Commodity Classification Expansion Based On User Log And Query Semantics

The commodity classification query of the e-commerce platform is similar to the characteristics of the document query,therefore,it can be included in the research field of information retrieval.There may be two types of problems when users use the e-commerce search engine to query: first,incorrectly entering the wrong query,which is inconsistent with the product classification in the database;second,entering an intentional query and the platform displays too many categories of products which do not match the user's interests.Inspired by the idea of query expansion in information retrieval,combined with the practical problems solved during the internship in e-commerce company,this paper proposes a commodity classification extension method based on user logs and query semantics,so that the query results are more in line with the real needs of users.The main work of this paper contains the following aspects:1.Predict commodity classifications for user queries.Based on the user log drawn from the retrieval background,the user query and commodity classification datasets are trained with FastText to obtain the commodity classification prediction model.By analyzing the characteristics of the user log file,this paper predicts commodity classifications for the user query in three levels.2.Expand commodity classifications for user queries.Based on the commodity classifications which have been queried in user history,the semantic similarity between current query and historical commodity classification is calculated with Word2 Vec,and the related commodity classifications are used as extension classifications.To fully exploit the semantic relationship between words,this paper utilizes reptile technique to collect corpus datasets in multiple e-commerce platforms.3.Experiment and system development.Through the first experiment,compare the performance of the commodity classification prediction models generated by training user logs under different conditions,and the model with the highest accuracy rate is selected.Then,through second experiment,compare the distributions of semantic similarity calculated by different dimensional word vector files,then determine three reasonable similarity thresholds.Finally based on the results of the first two experiments,the commodity classification prediction and expansion performance are displayed through the developed experimental system.