Research On The Effect Of Online Advertising Based On Eye Tracking

Currently,Internet plays an extremely important role in human life.Take online shopping as an example.When people want to buy something,they first use a search engine,and then make a right decision according to the recommendation results of the SERP(Search Engine Results Page).It is just the same when searching for questions' answers and raising some other queries.So,online advertising becomes a critical important part of merchant marketing.There are many fruits in online advertisement effect research areas,whichnot only provides users with more suitable recommendations but also helps merchants obtain better benefits.Although manyachievements have been made in this research domain,there are still some deficiencies.First of all,the prevailing evaluation indexes of the web advertising effectiveness such as click rate(CRT)are no longer valid.Secondly,the data sets used in research worksare non-public and have the characteristics of single-source,small scope of application and normative.Then,the web search results are considered as a whole,thereby ignoring the correlationsamongevery parts.Finally,overmuch undifferentiated hypothesis can not reflect the influence of many fine grained features on the attention and acceptability of advertisements..In fact,all of these differences lead to different levels of user attention and different acceptant degree for advertising,which in turn inevitably influence the accuracy of evaluation of online advertising effects.To solve the problems mentioned above,we propose a kind of evaluation method for web advertisement effect based on multimodal features.The main innovations of our method as follows.1.Propose a new online advertising effectiveness evaluation metriccalledobservation degree,thus predicting the effect of online advertising more accurately.First of all,this paper innovatively proposed the "observation degree" as the evaluation index of the online advertising effect in order to make up the shortcomings of the current research,and verified the effectiveness of the index through various experiments.Secondly,based on the "observation degree" metric,the prediction model is established,and further,its validity and stability are proved by experiments.This model includes four parts,which are mining multi-source features,classifying data sets by frequent patterns,establishing the stratified random forest model,and optimizing the results by comparing the prediction probability of two types of data sets.It is noted that the "observation degree" metric is the core and basis for all of our works.2.Analyze the user's attention distribution and put forward various attention effects.The multiple attention effects presented in this paper reveal the defects of the "no difference hypothesis" and play an important role in feature selection in online advertising prediction models.First of all,we divide the web page into different interest areas,and then obtain multiple parameters in different interest regions.Secondly,we make statistical tests and further quantitatively analyze features from three sides,which are the user's individual characteristics,the characteristics of the advertising itself and the eye movement parameters under the cross effect of the two above.Finally,according to the results of the test,we put forward a variety of attention effects,and explain the emergence of a variety of attention effects and their practical significance.3.Mine the temporal relationship between online advertising and web entries,and propose a temporal pattern correlation algorithm based on frequent patterns.The correlation between online advertising and webpage entries is also an important part of calculating the effect of online advertising.This algorithm mainly includes two aspects.We use frequent pattern algorithms to mine the association rules in interest areas,and then we propose the directional fixed length mining algorithm called DFBP which puts forward the most frequent browsing patterns of users.The algorithm gives the minimum length and minimum support of the browsing sequence,and gets the user frequent browsing mode sequence.In addition,in order to support the above innovative work,we set up a multi-source and large-scale online advertising evaluation dataset which makes up for the defects of single-source and incomplete in datasets of current researches.Therefore,it is the foundation for works in eliminating the non difference hypotheses.The dataset contains user individual information,advertising information,user eye movement information and mouse information.The collecting process of these kinds of information includes the following steps.First is verifying the importance and necessity of the collected information through literatures.Second is design of the specific process of information collection.Third is the quantification and standardization of collected data.Finally,we verify the importance of information to online advertising through correlation analysis.