Research And Implementation Of Search Advertising System Based On Convolution Semantic Model

The rapid development of the Internet has changed people's way of life.The endless stream of services not only facilitates people's lives,but also subverts the original industry.The term "internet thinking" has also been repeatedly mentioned.Internet products and services receive traffic and data,and how to monetize these traffic and data is the key to "internet thinking." Search advertising is the largest and fastest growing form of advertising in online advertising.Search engines have a large amount of user data and traffic,so Internet companies are eager to translate these data traffic into actual business benefits through the guidance of search behavior.Therefore,the core goal of search advertising is to predict the user's search purpose and deliver relevant advertisements based on the user's search terms,to ensure the interests of the advertiser and the advertising system,and to achieve a win-win situation among users,advertisers and advertising systems.It is not advisable to simply consider the interests of one party.The placement of search advertisements is a win-win situation after the tripartite game.In this paper,I describes the search advertisement and its related characteristics,in addition,according to the uniqueness of the search advertisement,I deeply analyzes the functional requirements and non-functional requirements of the search advertising system.Combines the key technologies of the online advertising system to the system,the overall architecture and the functions of each module are described,and the online advertising system for search engines is finally designed and implemented.This advertising system includes four modules: query processing,advertisement retrieval,advertisement sorting and advertisement management.Advertisers can create and manage ad plans,upload creatives,view ad serving statistics,and more in the ad management module.After receiving the user search request,the system will standardize and rewrite the query.The advertisement retrieval module performs matching according to the expanded keyword set to obtain an advertisement candidate set.The ad sorting module sorts by the eCPM,and returns the top ad to the search engine.In this paper,the semantic matching problem in search advertisement retrieval is compared.The effects of three semantic matching models are compared and the convolution semantic matching model is selected for this online advertising system.The model first uses the word2 vec algorithm to vectorize the query words and the advertisement title,and obtains the vectorized representation of the query and the advertisement title.The title is expressed as a low-dimensional semantic vector by the convolutional neural network for the advertisement title,and the query is performed by using the deep neural network.It is expressed as a low-dimensional semantic vector and the distance between two semantic vectors is calculated by the cosine distance.The click data is used for supervised training,and finally the semantic similarity model is trained.The model can be used to predict the semantic similarity of two sentences and obtain the low-latitude semantic vector expression of a sentence.