Research On Semantic Understanding And Advertising Ranking In E-commerce Sponsored Search

The problem of query semantic understanding and ad ranking in e-commerce sponsored search has always been a hot topic in both academic and industrial research.The two are interdependent,ranking is the ultimate goal,which can only be achieved with a good semantic understanding of query phrases,they are indispensable.The key technologies for semantic understanding are query intent prediction and semantic query expansion,and the ranking process involves advertising relevance calculation and advertising click rate estimation.The accuracy of semantic understanding directly affects the quality of the advertisements returned to the users and the accuracy of the ordering of the advertisements.However,the traditional method based on probability can understand the query semantics of the users to some extent,but it is not fine enough.The semantic understanding method of this thesis describes the user's query by extracting powerful features.The MCI(mutual click intent)is introduced to describe the degree of association between two query items,and the improved PageRank algorithm is used to calculate the contribution of the query item.Then,the features related to the degree of relevance and contribution are extracted,and the extracted two types of features are applied into the classifier to accurately predict the user's query intent.In addition,in the contribution solving model,there may be a problem of useless directed edge jamming algorithm iteration when querying the directed graph.The "pruning" operation is used to remove the useless directed edges and improve the accuracy of the contribution value solving,and ultimately improve the performance of the query intent prediction classifier.At the same time,in order to retrieve more relevant advertisements,the method makes the semantic extension of the query on the basis of the Word2 Vec model.Subsequently,we designed a series of experiments to prove that the "pruning" operation can improve the performance of the classifier and the semantic understanding method can effectively retrieve more and more relevant advertisements.The accuracy of ad ranking directly affects the user's query experience and the direct economic benefits of advertisers and advertising media.The main factors affecting the advertising ranking are advertising relevance,expected advertising click rate and advertiser bidding.Through comprehensive analysis,this thesis is based on GBDT model which is to achieve advertising click rate estimation,the contribution degree solving model and the standard TF-IDF model which are to calculate advertising relevance.In addition,the semantic understanding operation of the query makes the query entering the ranking module not a single query phrase,but a set of query phrases.Therefore,in the ranking process,it is necessary to solve the problem of ad ranking under the condition of multiple queries.The solution presented in this thesis is to use the query sub-phrase with the highest score,that is,the one that best represents the user's query intent in the query intent prediction process instead of the currently retrieved query phrase to participate in the calculation of the ad ranking.After obtaining the data of the influencing factors based on the above process,the ranking formula is applied to achieve the ranking of advertisements.The experimental results show that the method is effective.