System and method for identifying romantically compatible subjects


        First,explain the difference between our method with the others in the existing market.At present,most dating sitesare using the questionnaire type of personality test and try to match people by personality alone. These tests are generally multiple choice and require a lot of user input data to create any sort of matching model. This creates a difficult situation forquick and effective matching,is time-consuming,and the matching accuracy is extremely low.Only a few (such as TINDR) use the face recognition technology and the Likert Scale to determine the user's matching (this method is highly inaccurate and is mistake prone),and generally its using the calculation only in the form of statistical,its accuracy and effectiveness is also very limited too。 

        Our patents and methods are based on the face recognition combined with a binomial regression model,using the algorithm to determine each user's own preference or "type",by recording their simple point selections,using our statistical algorithm to make a quick and the exact choice. In addition,our system will continue to revise its matching algorithm utilizing new information until the user receives the most appropriate match possible。 

        Some sites use facial recognition to search but not to match people as mine goes on to find the people on the site that will like the user how initiated the search. We don't require any words,ratings,stars or any type of language to understand who a person finds attractive and vice versa.。


        This idea has been vetted by Statisticians and Scholars and it was concluded that it is mathematically correct. What the problem I have had is that most computer programmers don't understand statistics so it has been hard to find someone who could take a facial recognition program and combine it with my algorithm.Only a few can do it and they want a good sum of money to create even a test prototype.There are many talents and expertise in China,and it is believed that the sustainable development of this technology is very beneficial.


       In fact my program also uses a technology that can determine your personality and/or emotions just from a face picture. It is very Apple like in design,in that it is streamlined, requires no extra input from the user and does everything necessary to match two people without having language involved. From a mathematical perspective language is faulty and not quantitative,skewing the results. Imagine you are rating how hot you want food. Your 3 Stars might be 5 for me. Using a Likert Scale ends up with this problem.Therefore,our calculation method is a combination of the relevant areas of expertise with the scientific algorithm and get the accurate matching results,is currently the most humane and the most scientific advanced technology in the computing matching field.

Opinion on Patent No.: US 8,595,257 B1

System and Method for Identifying Romantically Compatible Subjects

By Stanley J. Smith

MetInfo enterprise content manager system | MetInfo CMSDR Stanley Smith Profile.pdf

        This patent describes a hybrid methodology for predictability testing and pattern matching of data derived from binary and ordinate data sets. The union of the data analysis is used to assert compatibility between individuals enrolled within a dating affinity framework. The basis of the methodology is from standard sampling theory, R2 testing from classical multi-variate analysis, binomial expansion expectation values, and likelihood ratio testing. The simplicity of factorial terms remaining from the binomial coefficients renders a simpler expression for the likelihood ratio. As an added measure, weight coefficients used in the binomial regression, provide further optimization through reinforced learning from mutually derived image processing metrics. The peak-to-minimum discrimination of the weighted binary vector covariance matrix can be optimized through back propagation of perturbed binary weights from feedback derived from image processing ordinate metrics cross-correlations.


        The target data is a combination of binary decision sequences and ordinate data based on quantitative matching parameters. The Wald Statistic is the discriminant used for asserting predictability that requires sufficient measurement variability in relation to static measurement error which is analogous to standard error. With a sufficiently large Wald Statistics, a first tier of statistically significance for predictability is achieved.


        The mutual affinity between parties that pertains to a successful dating dispositions is related to a dual joint hypothesis model which has potentially bivariate error. The patent design incorporates 3D tracking of eye movements, convergence image regions reflective of attraction, and deliberate sub-selections to augment conclusions from binary solicited questions about affinity. Features that would be useful for ranking images based on attraction could include dominant color distributions, homogeneity of color blends, symmetry, hair color, color contrast, facial feature geometric ratios, eye convergence, dimples, and smile inflections, for example.


        The improved predictability of the binary component is achieved by establishing a least-squares fit between with ordinate data derived by image matching preferences and the associated binary sequences. An improvement could be achieved through polynomial fitting.A machine learning adaptation of the ordinate matching parameters is achieved through a Pearson Correlation process (normalized cross-covariance) that allows for iterative binary weight optimization. Each sequence containing binary and pattern related matching data is used as vectors for which correlation feedback from the ordinate vector affects training of the weight coefficients of the binary vectors.


        In summary, the described patent methodology uses "binomial regression" which has inherent advantages for leveraging simple "Yes" or "No" data profile responses. With parallel ordinate image processing data available, several advantages result, namely,


            1) The weighting applied to the simple numeric attributes can be augmented by machine optimized fractional coefficients providing a wide dynamic range. Increasing the number of Yes/No questions also increases the dimensionality of the solution;

            2) Mutual affinity can be validated first hand by the matched individuals. This eliminates null-hypotheses used in the prediction technique allowing more "ground truth" confirmation for the weights.

A hybrid decision process using quantitative image processing metrics could also be augmented to provide statistically independent match recommendations.