The Research & Development division includes six workgroups (adForecast, adLearning, adLanguage, adOptim, adCription and adStat) working on research themes and applications in online advertising. These activities involve the presence of the research team and part of the programming department staff. The research fields include Statistics and Econometrics, Artificial Intelligence, Algorithms and Cryptography, Numeric Calculation and Optimization Theory.
The modeling econometrics and prevision adForecast group explores the econometrics methods of prevision and applicability areas in advertising online. The natural systems intelligence consists in the ability to preview the events and to evaluate quantitatively and qualitatively certain processes. The fundamental questions that we ask are: How can we measure the predictable behavior of Internet users? What are the factors that influence its decision? How can we predict certain phenomena that interact dynamically over time?
The scientific preoccupations of adForecast group include explorations of discret pattern decisions, the econometrics of chronological series, half parametric and nonparametric, panel data econometrics.
Artificial Intelligence and Data Mining technics ocupies a central position in the majority optimization applications that we develop. The frameworks that the machine learning is based on offers today the opportunity to develop some classified models and a more efficient numerical forecast imposed by classic econometrics. These nonlinear and nonparametric models are used mostly in prediction processes of demographic and behavioral tag.
Begining with 2007, our team members have had the preoccupation of exploring the biological calculation (neural and immunological) principles and their applications of machine learning. This way, a part of our team is interested in artificial immunological systems for numeric optimzation and shapes recognition. Recent supervised classification technics based on barycentric cytokine networks are shown to be top classifiers and a development of the capacities of artificial immune systems’ recognition.
Also, including the concept of quantum parallelism in iterative neural networks modeling offers a higher power of time series forecasting. The adLearning group is also researching aggregation and assembling technics (boosting, bagging, etc.) that use more classifiers to get classification improved performance.
Data mining, analyzing and processing large volumes of data is another concern of this group. The filtering methods, cleaning, sampling and reducing the data are essential, given their role in providing the data sets that will undergo the forcast processes.
Communication is one of the most important Internet features, primarly being the expression of intention. Understanding the intention of Internet users and the meaning that he has on detecting the sens of communication content is a major importance in online advertising. Multiple optimization processes of ads delivery depend on this rank of understanding online ad. AdLanguage group focuses on ultimately concepts on drawing the text significance, automatic summarization of web page content, developing metrics to determine the semantic closeness of texts, automatic translation and also using statistics methods in the automatic dressing of natural language.
Introducing these methods in software applications brings a new dimension in delivering ad content: users will receive the most relevant ad for their interest, meaning that the ad will display in the exact site sections where the content is similiar to banner’s message.
The adOptim workgroup determines the mathematical modeling of online advertising processes and optimal decesions or decrease errors for various problems. These optimal solutions can be defined through complex research technics (through directions, on regions, etc. ). A modeling example represents the error spreading in neural networks where the optimal problem requests finding the minimum points of an error function to a n-dimensional hyperplan.
The group preoccupations also include the modeling decisions using graphs theory elements, linear optimization problems, harmonious analysis.
Modern cryptography is based on mathematical and programmatic concepts having a main purpose the security of informational flow. The usefulness is shown in the security of e-commerce, personal data and their communication channels. Criptography’s modern directions are following criptosystems based on eliptic curves and quantum cryptography.
The market analysis group follows closely the study of a static population and the development of some original methods for their explorations. Online advertising is for people who are difficult to explore and whose needs and desires are in a non-stopping dynamic. The interdisciplinary knowledge of this segment has an essential importance and brings precious information that are valued by the research workgroup.
MiRaGe solution was developed for our research purposes. MiRaGe as a static tool offers the latest features regarding data analysis and processing (complex sampling technics, statistics hypothesis tests, ANOVA, linear regression and correlation analysis, neural networks, factor analysis, cluster analysis, etc).
The group also offers consulting services and statistical reviews for TradeAds Company clients.