Evaluation Of Internet Advertising Effectiveness By Eliminating The Indifference Hypothesis

The effect of Internet advertising is the most concerned topic in the field of Internet operations,because it directly affects the vital interests of Internet operators and the shopping experience of users.Many tasks of advertising operators have taken Internet advertising effects as an important reference basis.Therefore,evaluating the effectiveness of Internet advertising has become a hot research topic in academia and industry,and has very important research significance and commercial value.However,because the advertising effect of a web page is determined by many complex and heterogeneous factors,in addition to the advertisement itself,it is also related to factors such as page layout,user type,and browsing track.Therefore,how to find a more scientific and real evaluation index and evaluation model of Internet advertising is extremely challenging.First of all,Internet advertising effectiveness mostly uses the advertisement click through rate(CTR)and conversion rate(CR)as evaluation indicators,both of which are coarse-grained and inaccurate display evaluation indicators.The click-through rate indicator ignores ads that users notice and form a certain impression,but does not pay for specific click actions,while the conversion rate indicator confuses the user's browsing of web content with the browsing of ads;Secondly,the current evaluation of Internet advertising data sets has a single information source,poor integrity,and lack of sufficient implicit information,such as the user's emotions when paying attention to the web page,the user's own personality,etc.,and it is impossible to construct an accurate evaluation model based on multi-source heterogeneous data;Thirdly,the current evaluation methods of Internet advertising are mostly based on the assumption of user indifference and the assumption of page indifference.The user indifference assumption ignores the differences in behavior caused by different user types.Page indifference assumption assumes that the entire webpage is regarded as a whole,regardless of the effect of different layouts of the page and the location of advertisements on user behavior,it is difficult to reflect the behavioral differences of users in different interest areas of the webpage.Finally,the establishment of Internet advertising evaluation models is mostly based on static location features,lacking characterization of time series dynamic features,and ignoring user behavior as a time series.In response to the above problems,this paper proposes a method for evaluating the effectiveness of Internet advertising that combines multimodal features,multiple factors,and time series information.The main work and innovations of this article are as follows:(1)Propose a new evaluation index for Internet advertising effectiveness.In view of the single model of Internet advertising evaluation and insufficient real-time feedback from the evaluation system,this paper proposes an evaluation index of "Impression Space" Internet advertising effectiveness.First,use the psychological impression theory to construct the impression space,design a vector to simulate the brain impression space,the simulated impression space is reduced to a vector composed of multiple factors,and express the complex definition language through a vector;second,choose to browse In the process,subjective and external factors that may affect the establishment of impression space for Internet advertising may be used.Use comparative research and overall analysis methods to find out how these internal and external factors affect the construction of impression space,and select key,expressive,and diverse factors Through a series of permutation and combination methods to construct a multi-modal impression space;finally,the use of correlation analysis and deviation test in mathematical statistics to study the scientific and effectiveness of the impression space.(2)Large-scale multi-source heterogeneous data collection.In view of the lack of personalization of Internet advertising recommendations and the delay of advertising computing efficiency,this paper conducts multi-source heterogeneous data data collection.First,based on psychological behavior theory and methods,collect user information,including static information and dynamic information;second,use web design methods and eye tracking to collect information on web page layout content and user page operations;afterwards,use fuzzy set theory to The collected data is quantified and standardized;finally,through the correlation analysis method,multi-source heterogeneous data is integrated to form a large-scale online Internet advertising evaluation data set to verify the degree of impact of different data expressions on the effectiveness of Internet advertising,in order to eliminate users The indifference assumption and the page layout indifference assumption lay the data support.(3)Eliminate the assumption that users have no difference.In view of the lack of personalization of Internet advertising recommendations,this paper proposes a user type differentiation evaluation method to achieve user type division.First,use fuzzy decision-making methods to quantify multi-source heterogeneous user attributes;second,integrate the quantified attributes,define combined attributes,and use to divide the combined attributes into different domains;finally,use the optimized group square difference and clustering Method to find the best number of user types,based on this analysis of the various types of users.Experiments show that different types of users have different click behaviors and browsing habits.(4)Eliminate the assumption that there is no difference in page layout.Aiming at the problem of indiscriminate treatment of web pages,this paper proposes a research method of user's differential behavior based on interest areas.First,explore the layout of Internet advertisements and webpage entries,and define the page layout status of the webpage under the basic search engine;second,define the interest area of ??the page with attention to time;finally,filter the behavior data of users in the interest area and use association rules to mine the click behavior of different users Features;use multi-frequency sub-path algorithm to mine users' browsing preferences.The experiment proves that the user's focus area on the web page is F-shaped,and different behavior preferences will be generated in different interest areas.(5)Establish a cascading Internet advertising effectiveness evaluation model that integrates multiple factors,multi-category features,and time series features.Based on the above work,this paper proposes an evaluation model of Internet advertising effectiveness based on multi-modal features and time-series features.First,the Markov chain is used to find the timing characteristics of the features;second,the gradient boosting tree(Gradient Boosting Decision Tree GBDT)module,factorization machine(Factorization Machines FM)module,and field factorization machine(Field Factorization Machines)are constructed.FMM)module,partial Markov chain module(Partially Observable process Makorv POM)and other multi-module cascading deep learning framework;finally,establish a cascade model based on multi-modal features and timing features to achieveaccurate prediction of Internet advertising effects,and Analyze model complexity and model performance.