendation model is optimized.In the first aspect,according to the SWEM simple word vector model,the mean value can better represent the information of each word,and the original user behavior word vector is changed to the mean value of the user behavior word vector as input.In the second aspect the training process uses a two-layer ReLU function to inactivate a portion of the neurons,reducing the number of neurons during training.Combine the product classification model with the deep neural network recommendation model based on Word2Vec.The scores of the exposed products are obtained by the recommendation model,and are recommended to the users in descending order of the scores.At this time,the products with poor evaluation may be recommended to the user,and when the user browses the evaluation of the products,the evaluation spread is found,and the purchase is abandoned.At the end of the paper,the text classification model is used to remove poorly evaluated products and avoid recommending them to users.Finally,the proposed model of this paper is compared with two recommended models based on PMF and LibMF.Comparing the values of MAE and RMSE,it is found that the proposed algorithm has a more accurate recommendation effect.